GeForce 8800 GTX review: GeForce 8800 GTX
Because of design changes in the GeForce 8800 GTX chip’s new architecture, we need to consider some of this card’s specs differently than we have in the past. The basics are the same. The GeForce 8800 GTX has a core clock speed of 575MHz, and it comes with 768MB of DDR3 RAM clocked to 900MHz with a 1,800MHz data rate. That memory rate is a significant uptick compared to the 800MHz RAM in Nvidia’s last flagship card, the GeForce 7950 GX2. But one of the main differences in the GeForce 8800 GTX’s architecture lies in how we consider its pipelines.
We’ve been waiting for the GeForce 8800 GTX since we first got wind that Nvidia’s next-generation 3D card would be out before the end of this year. It’s everything we’d hoped it would be. For a suggested street price of $599, the GeForce 8800 GTX brings tremendous processing power to current-generation games. It’s also the first card to market that will support all of the 3D gaming-related features of Windows Vista and . The initial release of next-gen games is a bit far off. The poster child, the 3D shooter Crysis, is set to debut in March 2007, and even that game might not put all of the next-gen bells and whistles into play. Still, the GeForce 8800 GTX is so powerful, even compared to ATI’s fastest dual card combination, that there’s no reason to spend roughly $1,000 on a pair of Radeon cards when you can outperform them with a single $600 GeForce 8800 GTX. That and the fact that Nvidia has finally caught up to ATI’s image-quality advantages earn Nvidia’s newest card our Editors’ Choice award for high-end 3D graphics cards.
In the past, we’ve said that a 3D chip has X amount of pixel pipelines and Y pipes for shader calculations. But because of the new specifications of DirectX 10, the GeForce 8800 GTX employs what’s called a unified architecture. In other words, no pipe is geared toward a particular task. Instead, the GeForce 8800 GTX comes with 128 stream processors, which can dynamically process whatever info is thrown their way. This means that if your card is processing a shader-intensive scene, it can tap from more of the pipeline pool to process that image, rather than being capped at 24 or 48 pipes because some of the other pipes are set aside for geometry only. This capability should give game designers much more flexibility in how they design games, knowing that if they can balance the workload properly, they can pump a lot of processing power into a given calculation.
What’s perhaps even more impressive about the GeForce 8800 GTX is its sheer horsepower. Its transistor count sits at 681 million on a 90-nanometer manufacturing process chip. That’s more than the two 271-million-transistor chips on the GeForce 7950 GX2 combined. To power a single GeForce 8800 GTX card, Nvidia recommends a 450-watt power supply in a PC with a high-end dual-core chip and a typical combination of internal hardware. But the trick is that the power supply must have two PCI Express card power connectors to plug into the two sockets on the back of the card. Most modern power supplies should have the necessary connectors. If you want to add two 8800 GTX cards in an SLI configuration, however, you’ve got a challenge on your hands.
Nvidia hasn’t released a driver that will run the GeForce 8800 GTX in SLI mode as of the time of this writing, but it may have one out soon. Thus, we didn’t get to test it, but Nvidia did share the power supply specs with us. To run two GeForce 8800 GTX cards in SLI mode, Nvidia recommends at least a 750-watt power supply. But some of the recommended models on its SLI compatibility list go as high as 850 and even 1,000 watts. We suspect those higher-wattage recommendation will allow you some headroom for adding multiple hard drives and optical drivers, as well as very high-end quad-core processors. Still, it’s clear that building a next-gen SLI rig will be no small undertaking, at least for now. Heck, many midtowers PC cases are too small to accept a 1,000-watt power supply.
With no DirectX 10 games available to test on at the moment, we can’t speak to the GeForce 8800 GTX’s next-generation performance, aside from the fact that it’s the only card on the market that claims DirectX 10 compatibility. ATI’s next-gen card, code-named R600, was rumored to be released in January 2007, but we haven’t heard much about it so far. We imagine that ATI (whose acquisition by AMD has been finalized) will have a DirectX 10 card sooner or later, but right now Nvidia is the only vendor with something to show. And while we can’t really say who will win the battle for next-generation performance, the GeForce 8800 GTX dominates every single other card on the market right now.
One of the most important things to note about the GeForce 8800 GTX and its performance is that you would be smart to pair this card with a capable monitor that can go to resolutions of 1,600×1,200 or above. Nvidia calls this XHD (extreme high definition) gaming. Whatever you want to call it, if you’re not playing at high resolutions with antialiasing, anisotropic filtering, and other image-quality tweaks cranked, you’ll likely hit a CPU bottleneck, which means that you’re not giving the card enough to do. But when you get up to those high-quality settings, the results are amazing.
GameSpot was kind enough to provide us with benchmarks, as per usual. We suggest you check out their story, too; there are a number of screenshots taken during testing that show off the image quality. We’ll focus on frame rates. Our highlight here is . That game has been considered the pinnacle of DirectX 9-based game programming and has humbled even ATI’s mighty Radeon X1950 XT CrossFire setup, which can barely pass 60 frames per second (fps) with no antialiasing. But the GeForce 8800 GTX blew past ATI’s highest-end configuration, scoring 64fps on that test.
Oblivion also lets us highlight how the GeForce 8800 GTX has pulled Nvidia even with ATI on current-gen image quality. ATI has had an advantage on certain games, most conspicuously Oblivion, because through an unofficial patch, Radeon cards let you turn on antialiasing and high dynamic range lighting simultaneously. The resulting image looks noticeably better than if you can do only one or the other, as you can with the GeForce 7000-series cards. Not only can the GeForce 8800 GTX do both AA and HDR lighting, it also does them faster than a Radeon X1950 XT CrossFire rig. On that Oblivion test, the 8800 GTX scored an impressive 45 frames per second, which is much smoother than ATI’s 28fps.
You might notice that the GeForce 8800GTX doesn’t win on every single test. On Half-Life 2: Episode One at 8X antialiasing, an ATI CrossFire setup edged it out. It’s worth noting that the GeForce 8800 GTX hit 80 frames per second, though, so it’s not exactly slow. But even better, at 16X antialiasing ,which is more demanding, the GeForce 8800 GTX’s score stayed basically the same at 84fps, where the CrossFire cards’ scores dropped off. This lends to Nvidia’s argument that the GeForce 8800 GTX delivers better performance on extremely high image-quality settings.
The other test it lost on was Quake 4, wherein the Radeon X1950 XT CrossFire beat it at both resolutions by about 15fps or so. Again, even at 2,048×1,536, the GeForce 8800 GTX scored 68fps, so it’s by no means slow. But it’s also worth noting that a Radeon X1950 XT CrossFire setup costs between $900 and $1,000 for the master and the slave cards, and they can’t do DirectX 10. The $600 GeForce 8800 GTX and its forward-looking capabilities are clearly the better deal.
We should note a couple of final thoughts here. The first is that with the GeForce 8800 GTX, Nvidia is also unveiling something called CUDA, which stands for Compute Unified Device Architecture. Because of the 8800’s complexity, Nvidia is offering a framework for programmers to write software to use the GPU for intense number calculations. For gamers, Nvidia showed us how developers might use CUDA to really ramp up game physics calculations, but Nvidia is also offering this capability to the medical community and anyone else who might benefit from a combination of intense image-processing and number-crunching power. Nvidia is still getting the word out on CUDA, so there’s no way to check it out right now. Nvidia also unveiled its new PureVideo HD software as a component of its new universal ForceWare driver, which debuts today and includes support for the GeForce 8800 cards. PureVideo HD will run on both the 8000-series and the older 7000-series GeForce cards, and it’s designed to enhance HD video content coming from your PC. We have a sample system in to play with, and we’re in the process of putting it through its paces. Look for a blog post on our impressions of PureVideo HD next week.
