AMD A8-3870K Review | bit-tech.net
Written by
Antony Leather
February 6, 2012 | 08:17
Tags: #apu #comparative #cpu #f1 #gpu #llano #onboard-graphics #overclocking #performance #processor
Companies: #amd #hudson #intel
1 — AMD A8-3870K Review2 — AMD A8-3870K Test Setup3 — AMD A8-3870K Cinebench and WPrime Performance4 — AMD A8-3870K Image Editing and Video Encoding Performance5 — AMD A8-3870K Multi-tasking and Overall Performance6 — AMD A8-3870K Gaming Performance7 — AMD A8-3870K Power Consumption8 — AMD A8-3870K Performance Analysis and Conclusion
Manufacturer:AMD
UK Price (as reviewed):£104.39 (inc VAT)
US Price (as reviewed):$144.99 (ex TAX)
Our first look at AMD’s A-series APUs was a very positive one, with the A8-3850 proving to be capable of playing games at reasonable settings while also adding a fairly good smattering of other results in video editing and image editing benchmarks. While Intel has the lead in the latter, with its Sandy Bridge-based Pentium and Core i3 CPUs outperforming the A8-3850 by quite some margin, the AMD chip was on a different level when it came to games, making it a fantastic all-round APU that’s quite at home without a discrete graphics card.
Sadly, availability has plagued the A-series and for the last few months it’s been practically impossible to buy them. While we might have expected this anyway given its popularity (well, maybe not — even excellent products should actually be available), but unfortunately for AMD, the short supply was caused by 32nm yield issues at Global foundries — the fabrication plant responsible for manufacturing the A-series.
Click to enlarge
The A8-3850’s life was cut short, which is a shame because we suspect it would have sold bucket-loads, grabbing a handy share of the super-budget PC market in the process. This wasn’t to be, but thankfully, AMD appears to have sorted out supply issues with its latest APU in the range, the A8-3870K — there are currently plenty available from all the usual retailers.
What’s more, the new APU also sports an unlocked multiplier, meaning it can potentially be even better than the A8-3850. Amazingly it costs about the same too — at £104 at time of writing, this is barely any more expensive than its ill-fated sibling.
The unlocked multiplier also spells an end to issues that have surrounded overclocking these APUs in the past — AMD itself told us success was very motherboard dependant. If the A8-3870K works as well as it should, this will mean that even budget Socket FM1 boards could unleash some much needed extra grunt, which could see it gain ground on Intel’s budget CPUs in 2D tests.
Aside from the unlocked multiplier, the A8-3850 and A8-3870 aren’t much different — the latter has received a frequency boost to the CPU of 100MHz and now stands at a nice round 3GHz. This has the same amount of L1 and L2 cache (128KB L1 each core and 1MB L2 each core) and it has the same Radeon HD 6550D GPU under the hood, clocked at 600MHz.
Specifications
- Frequency 3.
0GHz - Core Stars
- Manufacturing process 32nm
- GPU AMD Radeon HD 6550D
- Number of cores 4 x physical
- Cache L1: 128KB (each core), L2: 1MB (each core)
- Packaging FM1
- Thermal design power (TDP) 100W
- Features SSE, SSE2, SSE3, SSE4a, AMD64, Cool’n’Quiet 3.0, AMD-V, MMX, UVD3
1 — AMD A8-3870K Review2 — AMD A8-3870K Test Setup3 — AMD A8-3870K Cinebench and WPrime Performance4 — AMD A8-3870K Image Editing and Video Encoding Performance5 — AMD A8-3870K Multi-tasking and Overall Performance6 — AMD A8-3870K Gaming Performance7 — AMD A8-3870K Power Consumption8 — AMD A8-3870K Performance Analysis and Conclusion
AMD A8-3870K: Coming Close To Radeon HD 6670 — Professional Help: Getting The Best Overclock From AMD’s A8-3870K
Skip to main content
Tom’s Hardware is supported by its audience.
When you purchase through links on our site, we may earn an affiliate commission. Here’s why you can trust us.
AMD’s Radeon HD 6670 DDR3 earns one recommendation after another in our monthly Best Gaming Graphics Cards For The Money column because, at $65, it represents a great baseline for gamers on a budget. It can handle most games at 1920×1080 using low detail settings, and more demanding games are still playable at 1680×1050 or 1280×720.