3DMark06
(Longer bars indicate better performance)
1,280×1,024 |
Nvidia GeForce 8800 GTX
10959
ATI Radeon X1950 XT CrossFire
10832
Nvidia GeForce 7900 GTX SLI
10655
Nvidia GeForce 8800 GTS
8890
ATI Radeon X1950 XTX
6792
Nvidia GeForce 7900 GTX
6453
Need For Speed: Carbon
(Longer bars indicate better performance)
1,600×1,200, 8X AA, 16X AF | 1,600×1,200, 4X AA, 16X AF |
Nvidia GeForce 8800 GTX
40
43
Nvidia GeForce 8800 GTS
28
31
ATI Radeon X1950 XT CrossFire*
26
16
ATI Radeon X1950 XTX*
25
30
Nvidia GeForce 7900 GTX SLI
17
18
Nvidia GeForce 7900 GTX
10
17
Half-Life 2: Episode One
(Longer bars indicate better performance)
2,048×1,536, 16X AA, 16X AF | 2,048×1,536, 8X AA, 16X AF |
ATI Radeon X1950 XT CrossFire*
49
98
Nvidia GeForce 8800 GTX
84
86
ATI Radeon X1950 XTX*
N/A
76
Nvidia GeForce 7900 GTX SLI
29
74
Nvidia GeForce 8800 GTS
61
63
Nvidia GeForce 7900 GTX
N/A
30
The Elder Scrolls IV: Oblivion
(Longer bars indicate better performance)
1,600×1,200, 4X AA, 16X AF, HDR On | 1,600×1,200 no AA, 16X AF, HDR On |
Nvidia GeForce 8800 GTX
45
64
Nvidia GeForce 8800 GTS
34
54
ATI Radeon X1950 XT CrossFire
28
51
Nvidia GeForce 7900 GTX SLI
N/A
51
ATI Radeon X1950 XTX
18
35
Nvidia GeForce 7900 GTX
N/A
32
Quake 4
(Longer bars indicate better performance)
2,048×1,536, 8X AA, 16X AF trans super | 1,600×1,200, 8X AA, 16X AF trans super |
ATI Radeon X1950 XT CrossFire*
84
115
Nvidia GeForce 8800 GTX
68
100
Nvidia GeForce 8800 GTS
49
73
Nvidia GeForce 7900 GTX SLI
41
60
ATI Radeon X1950 XTX*
51
56
Nvidia GeForce 7900 GTX
16
34
Company of Heroes
(Longer bars indicate better performance)
2,048×1,536, 4X AA, 16X AF | 1,600×1,200, 4X AA, 16X AF |
Nvidia GeForce 8800 GTX
45
66. 5
Nvidia GeForce 8800 GTS
16
65
Nvidia GeForce 7900 GTX SLI
29
47
Nvidia GeForce 7900 GTX
29
43
ATI Radeon X1950 XTX
26
40
ATI Radeon X1950 XT CrossFire
16
25
F.E.A.R.
(Longer bars indicate better performance)
2,048×1,536, 8X AA, 16X AF | 1,600×1,200, 8X AA, 16X AF |
Nvidia GeForce 8800 GTX
45
68
ATI Radeon X1950 XT CrossFire*
33
51
Nvidia GeForce 8800 GTS
33
48
8800 GTX Reviews | TechPowerUp Review Database
-
Review Database
- Graphics Cards
- NVIDIA
- 8800 GTX
Sort items by:
DateReview TitleSite NamePopularity
2007 | |||
---|---|---|---|
Oct 21, 2007 | Zotac 8800GTX Amp! Edition | DriverHeaven | |
Sep 10, 2007 | EVGA e-GeForce 8800 GTS KO ACS3 | Digit-Life | |
Sep 4, 2007 | Point of View GF8800 GTX EXO | SGOverclockers | |
Jul 30, 2007 | ASUS EN8800GTX TOP AQUATANK | Digit-Life | |
Jul 18, 2007 | Zotac GeForce 8800 GTX AMP! Edition | Bit-Tech | |
Jul 7, 2007 | PNY XLR8 8800 GTX OC | HotHardware | |
Jul 5, 2007 | Zotac GeForce 8800GTX AMP! Edition | Tweaktown | |
Jul 5, 2007 | ASUS EN8800GTX AquaTank | Hartware (de) | |
Jun 21, 2007 | ASUS 8800GTX 768MB | TheTechLounge | |
Jun 16, 2007 | XFX GeForce 8800 GTX Extreme | Digit-Life | |
May 30, 2007 | Zotac GeForce 8800 GTX OC Edition | TechPowerUp | |
May 25, 2007 | OCZ GeForce 8800 GTX | Legit Reviews | |
May 24, 2007 | Foxconn GeForce 8800 GTX | X-bit labs | |
May 22, 2007 | MSI GeForce NX8800GTX OC Liquid | Tweaktown | |
May 18, 2007 | Asus EN8800GTX 768Mb | Red and Blackness Mods | |
May 10, 2007 | ASUS EN8800GTX | TechARP | |
May 5, 2007 | ASUS EN8800GTX AquaTank | Hexus | |
May 4, 2007 | Foxconn GeForce 8800GTX | Hard h3o | |
Apr 17, 2007 | MSI NX8800GTX OC Liquid | HardwareZone | |
Apr 11, 2007 | ASUS EN8800GTX AquaTank | Bjorn3D | |
Apr 10, 2007 | Foxconn GeForce 8800 GTX | HotHardware | |
Mar 27, 2007 | ASUS EN8800GTX AquaTank/HTDP/ 768M | OC workbench | |
Mar 22, 2007 | Sparkle Calibre 8800 GTX 768MB OC | MVKTech | |
Mar 5, 2007 | Foxconn 8800 GTX | Bjorn3D | |
Mar 3, 2007 | Sparkle Calibre 8800GTX P880+OC | Bjorn3D | |
Feb 28, 2007 | OCZ Technology GeForce 8800 GTX | Overclock Intelligence Agency | |
Feb 28, 2007 | XFX NVIDIA 8800 GTX XXX | Technic3D (de) | |
Feb 21, 2007 | BFG NVIDIA 8800 GTX OC | Technic3D (de) | |
Feb 21, 2007 | EVGA e-GeForce 8800GTX KO ACS3 768MB | TechPowerUp | |
Feb 20, 2007 | Shootout at the 8800 GTX corral: ECS vs OCZ | Hexus | |
Feb 19, 2007 | MSI NX8800GTX-T2D768E-HD | PCStats | |
Feb 8, 2007 | XFX GeForce 8800 GTX 768MB | Hexus | |
Feb 8, 2007 | ASUS EN8800GTX | HotHardware | |
Feb 8, 2007 | Gainward 8800GTX | XSReviews | |
Feb 7, 2007 | Gainward Bliss 8800 GTX | Technic3D (de) | |
Jan 29, 2007 | EVGA e-GeForce 8800 GTX | Barry’s Rigs’n’Reviews | |
Jan 27, 2007 | XFX GeFORCE 8800GTX XXX | NeoSeeker | |
Jan 26, 2007 | XFX GeForce 8800 GTX XXX | Guru3D | |
Jan 25, 2007 | GeForce 8800 in SLI | Tech Report | |
Jan 25, 2007 | MSI NX8800GTX | NVNews | |
Jan 25, 2007 | MSI GeForce 8800 GTX | HotHardware | |
Jan 24, 2007 | BFGTech GeForce 8800 GTX Water Cooled | HardOCP | |
Jan 19, 2007 | ASUS EN8800 GTX | HardOCP | |
Jan 18, 2007 | BFGTech 8800 GTX Watercooled | Bit-Tech | |
Jan 18, 2007 | BFG GeForce 8800 GTX Water Cooled | Guru3D | |
Jan 18, 2007 | Foxconn GeForce 8800 GTX | HardwareOC (de) | |
2006 | |||
Dec 27, 2006 | XFX GeForce 8800 GTX SLI | Legit Reviews | |
Dec 27, 2006 | Foxconn GeForce 8800GTX | Digit-Life | |
Dec 26, 2006 | GeForce 8800GTX | Allround-PC (de) | |
Dec 20, 2006 | MSI GeForce 8800 GTX | TechSpot | |
Dec 15, 2006 | MSI GeForce 8800 GTX in 2560×1600 | Digit-Life | |
Dec 15, 2006 | Nvidia GeForce 8800 GTX | X-bit labs | |
Dec 15, 2006 | GeForce 8800 GTX | HardwareZone | |
Dec 14, 2006 | BFG GeForce 8800 GTX | Guru3D | |
Dec 10, 2006 | XFX 8800 GTX: SLI | Bjorn3D | |
Dec 10, 2006 | NVIDIA GeForce 8800 GTX | PCModdingMy | |
Dec 10, 2006 | BFG GeForce 8800GTX | ||
Dec 10, 2006 | Foxconn GeForce 8800 GTX | Hexus | |
Dec 7, 2006 | MSI NX8800GTX-2D3768E-HD | Viper’s Lair | |
Dec 1, 2006 | XFX GeForce 8800GTX | Nordic Hardware | |
Dec 1, 2006 | BFG 8800gtx | HardwareEcke (de) | |
Nov 11, 2006 | ASUS GeForce EN8800GTX/HTDP/768M | ||
Nov 11, 2006 | NVIDIA GeForce 8800GTX | OC workbench | |
Nov 11, 2006 | Leadtek WinFast PX8800 GTX TDH | TrustedReviews | |
Nov 10, 2006 | NVIDIA GeForce 8800 GTX | TechARP | |
Nov 10, 2006 | XFX GeForce 8800 GTX | Legit Reviews | |
Nov 10, 2006 | XFX GeForce 8800 GTX 768MiB | Hexus | |
Nov 10, 2006 | G80: NVIDIA GeForce 8800 GTX | Bit-Tech | |
Nov 10, 2006 | NVIDIA GeForce 8800 GTX | Hexus | |
Nov 10, 2006 | Sparkle GeForce 8800GTX | Bjorn3D | |
Nov 10, 2006 | XFX GeForce 8800 GTX | Bjorn3D | |
Nov 10, 2006 | Leadtek Winfast PX8800 GTX TDH | Bjorn3D | |
Nov 10, 2006 | NVIDIA GeForce 8800 GTX | Viper’s Lair |
Foxconn GeForce 8800 GTX Review
Foxconn GeForce 8800 GTX
Written by John Yan on 1/1/2007 for
PC
More On:
GeForce 8800 GTX
We’re on a Foxconn graphics card roll right now. First off, I reviewed their GeForce 7950 GT with 512MB. Today I look at the grand daddy of NVIDIA cards though. It’s been a few months since the release of the GeForce 8800 so with that here’s my review of Foxconn’s top end GeForce 8800 GTX.