Knowing that, we must admit to being surprised that an overclocked A8-3870K could come so close to our entry-level discrete recommendation. For $75 more than the graphics card alone, you get a quad-core processor and an on-die graphics engine capable of impressive feats, providing that you have fast-enough memory to let it shine. The good news is that 2000+ MT/s RAM has never been cheaper, and is now within the budget builder’s reach.
We tested five retail A8-3870K samples in order to generate our results, and we saw very consistent outcomes.
All five chips managed at least 900 MHz on their GPUs and 3.5 GHz from their x86 cores. Our best samples muster just over 960 MHz and 3.7 GHz from both subsystems, respectively, so there doesn’t seem to be a lot of variance. We think it’s reasonable to expect similar overclocking potential from most A8-3870Ks, so long as you invest in a decent aftermarket cooler.
How about a counter-point? Intel’s $70 Pentium G630, paired with a $65 Radeon HD 6670 DDR3, slightly cheaper RAM, and a low-cost motherboard promises slightly better performance in the games we tested. That’s not much of a surprise, though, considering the A8-3870K couldn’t quite match the Radeon in this story. The Pentium should also be faster in single-threaded apps, in addition to using less power all around. In fact, we ran a few tests on the new overclocked machine and compared them to the results we generated in February. The stock Pentium-based platform with AMD’s Radeon card turn out to be slightly faster than our most aggressively-overclocked Llano-based configuration.
Having said that, the A8-3870K’s four cores generate much better numbers in many threaded apps compared to the dual-core Pentium, especially when it comes to media encoding.
When the smoke clears, AMD’s A8-3870K is more compelling now than ever before, thanks in no small part to lower prices on high-performance memory. We were able to play three out of the four games we tested at 1920×1080 at smooth frame rates. Battlefield 3 nudged us down to 1280×720, but that remains an acceptable resolution on many high-def televisions. There’s definitely a lot more to this APU once it’s in the hands of an expertly-guided overclocking enthusiast.
- 10
Current page:
AMD A8-3870K: Coming Close To Radeon HD 6670
Prev Page Power And Temperature
Tom’s Hardware is part of Future US Inc, an international media group and leading digital publisher. Visit our corporate site .
©
Future US, Inc. Full 7th Floor, 130 West 42nd Street,
New York,
NY 10036.
AMD A8-3870K, A10-6800K and A10-7850K processors
All three AMD integrated platforms in top performance
year, at the end of 2012, it already began to give way to processors and boards with Socket FM2. But the Socket FM2 platform has not formally healed either — for about a year now, the phased introduction of FM2 + has been going on. However, this implementation is very gradual — the two platforms are compatible from the bottom up, i.e. older processors can be used on new boards, but not vice versa. Only new ones are suitable for new APUs, but the company is not expanding their range too hastily — for almost a year, the A10-7850K continues to be the fastest model, which is still not replaced, but in the younger families, the new architecture is almost not represented at all.
Is it even worth chasing after her? The company talks a lot about the HSA beneficiaries and the bright future of heterogeneous systems in general, but, in truth, real progress in this area continues to bear very modest results.
Yes, of course, compared to the situation three years ago, the use of OpenCL has become at least a little noticeable in mass software, but the history of GPGPU implementation has been going on for more than five years. All these years, we have heard promises that tomorrow heterogeneous computing will become massive and necessary for every user, but in fact, those who bought computers in those years already managed to change them 🙂 Not all, of course, but those who tried to get maximum performance for sure. But those who did not strive to conquer the peaks were hardly in a hurry to jump from platform to platform. Therefore, it became interesting to us — what are the «tops» of each of them? To find out, it is enough to test only three processors, which we will do today.
The A8-3870K is the fastest FM1 device, replacing the A8-3850 in the spring of 2012.
And at the end of the same year, as mentioned above, FM1 was replaced by the FM2 platform and a new flagship in the face of the A10-5800K, which in the summer of 2013 gave way to the A10-6800K. About half a year later, it turned out that FM2 would not be able to work with processors on new cores, so I had to switch to FM2 +, retaining, however, backward compatibility. But formally, the A10-7850K has become the top AMD APU, and there is no replacement yet. We have already tested it recently, however, due to the peculiarities of the motherboard used, the memory frequency turned out to be 1333 MHz, and not 2133 MHz, which, of course, affected many results and required the device to be “remeasured” 🙂
Thus, we got three processors for two and a half platforms, which we will be testing today.
Test methodology
To evaluate performance, we used our performance measurement methodology using iXBT Notebook Benchmark v.1.0 and iXBT Game Benchmark v.1.0. We normalized all test results in the iXBT Notebook Benchmark v.
1.0 against the results of Pentium G3250 with 8 GB of memory and Intel SSD 520,240 GB, while the method of calculating the integral result remained unchanged. Another program that we added to the test suite as usual is the Basemark CL 1.0.1.4 benchmark, created to measure the performance of OpenCL code.
iXBT Notebook Benchmark v.1.0
The application is able to load at least 16 computation threads, and it actively uses OpenCL, so the results of all three APUs differ significantly. As expected, the 7850K turned out to be the fastest, but it is somewhat embarrassing that its advantage over the 6800K is only 5%, while the transition from FM1 to FM2 can give more than 15%.
In content creation, the newest AMD APU loses to its predecessor. And how do these programs differ from converters? They also make good use of processor cores, but the contribution of the GPU here is much less significant. Thus, it can be argued that the processor part even became slower when switching from 6800K to 7850K, and all cases when the latter manages to win (or at least not lose) are mainly associated with heterogeneous calculations.
Whether this is true or not, we will check with the help of other tests. In the meantime, we note that the transition from FM1 to FM2 is still capable of providing about 15%: this is too little if the first platform is already available, but not so bad in global terms.
Photoshop is not one of the programs that use OpenCL too actively, but, nevertheless, the 7850K did the job a little faster in it. Well, it means that sometimes the architecture improvements in Kaveri affect so much that they are able to overcome the clock frequency lag. But the «athlon-like» architecture of Llano is a definite outsider here — 30-35% of the difference speaks for itself.
But sometimes the A10-7850K’s stage presence ends in disgrace — as in this case. Of course, he overtook the A8-3870K, but he did it sluggishly and reluctantly, settling almost in the middle between the mentioned «old man» and also not the new A10-6800K. However, even the latter falls far short of the banal budget Pentium G3250 (which, remember, in this cycle we took as a starting point of 100 points), which is due to the application’s not too high need for multithreading.
In a word, the performance of each computational thread in AMD processors is still very low, which can only be compensated for by high clock frequencies. And you can not compensate 🙂
But it also happens that four «half-cores» overtake four cores, even when fully engaged. Moreover, four «half-cores» of a lower clock frequency can overtake similar ones operating at a higher one. True, quite a bit, and, again, in fact, we are talking only about competition with dual-core Intel processors. Moreover, with «completely dual-core» — such as Celeron and Pentium, since Core i3 with its four computation threads will be faster.
The situation is similar to the previous case, although here a newer processor is also slower. In general, it’s easier to think of them as roughly equal on average — sometimes a little faster than one, sometimes a little faster than the other. The better devices for FM1 are faster, but this was achieved, we repeat, already at the stage of transition from FM1 to FM2, and the subsequent tuning of the platform did not increase the advantage.
To test two out of three processors, we used one motherboard, so the conditions here are as equal as possible. Yes, and the result is about the same. The A8-3870K worked on an older motherboard with an old chipset and slow memory, but in principle nothing can be changed here, so we can state that the «in everyday life» systems with it are about 20% slower — similar to other tests.
The result is logical: changing FM1 to FM2 provides potentially about 20% of performance and allows you to at least break away from the Pentium, while the introduction of FM2 + does not give anything at all 🙂 As for fast memory, it is also useful for general-purpose applications: DDR3-2133 provided 21% superiority over the Pentium G3250, and in our earlier tests with DDR3-1333, the A10-7850K outperformed this processor by only 6% in the overall standings.
OpenCL
Switching to Kaveri can give something with the active introduction of heterogeneous computing — as you can see, the speed of interpreting OpenCL code in the A10-7850K is almost one and a half times higher than that provided by the A10-6800K, and A8-3870K it overtakes almost twice.
An excellent result, which spoils only one thing — in practice, most of the time, processors still have to «turn» the usual x86 code. It is not possible to «accelerate» using GPGPU everywhere, and even where possible, improvements have only a local effect. In other words, if the code that takes 10% of the total execution time is accelerated even by an order of magnitude, only 9 will be saved.% time. It doesn’t matter what you should always keep in mind when considering ideal cases: in specialized benchmarks or algorithms that fit especially well on GPGPU.
Low resolution games
The 1366×768 mode is the second most popular among gamers according to Steam statistics, but this is primarily due to the popularity of laptops (including those that allow you to somehow play games). For desktop systems, it is somewhat atypical, but if you really have to rely on the strength of the built-in video core, then there is no time for fat.
This is what the first game we took for comparison immediately demonstrates: as we see, even at low resolution and at minimum settings, we can get good, but not ideal, performance.