The Foxconn GeForce 8800 GTX is the top of the line enthusiast card currently on the market. Featuring 128 individual stream processors running at 1.35GHz and 768MB of GDDR3 memory, the Foxconn GeForce 8800 GTX is a DirectX10 capable card. The core runs at 575MHz with the memory running at 1800MHz DDR. Compared to the GTS version, the GTX runs 13% faster in the core, 11% faster for the stream processors, and 11% faster for the memory. The memory interface is 384-bit for the GTX while the GTS is at 320-bit. You can read about most of the new features of the GF8800 from my preview so I won’t go too much into it again. The GeForce 8800 GTX is the same as the GeForce 8800 GTS I previewed but faster.
So what’s different between this and the GTS version. For one, the GTX card is a bit longer. You can see in one of the comparison photographs just how much longer it is. Another change is there are TWO power connectors on this card. The GTX requires more juice and you’ll need two power connectors for this card to work. While most cards have their power connectors facing to the right of the card, the ones on the GTX face up so that you can attach them a little easier. I’ve found that a few cases interfered with the horizontal facing power connectors and this change will help in those situations especially with two of them to connect. Finally, there are two SLI connectors at the top instead of one. No support now for the secondary SLI connector but it’s there for the future. You can plug either one in for SLI to work between two GTX cards. You can imagine the monstrosity once you can connect more than two of these cards together and the number of power lines being strung through the inside of your computer. I’m thinking they should just build an interface and case to allow another power supply to be installed so that the video cards can have it’s own dedicated power rather than sharing it with the entire system.
Physically, the card is pretty much the same as the reference cards out there. The over-sized cooler is big but it runs very quiet. I could hardly hear the fan spinning even under load. The venting of the hot air out the back of the secondary slot will help keep the inside of your case cooler. There is one unique twist that Foxconn has added to the cooler though. Situated around underneath the outer shell are blue LEDs that light up when the computer is turned on. There aren’t any blinking actions to make it annoying so if you have a window inside your case and love to show off lights, this card will emit a blue glow for you. If you know a little bit about electronics, you can adjust the lights yourself to suite your needs.
If you want to watch high definition video on your high definition display, then you’ll be glad to know that this card is HDCP compliant so as long as you connect it to an HDCP compliant display source then you’ll be able to enjoy Blu-Ray or HD DVDs in full resolution. I’ve been eyeing the Xbox 360 HD DVD drive for my console and PC and this card will come in handy for HD viewing. You’ll still need software capable of playing the new formats of course such as Cyberlink’s Power DVD 7. With this card, you’ll be able to enjoy some smooth HD playback as it will help decode the higher bandwidth video.
Dual-DVI connectors and various TV outs via a pigtail lets you connect this card in various ways. The Dual-DVI connection is capable of a maximum resolution of 2560×1600 with 32-bit color at 60Hz. If you need to connect this card via the old VGA connector, Foxconn has included two DVI to VGA adapters for you.
Foxconn bundles their video cards with two utilities and a USB gamepad. The two utilities are RestoreIt! and Virtual Drive. RestoreIt! is similiar to the more popular Symantec Ghost while Virtual Drive lets you copy a CD or DVD to the hard drive and run it from there. These two can come in handy and they’re free. The USB gamepad is pretty serviceable so if you need another controller then you’ll get one with the card. It’s modeled somewhat after the PlayStation 2 style with the dual analog sticks and four trigger buttons along with the four top buttons and directional pad.
First up is Futuremark’s 3D Mark 06.
3DMark®06 is the worldwide standard in advanced 3D game performance benchmarking. A fundamental tool for every company in the PC industry as well as PC users and gamers, 3DMark06 uses advanced real-time 3D game workloads to measure PC performance using a suite of DirectX 9 3D graphics tests, CPU tests, and 3D feature tests. 3DMark06 tests include all new HDR/SM3.0 graphics tests, SM2.0 graphics tests, AI and physics driven single and multiple cores or processor CPU tests and a collection of comprehensive feature tests to reliably measure next generation gaming performance today. We tested at the standard 1280×1024 resolution.
Quake 4 is Raven Software’s true sequel to the id classic. The game uses an improved Doom 3 engine for some great graphics. For the test we ran a demo featuring a few enemies and some squad mates. We set the graphics qualities at maximum and ran it on three different resolutions.
One of the surprise hits out of Monolith was F.E.A.R. This supernatural FPS looks incredible and really pushes a video card to its limits. For the benchmark, we ran three resolutions using the in game benchmark with all the settings set at max.
Prey has been in development for many years but the folks at Human Head finally released the game this year. The game utilizes the Doom 3 engine like Quake 4 and features the really cool Portal technology to garner some interesting game play aspects. All settings were set to maximum and three resolutions were chosen for the test.
Company of Heroes is an RTS that really pushes video cards. The game by the fine folks at Relic Entertainment is set in WWII and features deformable terrain as well as great physics. The level of detail in the game for an RTS is amazing. For the tests, we set everything at maximum or ultra to ensure that the card was taxed as much as possible
While there’s still the traditional anti-aliasing and anisotropic filtering modes, the GeForce 8800 introduces CSAA. We’ll test the regular modes first and then I’ll show you how NVIDIA’s CSAA doesn’t bog the card down as much while offering some very nice anti-aliasing quality. First up is 4xAA with 8xAF.
Next up we turn it up a little with the traditional 8xAA with 16xAF.
Now we move onto NVIDIA’s latest achievement in image quality and that’s Coverage Sampling Anti-Aliasing. I kept the anisotropic filtering at 16X for these tests and used 16x for CSAA. You might be surprised by the results below.
NVIDIA’s new antialiasing technique really shines with some great performance. With this card, you can turn up everything and get some great framerates at high resolutions. It’s a truly impressive result and one of the aspects that really makes this card stand out among the others.
The power of this card is truly impressive. When you look at the performance gains over the GeForce 7950GT as well as the GeForce 8800 GTS, the GeForce 8800GTX is hands down the best card to own right now. I don’t have the Radeon X1950 XTX with me to compare anymore but I’d sure like to see how much this card would beat ATI’s current flagship card. To be able to enable NVIDIA’s new 16x AA mode and get great playable performance at high resolutions is just outstanding. Any other card would be brought to it’s knees even at a lower AA mode running at 1600×1200. Foxconn gives you a card with a unique bundle and blue LEDs that light up the card. If you can afford it, it’s definitely a great purchase and Foxconn’s card ran solid throughout testing and in general usage. I’ll be putting up a second part of this article showing how much gain you get by SLI’ing two of these bad boys. For now, Foxconn’s offering is a real winner.
It’s damn expensive but it’s damn fast. Foxconn has a unique bundle and the LEDs make this card glow. If you have the money, go pick yourself up one for the ultimate in gaming performance.
Rating: 9.5 Exquisite
* The product in this article was sent to us by the developer/company.
About Author
I’ve been reviewing products since 1997 and started out at Gaming Nexus. As one of the original writers, I was tapped to do action games and hardware. Nowadays, I work with a great group of folks on here to bring to you news and reviews on all things PC and consoles.
As for what I enjoy, I love action and survival games. I’m more of a PC gamer now than I used to be, but still enjoy the occasional console fair. Lately, I’ve been really playing a ton of retro games after building an arcade cabinet for myself and the kids. There’s some old games I love to revisit and the cabinet really does a great job at bringing back that nostalgic feeling of going to the arcade.
View Profile
GeForce 8800GTX — lost the throne, but retained the crown
- Introduction
- How it was
- Radeon HD 2900 XT — «Suppressed Revolution»
- 8800 GTX three years later
- Enemies of the gray king
- Test applications, test bench configuration
- Overclocking 8800GTX
- Testing
- Summing up
Not so long ago (in November) the G80 video chip turned exactly three years old. This is a huge time for the graphics chip and for the video card as a whole. Usually one year is enough for the once powerful and productive video chip to move from the «high society», at best, to the bottom of the middle class. In three years, the manufacturer manages to update the lineup of its video cards at least two or three times, a new API, new generations of CPUs may well appear, not to mention the release of a whole bunch of new generation games. If a video card becomes obsolete so quickly in a year, what is there to talk about when three whole years pass ?! Often it turns into a piece of textolite of little use, and shows only a shadow of its former glory — neither new games, nor high resolutions will be beyond its power. Recall at least Geforce 3Ti, released in 2001. Was he capable of such heavyweights as Doom 3, Half-Life 2 or FarCry? Well, they certainly ran, if we talk about the fact of work, but it is unlikely that users with Geforce 3Ti/Radeon 8500 could afford high graphics settings, anti-aliasing or anisotropic filtering.
However, we have a different situation. Despite the fact that Geforce 8800GTX is 3 years old, it is not only able to demonstrate high FPS in the most modern games, but also allows you to set high resolutions, turn on anti-aliasing and be content with free anisotropic filtering.
Moreover, about 10% of all users of the Steam online service are «armed» with video cards based on Geforce 8800 — this already says a lot.
However, first things first…
recommendations
In order to help the reader remember the dawn of the 8800GTX, we offer something to remember from our previous articles:
- 04.12.2006 Comparative testing of 8800GTX and X1950XTX
- 02/08/2007 Review of several Geforce 8800GTX
- 04/27/2007 FAQ on 8800GTX
- 07/09/2007 Overview of G80 and R600 architectures
- 08/01/2007 Overclocking, observations and tricks 8800GTX
- 08/26/2009 HD 4870 1GB vs 8800GTX
models
The history of the legendary G80 dates back to 2004, when Nvidia had just begun designing the latest GPU, which would depart from the usual “pixel-vertex” architecture and introduce the implementation of a new, more flexible, more advanced and advanced architecture — unified.
It should be noted that the implementation of the new architecture in the face of the G80 (or NV50) began almost immediately after the release of the NV40, which at one time gave life to the productive GeForce 6800 line. pipelines and GDDR3 memory brought support for DirectX 9with and Shader Model 3.0
The NV40 had quite a lot of potential. Its architecture was developed several times — first with the release of the G70, which was essentially the same NV40, but “on steroids”, with an increased number of pixel and vertex pipelines, and then with the G71 with a reduced technical process and increased frequencies.
After a long exchange of blows between ATI and Nvidia, when their video cards robbed each other of the championship belt as they grew in power, the Californians finally managed to make a revolution. A real one that hasn’t been seen since the GeForce 3Ti or Radeon 9700. Nvidia has announced the most powerful, hottest, largest and most expensive desktop GPU in history, the G80, which is also the first DX10 compatible video chip and the first streaming architecture desktop GPU. (Before that, in 2005, the ATI Xenos R500 chip was released for the Xbox 360, which has 48 superscalar processors (48×5 = 240 scalar) and absorbed the best from the R580 and R600).
What was so revolutionary about this huge, hot and power-hungry chip? First, the hardware — 128 programmable stream processors that play the role of both pixel and vertex pipelines; 32 texture units, 24(!!!) ROP’s. Add to this a 384-bit bus and 768 megabytes of video memory — a record amount for a single desktop GPU at the time.
Secondly, G80, as already mentioned, became the first DX10 compatible video chip, providing hardware support for Shader Model 4.0
The unified shader architecture surpassed even the wildest expectations — the 8800GTX turned out to be two or three times more productive than the former tops! It was a phenomenal success, because in all respects the new generation of video cards from Nvidia overtook its time by at least a year. Although the design of 8800GTX was not perfect — the length of the PCB had impressive dimensions — 27cm, the chip was manufactured by 90nm process technology, the video card needed as many as two 6-pin power connectors, and a huge cooling system that covered not only the chip itself, but also the power subsystem with memory chips, made the 8800GTX extremely heavy. What can we say about a separate NVIO chip, which, due to the enormous size of the G80, had to be moved outside the main chip. And the full potential of the 8800GTX could only be unleashed on ultra-modern Core 2 Duo processors operating at >3GHz.
But they turned a blind eye to all these shortcomings — the excellent performance magically turned the disadvantages of the 8800GTX into its features
A year later, Nvidia announced a new chip — G92, which, in fact, was the same G80, but with cosmetic changes. Instead of 24 ROP’s, it was decided to leave only 16, a 256-bit bus and higher-frequency GDDR3 memory replaced the non-standard 384-bit one. The amount of video memory was reduced from 768 megabytes to 512, and a thinner 65nm process technology allowed not only to reduce heat dissipation and GPU size, but also to increase clock speeds and integrate NVIO into the main chip. The release of the 8800GTS 512 MB was marred only by its dubious name, but in all other respects it was an excellent product — the PCB was gracefully shortened, the price was not too high, and the performance was very close to the 8800GTX. But here is the rebranding and appearance of 9Nobody expected 800GTX. It was the same 8800GTS 512 mb, but on a longer PCB, with a reinforced power subsystem, a new cooling system, and higher clock frequencies. For some time, it became the flagship of the entire GeForce line, although in fact it turned out to be cut down compared to the 8800GTX\Ultra and, often, could not surpass them in performance.
Nvidia programmers even resorted to “slowing down” the 8800GTX at the driver level, thereby artificially increasing performance 9800GTX.