Moreover, this can be done even on the old A8, and the newer A10 only increase the «margin of safety», but nothing more: the high-quality mode is far away even for the A10-7850K.
The situation is similar in Bioshok. And it doesn’t matter that in the quality mode the A10-7850K outperformed the A10-6800K by as much as 20% — there are not enough of them for such a mode. As quality decreases, as we can see, the difference in performance quickly decreases.
Here you can drive in «tanks» in the quality mode. The A8-3870K copes with this, however, without much margin, which is likely to be true for other A8s, but the A10 is quite enough. And there is not much point in reducing the quality — the game is very processor-dependent, and its engine is still radically single-threaded, so you should not expect any mind-blowing results from AMD processors, regardless of the video card used. And with that said, there is nothing surprising that the fastest was not 7850K, but 6800K;)
Although the game itself works fine on the A8-3870K, the benchmark for it does not run on many processors, since it requires mandatory support for AVX commands.
Therefore, you can only compare two A10s, but this is not very interesting — both of them are “too much” for the minimum settings, but not enough for the high quality mode. In general, something in between will be used in practice 🙂
Metro at the very least works on A8 and well — on A10, but here it will not be possible to aim at high quality mode very soon.
But Hitman’s requirements are even higher, so the A10 (not to mention the A8) doesn’t have a special margin even «at the minimum». But you can play on all processors.
In general, the conclusion is simple — at low resolutions and at minimum settings, all APUs starting from A8 cope with most games. The appearance of A10 along with the release of FM2 increased the performance of the platform, the transition to FM2 + has not yet yielded anything. But high settings are not yet possible to use even in this resolution. More precisely, in old and / or simple games — quite, but nothing more.
High-definition games
So, we won’t be able to increase the quality yet.
Is it possible to increase the resolution to normal at least on the desktop A10?
FPS on the verge of «playability», but on the right side of this line. Of course, «at the minimum», but at least that’s not bad.
The situation is even better — here, without any exaggeration, the normal frame rate.
In «tanks» there is even some margin for quality improvement. However, the game is very dynamic, so there is no need for FPS drawdowns, so it’s better not to chase visual effects. But here the resolution can not be reduced — it’s already good.
On the other hand, you don’t need too high a frame rate in racing circuits, so you can already improve the quality here. Not to the maximum, of course, but “at the minimum” it is not necessary to suffer.
And two games are still not giving up even on the top A10 and even on the minimum settings. However, you can probably play Hitman: 28-29 average FPS is enough for this, in principle, especially since the minimum frame rate does not fall below 20.
Thus, the general verdict is simple — in principle, you can play at minimum settings with FHD resolution not in all, but in many games. Naturally, in this case, we are almost never talking about high quality. With the exception of games with very very weak graphics, which is natural.
Total
As you can see, despite all the “progressiveness”, processors based on the Kaveri core are absolutely not interesting at least for owners of boards with an FM2 socket: to switch to them, you will have to change the board, while older Richland models are able to provide the same the level of performance of the processor part and almost the same graphic and without replacement. And given the current retail prices, it can even be argued that desktop Kaveri is not needed … by anyone. Indeed: A10-7850K is much more expensive than A10-6800K. There is an opportunity to save even more — by purchasing A10-6790K or A10-5800K in general: they are slower due to the lower clock frequency, but much cheaper — one and a half times or more.
Well, the frequency lag can be compensated for by a little overclocking, since all the mentioned APUs have all the multipliers unlocked. A10-7850K, of course, can also be overclocked a little, but it will be too expensive to pay for it. So much so that the price is comparable to, for example, a set of Athlon and a fast (at least much faster than any integrated graphics) discrete graphics card, the use of which is quite possible even in some mini-ITX systems, not to mention MicroATX. In general, with such subtle improvements, the A10-7850K and its counterparts should cost the same as their predecessors, but not more expensive — only in this case the purchase will be justified.
Hard? Yes, but that’s the way it is. Moreover, we note that above we spoke only about intracompany competition, without provoking the «wars for faith» so beloved by many — adding processors from Intel to the list of considered solutions can only aggravate the situation. But today we will not deal with this, since we are mainly limited to AMD processors (or rather, APUs).