Two things prevented the absolute success of the 8800GTX. The first is, of course, a rather high price. But the second was more prosaic. The thing is that ATI, already absorbed by AMD by that time, was preparing a super-secret and, according to rumors, super-powerful weapon, codenamed R600. It was the first of the scalar architecture and, like the G80, was in development for a long time. Its alternative, slightly lighter and modified version of the R500 Xenos, by November 2006, was already widely used in the new Xbox 360, demonstrating great features and amazing performance. Thus, a fairly large audience of AMD fans was waiting for a response, which was to follow immediately after the release of 8800GTX\GTS. Many factors spoke in favor of the R600. First, the price — it was an order of magnitude less than the 8800GTX and amounted to about 399$. Secondly, the hardware component is a 512-bit memory bus and 320 scalar processors. It would seem that in all respects, the new product should have outperformed the Nvidia flagship, especially since it was half a year late — enough time to “tighten the screws” and release a product that is far ahead of its competitor in performance. Nvidia even went on the counterattack with the 8800ULTRA, an overclocked version of the 8800GTX. However, time went on, the R600 was late, and when its time finally came, it turned out to be not at all the product that ATI fans were waiting for. In addition to loyal fans, there were also independent citizens who wanted to get the maximum performance and waited for ATI’s response in order to choose the “best of the best”. But the miracle did not happen — video card 2900XT, which became the embodiment of the R600 architecture, turned out to be not so powerful. Yes, its chip was released on 80nm process technology. The length of the PCB was much shorter than that of the competitor, the 512-bit bus provided a throughput of 100Gb / s, and 320 scalar processors in synthetics showed miracles. But if everything was so great, probably you, dear readers, would not read these lines.
As it turned out, the only real trump card of the 2900XT was the memory bus. 512 bits is a lot, and even with fairly low-frequency GDDR3 memory, there was excellent bandwidth, and the video card did not lack it at all. However, only 16 raster operation units and, most importantly, only 16 texture units (the 8800GTX\ULTRA had twice as many), did 2900XT was extremely vulnerable in scenes with massive textures, resulting in a severe FPS drop. In addition, due to R600 architecture peculiarities, video cards based on HD 2900XT crashed loudly in anti-aliased modes. The fall reached 40-45%, while the competitor did not exceed 25%. Scalar processors (of which there were actually 64 (x5)) although they showed good speed, especially in scenes rich in special effects (working with shaders), they could not outweigh the two main drawbacks of the video card and provide a tangible advantage. All that fit 2900XT is the role of the rival 8800GTS 640mb. Even if the competition in terms of performance was quite successful, the timely release of the 8800GTS 320, which has almost identical performance to the 8800GTS 640mb and an extremely tempting price, reduced the few chances of 2900XT for success to almost zero.
The long-awaited revolution failed, 8800GTX\Ultra remained the most powerful video card.
A little later, exactly two months before the announcement of the RV670 (a slightly lightweight R600 made according to the 55nm process technology), ATI put all the R600 chips under the knife, releasing two fairly powerful and very inexpensive products — HD 2900PRO and HD 2900GT. The first differed from the original XT only in a 256-bit bus and slightly lower frequencies (although the first versions of the 2900PRO had a 512-bit bus), and the second, weaker 2900GT product, already lost 80 scalar processors and 4 texture units at the hardware level, having received even lower clock speeds. The result is an attractive price ($249 and $199, respectively) and a performance level that surpasses the entire Middle-End, including 8800GTS 320/8600GTS.
It’s worth noting that despite the failure of the R600 and the dubious success of the RV670, the scalar architecture should not be considered unsuccessful and less perfect than Nvidia’s streaming (or shader, if you like) architecture. The release of RV770 showed that the elimination of weak points — slow AA operation (by redesigning the algorithm) and a small number of texture units, can unleash the full potential of this architecture on the positive side. A gradual «build-up» of the execution units (scalar, texture and raster) makes it possible to achieve an almost linear increase in performance (let’s pay attention to the HD 5870). In addition, the architecture design allows the chip to operate at higher clock frequencies than its competitor (compare at least RV790 and G200b). Although the shader unit of Nvidia chips operates at twice the frequency compared to the GPU.
The WinFast 8800GTX we are reviewing today has been returned from the other world. As many have already guessed, we are talking about a well-known problem — unsoldering some contacts of a huge chip from the PCB board, which leads to the appearance of artifacts, incorrect operation of the video card or even its “death” — the absence of a signal on the monitor. Although, it is worth making a small correction — this leads rather not to “death”, but to “coma”. There are several ways to return the 8800GTX from the other world, but all of them come down to “warming up” the chip, thereby allowing the loose contacts to be “soldered” again:
- “Warm up” the chip under a special IR lamp. Its high temperature will allow the tin BGA balls to take a plastic shape and re-attach to the PCB.
- A more homely way that most people use is to “warm up” the video card in a conventional oven. We get the same effect — the contacts are soldered in place.
But these methods have two big drawbacks. Firstly, using a powerful IR Lamp, you can burn the textolite. Secondly, heating in the oven can melt the plastic parts of the video card — DVI and TV-Out connectors, not to mention the fact that PCB elements can easily “float” due to high temperature — the soldering will melt, and a separate transistor or memory chip will slide off its rightful place. And it’s not a fact that with such a “warm-up” an even greater number of contacts on the chip or board elements are not soldered.
So what to do? The answer is quite simple and elementary — soldered contacts should not be soldered, but simply “pressed”. This is not difficult to do — change the cooling system to something not very expensive and efficient (Zalman VF1000, for example), screw it to the board and tighten the screws harder. As a result, the radiator of the cooling system will press the huge chip to the board, and the contacts will naturally fall into place. You can guarantee 100% of this method only if you are sure that the matter is in the soldered contacts of the chip, and not in a failed memory chip or a burned out NVIO. To check this, it is enough to get hold of a simple CO similar to Zalman VF700. Naturally, it will definitely not be enough to fully cool the chip, but it will be quite enough to boot Windows and pass a short 3D test.
After wiping off a thick layer of dust and giving the board a real alcohol bath with further drying, it returns to its former gloss of 2006. The native cooling system is no longer needed — it will not add efficiency in work, you need to look for a replacement.
It was found quite quickly — a set of branded Zalman heatsinks, specially for 8800GTX\8800GTS — ZALMAN ZM-RHS88 and a great cooler Z-machine GV1000. An excellent choice for the 8800GTX. Not the cheapest of course, but one of the most effective.
However, let’s move on from theory to long-awaited practice. Three years have passed, let’s take a look at what the 8800GTX is capable of today — it has the latest games and two resolutions 1280×1024 and 1680×1050, including modes with AA and AF. The following video cards will be put forward as rivals:
Radeon HD 2900XT — a “repressed revolution”, inferior to 8800GTX from birth. Let’s see how the “unsuccessful response to the G80” will manifest itself today, and how much it will lose to our hero in modern games.
Radeon HD 2900Pro — almost the same card but with a 256-bit bus. We introduce it for pure interest — let’s see how far-sighted the ATI engineers were when they endowed the 2900XT with a 512-bit bus. Will a higher memory bandwidth provide an advantage? Let’s try to figure it out.
GeForce 9800GTX\GTX+ is a direct competitor to 8800GTX. It came out almost a year and a half later and turned out to be cut in terms of memory bandwidth, memory size and raster blocks. But it has higher clock frequencies. So what is more important — hardware stuffing or frequency advantage?
HD 4850 is another direct competitor for today’s hero. This was supposed to be the R600, but, alas, it did not.
HD 4870 — We add this video card as a reference card that provides the highest performance among today’s participants. The 8800GTX will strive for it.
8800GTX — he will perform in two roles. In the first, it will be a standard 8800GTX. In the second, 8800GTX overclocked to 8800ULTRA clock speeds. Well, at least let’s try to disperse it like that …
- 3DMark 2005, 3DMark 2006, 3DMark Vantage. Unremarkable synthetics. From year to year, it is used to roughly estimate the power of a video card. Why approximate? Because the results of synthetics do not always coincide with the results of gaming applications. However, some conclusions can be drawn.
- Crysis is the most demanding game to date. She has no competitors, this is the minimum performance benchmark that any video card should strive for. The higher the FPS, the more power reserve a particular video card has.
- FarCry 2 is a beautiful and special effects-filled game that squeezes all the juice out of video cards. Not Crysis of course, but still one of the most demanding games to date.
- Batman: Arkham Asylum is without a doubt an outstanding game that has won the hearts of many players. It has gorgeous graphics that demonstrate all the beauty of Unreal Engine 3. Since a huge number of games are now being released on this engine, we decided to include it in testing. One of the most demanding games on this engine.
- Risen is not the most perfect game in terms of graphics, but it still has some nice «chips» like good shadows and good lighting.