So let’s just summarize that even for a refined fan who chooses strictly from the assortment of one of the main vendors, there are almost no cases in which the purchase of a device for FM2 + is justified at the moment. Although this does not negate the purchase, other things being equal, of a board with this connector, it is quite possible that the situation will change in the future. But hardly much — the radical improvement of the GPU in practice is offset by the low speed of working with memory. It’s better not to save money on the latter, but also not to hope for improvements in the future: the increase in frequencies will be associated only with the transition to DDR4, which will most likely still require a new socket.
Assessing the company’s progress over three years… In general, not too high. Yes, of course, Llano processors used a very archaic microarchitecture inherited from the 2008 Phenom II development (and in practice dating back to even earlier times), so the update of the processor part was very relevant.
However, in practice, it increased performance by only about 20%, and this was already done in the Trinity / Richland APU, i.e. at the first step of the update — the second did not give anything at all. The graphical part has also improved, which made it possible to “build on” the A4-A6-A8 line with another step in the form of A10, but this was already done during the development of Trinity — there is almost no practical benefit from Kaveri, since GPU performance already greatly limits memory performance. Well, taking into account the fact that, as mentioned above, processors for the «full» FM2 + are still too expensive, the effect of the latest upgrade of the APU family is generally negative.
In general, as we can see, not only Intel, but also AMD has been marking time for a long time. But who is bigger is not clear yet: for this, it is necessary to test in practice the decisions of the largest processor manufacturer, which we will deal with in the near future.
Review of the AMD A8-3870K processor: specifications, tests in benchmarks
The A8-3870K processor was released by AMD, release date: December 2011.
The processor is designed for desktop computers and is built on the Llano architecture.
Processor unlocked for overclocking. The total number of cores is 4. The maximum clock frequency of the processor is 3 GHz. Technological process — 32 nm. Cache size: L1 — 128 KB (per core), L2 — 1024 KB (per core).
Supported memory type: DDR3.
Supported socket type: FM1. The maximum number of processors in the configuration is 1. Power consumption (TDP): 100 Watt.
Benchmarks
| PassMark Single thread mark |
|
|||||
| PassMark CPU mark |
|
|
||||
| Geekbench 4 Single Core |
|
|||||
| Geekbench 4 Multi-Core |
|
|||||
| CompuBench 1. Face Detection |
|
|
||||
| CompuBench 1.5 Desktop Ocean Surface Simulation |
|
|
||||
| CompuBench 1.5 Desktop T-Rex |
|
|
||||
| CompuBench 1.5 Desktop Video Composition |
|
|
||||
| CompuBench 1.5 Desktop Bitcoin Mining |
|
|
||||
| GFXBench 4.0 Car Chase Offscreen |
|
|
||||
| GFXBench 4.0 Manhattan |
|
|
||||
GFXBench 4. 0 Car Chase Offscreen |
|
|
||||
| GFXBench 4.0 Manhattan |
|
|
237 Frames/s 
344 mHash/s | Name | Meaning |
|---|---|
| PassMark — Single thread mark | 1236 |
| PassMark — CPU mark | 2331 |
| Geekbench 4 — Single Core | 373 |
| Geekbench 4 — Multi-Core | 1172 |
CompuBench 1. 5 Desktop — Face Detection |
2.914 mPixels/s |
| CompuBench 1.5 Desktop — Ocean Surface Simulation | 9.482 Frames/s |
| CompuBench 1.5 Desktop — T-Rex | 0.191 Frames/s |
| CompuBench 1.5 Desktop — Video Composition | 0.400 Frames/s |
| CompuBench 1.5 Desktop — Bitcoin Mining | 3.344 mHash/s |
| GFXBench 4.0 — Car Chase Offscreen | 762 Frames |
| GFXBench 4.0 — Manhattan | 1721 Frames |
| GFXBench 4.0 — Car Chase Offscreen | 762.000 Fps |
GFXBench 4. 0 — Manhattan |
1721.000 Fps |
Features
| Architecture name | Llano |
| Production date | December 2011 |
| Place in the ranking | 1990 |
| Applicability | Desktop |
| Support 64 bit | |
| Crystal area | 228 mm |
| Level 1 cache | 128 KB (per core) |
| Level 2 cache | 1024 KB (per core) |
| Process | 32nm |
| Maximum frequency | 3 GHz |
| Number of cores | 4 |
| Number of transistors | 1178 million |
| Unlocked | |
| Supported memory types | DDR3 |
| Maximum number of processors in configuration | 1 |
| Supported sockets | FM1 |
| Power consumption (TDP) | 100 Watt |
| AMD Virtualization (AMD-V™) | |
Navigation
Select Processor
Processor Comparison
AMD A8-3870K vs.