- Resident Evil 5 is a platform game, but with amazing graphics. To date, one of the most beautiful games.
- GTA 4 is the same Grand Theft Auto 4. The latest patches smoothed out some of the shortcomings and significantly increased the FPS, but the game for the most part still remained quite resource-intensive. Many users, when upgrading, strive to get the highest frame rate in this particular game.
All benchmarks (3DMark) were tested at standard settings, in games the value High or Ultra High (ie the maximum possible) was used for all graphics settings. Optionally, some games were tested in AAx4 AFx16 mode.
Test stand configuration:
- Processor — Core i7 920 3400 MHz
- Motherboard — Asus P6T
- RAM — 3 x 1 GB DDR3 RAM 1333 MHz
- Hard Drive — WD 500 GB HDD
- Power Supply — Thermaltake 650 Watt
Operating system — Windows Vista Ultimate SP2 32 bit
It was decided to overclock the 8800GTX exactly to the clock frequencies of 8800 ULTRA. Why not more? The answer is simple — because, firstly, not all 8800GTX can be overclocked to the ULTRA version frequencies, and conquering its frequencies is already a «feat», and, secondly, this will be quite enough, we will get a significant increase in speed due, mainly , increased frequencies of the shader unit.
So, our hero reached the frequencies of the ULTRA version (612/1512/1080MHz) without any problems, which is very, very good.
Map |
2900Pro |
2900XT |
HD4850 |
HD 4870 |
9800 GTX |
9800 GTX+ |
8800GTX |
8800 ULTRA |
Video chip |
R600 |
R600 |
RV670 |
RV670 |
G92 |
G92 |
G80 |
G80 |
Process |
80nm |
80nm |
55nm |
55nm |
65nm |
65nm |
90nm |
90nm |
Number of transistors, million |
700 |
700 |
956 |
956 |
754 |
754 |
681 |
681 |
Chip area (mm2) |
420 |
420 |
256 |
256 |
330 |
330 |
484 |
484 |
Number of unified processors, pcs. |
320 (64×5) |
320 (64×5) |
800 (160×5) |
800 (160×5) |
128 |
128 |
128 |
128 |
ROP’s, pcs. |
16 |
16 |
16 |
16 |
16 |
16 |
24 |
24 |
Texture blocks |
16 |
16 |
40 |
40 |
32 (64) |
32 (64) |
32 |
32 |
GPU frequency |
601 |
750 |
625 |
750 |
675 |
765 |
576 |
612 |
Shader clock |
601 |
750 |
625 |
750 |
1688 |
1898 |
1350 |
1512 |
Video memory frequency |
1600 |
1660 |
2000 |
3600 |
2200 |
2200 |
1800 |
2160 |
Memory type |
GDDR3 |
GDDR3 |
GDDR3 |
GDDR5 |
GDDR3 |
GDDR3 |
GDDR3 |
GDDR3 |
Memory size |
512 |
512 |
512 |
512 |
512 |
512 |
768 |
768 |
Memory bus, bit |
256 |
512 |
256 |
256 |
256 |
256 |
384 |
384 |
Subscribe to our channel in Yandex. Zen or telegram channel @overclockers_news — these are convenient ways to follow new materials on the site. With pictures, extended descriptions and no ads.
Core clock reduced to 500 MHz, stream processors clocked to 1200 MHz, and 640 MB GDDR3 memory running at 800 MHz (1600 MHz DDR). GeForce 8800GTS will be sold at a recommended price of $599.
So, first, some interesting facts about the GeForce 8800GTX video card, its availability in stores, etc.
At the moment, all available G80-based boards look like twin brothers, the only differences may be in the design of the cooling system. In addition, the entire first generation GeForce 8800GTX and GTS will exactly match the reference design and specifications, any changes in the design of printed circuit boards and frequency parameters of the cards are prohibited by manufacturers. In the future, the release of the second and third «waves» of video cards is expected, and then, perhaps, they will be produced by third-party manufacturers at their own facilities and will delight us with some differences. However, Foxconn’s capacity is quite enough to provide ready-made boards for all manufacturers, and nVIDIA will protect itself from interference by third-party companies in the circuitry of its products. After the third release wave of the G80, an updated core is expected to be released.
It further states that the cards will be available in stores on the day of the announcement. However, they are unlikely to be available everywhere, since the number of video cards is very limited. In particular, according to our sources in the UK, 200 copies of GeForce 8800GTX have been provided for the entire island nation. Interestingly, our colleagues reported in their news feeds that already on Saturday video cards of this generation from Leadtek were sold on the Taipei market for about $800. Most likely, it was either about some dishonest nVIDIA partner who received a video card for trial and testing, or about a banal theft from a warehouse. It is unlikely that Leadtek would risk its reputation for the dubious pleasure of getting a little extra profit.
Another alarming fact: the Singapore site VR-Zone.com reported on Monday that all GeForce 8800GTX are being recalled by manufacturers due to the fact that resistors with the wrong nominal resistance could be installed on them, which may affect the performance of the board. This does not apply to the GeForce 8800GTS. Further clarification of the information made it clear that the defect does not apply to all manufactured video cards, but only to those shipped to several nVIDIA partner companies. In particular, the brands MSI, Gigabyte and eVGA are named, although eVGA is in a hurry to justify itself and declares that the defect was only in the early samples of its production, which will not be found in retail. The Russian representative offices of MSI and Chaintech also announced that boards containing a defect will not enter the Russian market. There is no information about the presence of defects in the products of other companies.
nVIDIA PR man Adam Foat’s interview with Bit-tech. net confirms the defect and reports that some boards are being recalled for resistor replacement. However, according to many analysts, this defect is unlikely to appear in retail cards, as it was discovered even before the bulk of the cards were shipped to partners. Therefore, all boards on store shelves will work as they should.
In general, despite all these ups and downs and clouds that have gathered over the GeForce 8800 on the eve of the release, our laboratory was able to get a working reference GeForce 8800GTX video card at the nVIDIA Moscow office in time to bring as much information about it to you as possible, our dear readers.
Well, let’s get started.
The board is made on a black textolite — a definite departure from the traditional green color for nVIDIA. In our opinion, this color is even pleasant: a strict black textolite gives the impression of a serious powerful product, it’s not for nothing that Asus manufactures its Deluxe class motherboards on black textolite. By the way, all new nVIDIA products presented today are made on black textolite, including reference motherboards based on nForce 600i and nForce 680i SLI chipsets.
The first time we saw the board, we were shocked by its size. Before this, the only board of this size that we held in our hands was the Wildcat Realizm. Once again, you are convinced that a photo is one thing, and the product lying in front of you is a completely different thing. It’s no joke: the length of the board is 265 mm, including the frame. Having inserted it into the case, all the time in the depths of your soul you expect it to fall off the motherboard along with the connector. Not a very convenient video card for those testers who keep the motherboard on the table and stick motherboards into it «for profit», without a case.
Since we got a reference video card, there were no accessories with it.
Video card equipped with two Dual-link DVI outputs and a 7-pin S-Video output that supports AACS protected HD video output and HDMI 1. 3 support.
The cooling system occupies an additional slot on the motherboard. The cooler heatsink is quite complex and should provide good core cooling. The copper insert in the base of the cooler takes heat from the core and transfers it to the aluminum fins of the radiator, which are fastened to it with two heat pipes for better heat transfer. At the same time, the aluminum base of the cooler removes heat from 12 memory chips placed in a separate RAMDAC chip and from the field-effect transistors of the power converter, arranged in four rows of three. The already familiar white thermal pads are responsible for the transfer of heat from the chips to the radiator.
A turbine fan located at the very end of the heatsink blows air through the fins and blows it out of the PC case. Five slots were made on the casing of the cooling system, however, the functional meaning of this solution is not entirely clear to us. Perhaps, at maximum fan speed, the air simply does not have time to escape through the grate in the mounting bracket, and through these slots it will be thrown into the case so as not to reduce the cooling efficiency. The fan is connected to a four-pin connector that allows you to control the speed.
The cooling system is remarkably quiet, while still providing fairly good cooling for the graphics card. The only complaint is that the design does not provide any cooling for some elements of the power circuits, this is clearly seen on this photo:
Having removed the cooler, we were able to examine the printed circuit board and the G80 core in detail.
So, in the middle of the board we see just a huge G80 core. Almost 700 million transistors, even with a manufacturing process with a tolerance of 90 nm, take up a considerable area, besides, separate frequencies for individual parts of the chip should require additional costs. It will be interesting to look at the size of the core after the planned transition to the 80-nm process technology by nVIDIA. The chip is covered with a distribution cover, which improves heat transfer from the core to the cooler due to the larger area, and in addition it also protects the core (most likely, with such a complexity of the core, the number of dummy transistors is extremely small, and chipping the core is easy).
Interestingly, the chip is protected by an additional aluminum frame, fastened with eight screws, so the chance of damaging the chip by misaligning the heatsink is negligible.
Our copy of the chip was made in Taiwan on the 37th week of this year, that is, September 11-17. Quite recently, however, this is already a revision of the A2 core. The core, as promised, operates at a frequency of 575 MHz.
Another surprise for those who do not follow the news on the Web: to the left of the core is another small chip on a green substrate. What versions of its purpose have not been put forward: a chip for processing physics, and a switch for Quad SLI, and even a transitional bridge for the AGP bus, although this is already completely ridiculous … As it turned out, this is a RAMDAC isolated from the core (the unit responsible for image output directly to the monitor). The fact is that shielding the RAMDAC from interference induced by stream processors operating at a frequency of 1350 MHz turned out to be more difficult than just taking it out of the core. Moreover, nVIDIA introduces 10-bit color addressing, and it does not need image distortion due to noise.
12 Samsung K4J52324QE-BJ1A chips are soldered on the front side of the board. These are chips with a capacity of 512 Mbit with an organization of 16×32 Mbit, with a nominal frequency of 1000 MHz and an access time of 1.0 ns. The chips are packaged in 136-pin FBGA Lead-free packages. The memory runs at 900 MHz, which leaves room for overclocking. The total amount of installed memory is 512*12=768 MB.
By the way, these chips are not even on the Samsung website, we could only decipher the characteristics by the code.
Many readers will wonder why NVIDIA didn’t install GDDR4 memory chips on the GeForce 8800GTX. This decision can be explained by several factors at once: the high price of this type of memory in high-capacity chips (512 Mbit each), high heat dissipation and power consumption, and long delays in accessing these chips. For the G80 architecture, fast access to memory cells and data sampling is important, and the strong point of GDDR4 at this stage is the streaming transfer rate, which is not so critical at such frequencies and such a bus width.
We would like to draw your attention to the exceptional complexity of the power supply circuits, already quite comparable to circuits on motherboards. No wonder: for a system that will include one GeForce 8800GTX video card, the necessary requirement for the power supply is a power of 450 W or more and a current of at least 30A delivered through the 12 V line. GeForce 8800GTS is a bit more modest: 400W and 26A respectively. For SLI arrays, PSU monsters are recommended from the list available on the nVIDIA website. Note that the minimum power of the PSU present there is 750 watts. According to preliminary estimates on the Web, one GeForce 8800GTX consumes 170 W (140 W for the core only).
Two 6-pin power connectors are located on the top side of the board to provide power to the graphics card. If your power supply is not equipped with these connectors, 2 adapters from a standard molex connector will be supplied in the retail kit.
Bottom right is a piezoelectric speaker that emits a heart-rending squeak when there is a lack of power or if one of the power connectors is not connected.
At the top left we see two MIO connectors for connecting SLI bridges. “What are two for?”, you ask. We will answer: «For Quad SLI systems.» By connecting video cards in a checkerboard pattern, you can build an array of 4 video cards of the GeForce 8 class, and achieve downright unprecedented performance. It is possible to process physical effects on a third GeForce 7x video card using motherboards based on the nForce 680i chipset, while the board is placed between two main video cards in SLI, a bridge is not needed for it.
We look at the back side of the printed circuit board:
There are a huge number of small elements of the power circuits: resistors, capacitances, field effect transistors and a couple of controllers. A sticker with the BIOS version turned out to be a nice addition. The RoHS sticker was somewhat surprising: even when designing such a complex product, Foxconn was able to refuse components that are potentially harmful to the environment.
We found some new settings in the new drivers that nVIDIA provided with the graphics card. A new option for setting image smoothing has appeared: in addition to the usual Application-controlled and Override any application setting, we see the Enhance the application setting item. This option activates the 8x, 8xQ, 16x and 16xQ anti-aliasing modes and in some cases allows you to significantly improve the quality of anti-aliasing without significant loss in speed. In addition, drivers with DirectX 10 support will have corresponding settings that regulate the newly introduced options.
Well, let’s start testing. In this review, we will only run express testing on the most popular test applications to evaluate the progress of nVIDIA in performance. A full-scale review with a study of the processor dependence of GeForce 8800 video cards will be prepared and published a little later, in which we will also fully evaluate the overclocking potential of the board.
Overclocking
At the moment we managed to overclock the board to 620MHz for the processor and 1900MHz for the memory. That is, the frequencies increased by 45 MHz for the chip and 50 MHz (100 MHz) for the memory. This overclocking potential cannot be called stunning, but overclocking should give a performance increase of at least 5-7%. Pay attention to the fact that only the 3D unit and memory can be overclocked so far, the frequencies of the pipelines cannot be controlled yet, because they depend on the main core frequency.
Our graphics card performance test bench consists of the following components:
- Processor: Intel Pentium D 955 3.4 GHz (dual core).
- Maternal board: Intel D975XBX (i975x)
- RAM: 2×512 MB Kingston DDR2-533
- HARD DIS: Western Digital 80 GB IDE
- Block: FSP600-80GLN
ON rather weak. In the next review, testing will be carried out on a quad-core Core 2 Extreme QX6700 processor, running at a frequency of 2.93GHz.
To compare the performance, we took the most productive single-chip video adapter from ATI today — Radeon X1950XTX by Sapphire. The video card belongs to the Toxic series, is water-cooled and pre-overclocked at the factory to a core frequency of 700 MHz (GDDR4 memory runs at standard 1000 MHz). Representatives of the previous generation of chips from nVIDIA will also take part in the race: GeForce 7900GTX and GeForce 7900GT.
The test package consists of the following applications:
In heavy mode, the picture repeats itself, but we begin to notice a drop in the power of competitors at 1600×1200, which is not observed in GeForce 8800GTX. The third graphical test again puts the beginner on a pedestal, and again the CPU does not allow him to turn around.
According to the overall test results in 3DMark’05, even with a frankly insufficient CPU, we see a solid gain for GeForce 8800GTX. The 9500-9750 points earned by the video card will easily grow to 11-12 thousand points when using more powerful processors.
Futuremark 3DMark’06
In 3DMark’06 the score becomes even more devastating, the gap of 2350 points in hard mode at 1600×1200 in favor of GeForce 8800GTX is very pleasing. It seems that very soon the «top» video cards of the previous generation will begin to move into the middle price segment, cut down and slow down, acquiring the suffixes GS, GTO, etc., which will greatly please buyers, and especially overclockers.
Codecult Codecreatures
In this test, we observe the superiority of nVIDIA architectures over the R580+ core from ATi in a specific profile — a detailed scene with many textures and using multipass texturing. As you know, the strong point of the Radeon X1950XTX — work with complex shaders, which are just not in this engine. As a result, in light mode, we see that all video cards “rested” on the central processor, but Radeon is still somewhat behind. In heavy mode, starting from a resolution of 1280×1024, the competitors of the new nVIDIA start to work at full strength and fall behind, and the GeForce 8800GTX demonstrates even performance within the measurement error at all resolutions, outperforming its competitors by at least 22 percent at 1600×1200. To ATi’s credit, we note that in heavy mode, it is their video card that is closest to the leader.
SpecViewPerf 8
This test package is already quite outdated and in some profiles does not give the expected increase either from the frequencies of the video processor or memory, or from more intelligent architectures. As a result, the G80 only wins in two tests out of five (but by a very noticeable margin), and sometimes even ends up in last place. At the same time, the «emphasis» of the G80 on the central processor is noticeable by the fact that the transition to heavy mode does not affect the performance at all. Interestingly, when anti-aliasing and filtering are activated, the effect is diametrically opposite to that observed earlier: instead of a radical increase in the lead, the GeForce 8800GTX remains in place, while competitors approach it a little in those tests where it leads, but naturally cannot overtake it.
Loss to video cards of the previous generation, most likely due to the fact that the beta version of drivers that support new nVIDIA video cards is not yet debugged and optimized enough.
Well, let’s finally move on to the tests for which these video cards are designed — to games!
Call of Duty 2
In Call of Duty 2 in easy mode, GeForce 8800GTX and Radeon X1950XTX are on par due to insufficient CPU performance. The previous generation nVIDIA graphics cards lag behind by a fairly small amount, which grows with increasing resolution. In heavy mode, the situation remains the same, only in 1600×1200 Radeon X1950XTX is slightly behind GeForce 8800GTX. We emphasize that the first lags behind the second, and not the second overtakes the first, the G80 still shows results that are identical to working in easy mode.
Doom 3, Quake 4
We decided to combine these tests into one section because Quake 4 uses a slightly updated Doom 3 engine, and, as you can see on the graphs, the results in this game are absolutely identical to those in Doom 3. By the way, then the same can be observed in another game based on this engine and used by some of our colleagues — Prey.
A rather strange picture in both light and heavy modes, the GeForce 7900GTX outperforms the newcomer. Only at 1280×1024 and 1600×1200 in heavy mode the competitors lag behind GeForce 8800GTX. Pay attention to the very slight drop in GeForce 8800GTX performance with increasing resolution: a very wide memory bus and high frequency do their job. Such strange results may be related to the “dampness” of the drivers or a banal feature of the processing of this engine by the G80 architecture. Another thing to think about is that the scenes produced by this engine require a lot of vector calculations (bump-mapping, complex work with color), and the G80 consists entirely of scalar processors. The R580+ also has a strong propensity for scalar math work and loses to absolutely everyone in these tests.
Next in our test suite is always First Encounter Assault Recon (F.E.A.R.), which is a favorite test for graphics cards on most sites. However, in the process of testing we came across an inexplicable fact: there was no consistency in the results demonstrated by GeForce 8800GTX. To be sure, we repeated the same test several times, and the results in most cases were in the range of 120-130 fps (oh, how nice it sounds, 130 fps in F.E.A.R.), however, repeatedly, they inexplicably dropped to 60 fps. Therefore, at the moment, before the appearance of the official version of the driver and, possibly, before the adaptation of the F.E.A.R. For the G80 architecture, we remove this test from the agenda. It’s a pity.
Well, let’s not grieve over this for a long time, but follow the performance of the new product in the Source engine, on which the Half-Life 2 game is based.
Half-Life 2
series, Source’s love for the ATi R5xx architecture remains strong. Now the Radeon X1950XTX acts as a tank, «resting» against a slow processor and noticeably outperforming all competitors, including the G80. In heavy mode, the situation improves with increasing resolution: the power of the memory subsystem and 64 texture units save the beginner.
Chronicles of Riddick: Escape from Butcher Bay
Finally, the game where the GeForce 8800GTX won with a crushing score, outperforming its competitors by at least 1. 5 times! Still, the game engine is rich in shaders and was originally developed for game consoles, so the unified shader architecture of the G80 suits it perfectly. Still, the GeForce 8800GTX is limited by the performance of the CPU: both in easy and hard modes, the performance is almost identical. Only at 800×600 in light mode we see an incomprehensible performance gap. We suspect that with the replacement of the processor, the leader’s lead will grow to indecent proportions and exceed the two-fold mark.
Processor dependence
In fact, the new GeForce 8800GTX 3D accelerator is not only a new era in the field of 3D graphics processing, it is also a new bar for other components of a modern PC. The thing is that the performance of 3D graphics accelerators every year is more and more ahead of the performance of central processors, and this cannot but affect the overall performance in general. Any powerful 3D accelerator needs a powerful CPU, and in the case of the GeForce 8800GTX, the CPU performance should be no less than that of the recently announced Intel QX6700, and it can easily be not enough to provide such a powerful video subsystem with data. There is no need to talk about the GeForce 8800GTX Quad SLI or even just SLI, since in this case no adequate processor has yet been invented.
That’s why you need to firmly understand that the nVIDIA GeForce 8800GTX, like any precious stone, needs the right cut and setting, that is, it will be able to show its power on existing applications only in the toughest modes and the highest resolutions. Everything else is just tests for processor dependence of drivers and they have a very distant relation to the performance of the tested accelerator. In most cases, on our CPU, the nVIDIA GeForce 8800GTX showed its power only at 1280×1024 and 1600×1200 resolutions with anti-aliasing and anisotropy enabled.
In the future, the situation may change, as game creators are well aware of the current situation and will try to shift the maximum amount of load to the graphics processor, since DirectX 10 has all the possibilities for this.
Conclusion
The video card GeForce 8800GTX presented today can rightfully be called the most productive single video card in the world today. In most test applications, it won a convincing victory over the direct competitor from ATi (of course, the R600 core will be the direct competitor of the G80, but there is still a lot of time before its presentation, but for now the most powerful video card from ATi is the Radeon X1950XTX). The superiority over GeForce 7x video cards is also very significant, and allows us to hope for an early price reduction for top-end video cards of this generation and their transition to the middle price range.
Let’s note some difficulties that a buyer will face when he decides to purchase (and finds in stores) this video card. To date, it is supported only by the beta version of the ForceWare driver, which is alarming due to its frankly disastrous performance in some games. In addition, game developers will have to do additional work and release patches that improve the compatibility of game engines with the new architecture. Another difficulty is that for balanced operation the GeForce 8800GTX requires a very powerful CPU and a high-quality powerful power supply, so the final price of a system using it will not be at all budgetary.
In the meantime, we can only congratulate nVIDIA Corporation for being the first in the world to release a graphics card that fully supports DirectX 10, unconditionally supports the upcoming Windows Vista, and takes its rightful royal place on the podium of the most powerful graphics cards. It remains only to wait for the final products to appear in stores and look at the prices, which in the first couple of months, according to the laws of the market, will be very slightly similar to the recommended ones.
performance overview and gaming performance tests
The GeForce 8800 GTX video card was released by NVIDIA, released on November 8, 2006. At the time of release, the video card cost $599. The video card is designed for desktop computers and is built on the Tesla architecture codenamed G80.
Core frequency — 576 MHz. Texturing speed — 36.8 billion / sec. Number of shader processors — 128. Floating point performance — 345.6 gflops. Technological process — 90 nm. The number of transistors is 681 million. Power consumption (TDP) — 155 Watt.
Memory type: GDDR3. The maximum memory size is 768 MB. Memory bus width — 384 Bit. Memory frequency — 900 MHz. The memory bandwidth is 86.4 GB / s.
Benchmarks
PassMark G3D Mark |
|
|||||
PassMark G2D Mark |
|
|||||
GFXBench 4.0 T-Rex |
|
|
||||
GFXBench 4. 0 T-Rex |
|
|
Name | Meaning |
---|---|
PassMark — G3D Mark | 597 |
PassMark — G2D Mark | 105 |
GFXBench 4.0 — T-Rex | 3349 Frames |
GFXBench 4.0 — T-Rex | 3349.000 Fps |
Features
Architecture | Tesla |
Codename | G80 |
Production date | November 8, 2006 |
Price at first issue date | $599 |
Place in the ranking | 1144 |
Type | Desktop |
Core frequency | 576MHz |
Number of CUDA conveyors | 575 |
Floating point performance | 345. 6 gflops |
Process | 90nm |
Number of shaders | 128 |
Texturing speed | 36.8 billion/sec |
Power consumption (TDP) | 155 Watt |
Number of transistors | 681 million |
Video connectors | 2x DVI, 1x S-Video |
Interface | PCIe 1.0 x16 |
Length | 270mm |
Additional power connectors | 2x 6-pin |
DirectX | 10.
|