155GB/s DDR4 12900K AIDA64-Cache and Memory Benchmark
JavaScript is disabled. For a better experience, please enable JavaScript in your browser before proceeding.
My word, 155GB/s (!?!?)
See less
See more
1
Reply
Save
Like
1 — 18 of 18 Posts
If you want fake Aida numbers:
Aida64 is useless when overlocking BCLK..
See less
See more
Reply
Save
Like
Reactions:
5
Not legit. You need at least 7500MHz memory to even be within striking distance. There is a mathematical formula you can follow.
Estimating his real numbers are probably ~80GB/s
Reply
Save
Like
Reactions:
1
You just gotta divide the bandwidth by the difference in BCLK from 100.0, and multiply the latency by the same difference.
Reply
Save
Like
Reactions:
1
Aida is about as useful as userbenchmark.
Reply
Save
Like
Reactions:
1
bhav said:
Aida is about as useful as userbenchmark.
Click to expand…
At this point, I completely agree.
Reply
Save
Like
Anyway 4800CL17 is around 73-75k bandwidth in valid results (Intel MLC), scaling proportionately from 3600CL16ish 59k, 4000-4200 62-65k.
Reply
Save
Like
storm-chaser said:
At this point, I completely agree.
Click to expand…
And good god wait what 28ns on a Z690 at 4800, in G2 on DDR4???
🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪
Reply
Save
Like
bhav said:
And good god wait what 28ns on a Z690 at 4800, in G2 on DDR4???
🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪
Click to expand…
right lol
Wouldn’t that just be peachy
Reply
Save
Like
storm-chaser said:
right lol
Wouldn’t that just be peachy
Click to expand.
..
Or is OP first person in the world running DR Samsung B die 4800CL17 in Gear 1
Reply
Save
Like
Reactions:
1
Who’s up for fake #s aida64 competition?
😆😆😆
Reply
Save
Like
Idk, it could be real, hard to really say. I know Iv been playing with the B clock and had a pretty big improvement going from 100-104. So going to 219, Which i have to say seems alittle impossible, But im a novice at messing with B clocks. Anyways here is my results on my 11900k using gear 1 and the using all the same timings, just changed it from 100 to 104.
See less
See more
Reply
Save
Like
Reactions:
1
Taint3dBulge said:
Idk, it could be real, hard to really say. I know Iv been playing with the B clock and had a pretty big improvement going from 100-104. So going to 219, Which i have to say seems alittle impossible, But im a novice at messing with B clocks.
Anyways here is my results on my 11900k using gear 1 and the using all the same timings, just changed it from 100 to 104.
View attachment 2587278Click to expand…
The BCLK bug I think is only present on 12th / 13th gen.
Reply
Save
Like
Taint3dBulge said:
Idk, it could be real, hard to really say. I know Iv been playing with the B clock and had a pretty big improvement going from 100-104. So going to 219, Which i have to say seems alittle impossible, But im a novice at messing with B clocks. Anyways here is my results on my 11900k using gear 1 and the using all the same timings, just changed it from 100 to 104.
View attachment 2587278Click to expand…
Nope it’s impossible to hit those numbers with a DDR4 dual channel kit. Keep in mind, this result pretends to beat quad channel DDR5 7000 in memory bandwidth. That is astronomically wrong.
Also my Z840 with 8 channel effective memory throughput doesn’t even come close either.
These numbers are more realistic, if you are interested in what the bandwidth/latency SHOULD actually look like.
See less
See more
Reply
Save
Like
Wool Clad Wolf said:
Click to expand…
storm-chaser said:
Nope it’s impossible to hit those numbers with a DDR4 dual channel kit. Keep in mind, this result pretends to beat quad channel DDR5 7000 in memory bandwidth. That is astronomically wrong.
Also my Z840 with 8 channel effective memory throughput doesn’t even come close either.
These numbers are more realistic, if you are interested in what the bandwidth/latency SHOULD actually look like.
View attachment 2587288Click to expand…
Damn, that’s one mean 9600kf,
I just ran it with Default/auto bios setting
domdtxdissar said:
If you want fake Aida numbers:
View attachment 2587260
Aida64 is useless when overlocking BCLK..
Click to expand…
that’s what I assumed was most likely to blame, but also the only reason my IMC’s are
bhav said:
And good god wait what 28ns on a Z690 at 4800, in G2 on DDR4???
🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪 🤪
Click to expand…
8 CORES 8THREADS,
~4700ish G1 in bios but my windows boot manager couldn’t find
Reply
Save
Like
Reactions:
1
Wool Clad Wolf said:
~4700ish G1 in bios but my windows boot manager couldn’t find
Click to expand…
No it isnt!
Reply
Save
Like
was since tinkering with a kinda posting ~4700mhz ~ish G1 in bios, but my windows boot manager was unable to locate my fresh windows install,, i may of blundered the RST driver setup during the iso install or,,😅
When checking for newer RST driver on MB webpage a Fresh BIOS «»»Improved DRAM compatibility»»»
That’s one mean 9600kf had one on an Asus z390ws pro with the «not» gen3. 1/gen4 pciex4 slot and optimizied DMI routing THB4 aic DP1.4alt(+DSC?) almost fed my 5k-g9 via a 1070 as well as my 3080ti is now,, and it just workeed,,, a ****ing mess,, intel probably was **** with recent chipset support so there never realised ARC would have something special, then again that board was selling >$1000 used when i looked to replace mine.was hoping
Wool Clad Wolf said:
was tinkering with a kinda posting ~4700ish G1 in bios, but my windows boot manager was unable to locate my fresh windows install,,, but
I just ran it with Default/auto bios setting Looks bad >:80 latency and it crashed
Bclk’s the best I just love having to toggle the PSU’s power switch because the system «can’t» fully restart itself. @ Least it sometimes cuts a chassis fan to let me know when it’s finished memory trainings…
Bclk of course,, adjustments often don’t get factored in ..ie. Task managers Perf Monitor-Cpu Freq. You have to give it to AIDA mofos count twice!?!
Assumed it was miss calculating, something.. but BCLK is also the only reason my IMC’s are only this ****?
~~1725mc // 3550Mhz ddr4 was my previous Gear1 booting config
8 CORES 8THREADS, My E cores Are beach quality silicon,, my best E is stuck in a cluster with my three wosrt,,, so Per E core mostly just conflicts with intels per cluster i know how many of the enabled ones are on but not witch ones….
~4700ish G1 in bios, but my windows boot manager was unable to locate my fresh windows install,,, but
Click to expand…
See less
See more
Reply
Save
Like
Wool Clad Wolf said:
That’s one mean 9600kf
Click to expand…
Thanks! It’s currently the highest clocked 9600K/KF on HWBOT not using liquid nitrogen.
See less
See more
Reply
Save
Like
1 — 18 of 18 Posts
- This is an older thread, you may not receive a response, and could
be reviving an old thread. Please consider creating a new thread.
Top
How Low Can You Go? AIDA64 Memory Latency Challenge | Page 4
JavaScript is disabled. For a better experience, please enable JavaScript in your browser before proceeding.
Leaderboard Updated December 11th 12:00pm
Got to keep myself busy with comps.. And plus after the competition is over, this thread can still be used to post your latency #s as well.
I figured it was time for something AIDA64 related — and that’s memory latency. I came to this conclusion after working with some experts over in the DDR4 stability thread who were kind enough to help me earlier today get my memory 100% dialed in. Obviously, I am not going to win this, but at least I have a horse in this race now, so it makes the idea more appealing.
1)Include AIDA64 Cache and Memory benchmark snip
2)Include CPU and MEMORY and SPD tabs of CPUz
3)Any other relevant OC information INCLUDING quantity
4)Winner will be the member with lowest latency overall
5) Please include details on your ram. I want to have a category on that so if people want it they can identify exactly what it is.
-
1638930301805.png
93.2 KB
Views: 59
See less
See more
5
Reply
Save
Like
Reactions:
1
61 — 80 of 95 Posts
I can’t believe I never really tried to get DDR4-3333 working on this chip. I didn’t even have to touch any voltages, it saw all 64GB and booted right up. Running Karhu right now and just passed 1000%, so it might be 24/7 stable as well. I’m getting ready to ditch this platform, but after all these years I’m still finding ways to squeeze more out of it.
See less
See more
Reply
Save
Like
Reactions:
1
This 4600c18 1.48v, 47.6ns is rather my 24/7-setup.
2x8GB G.Skill TridentZ Neo 3600Mhz 16 16 16 36 @ 4600Mhz 18 18 18 18 36 1T 1.48v
See less
See more
Reply
Save
Like
Nothing special but I always enjoyed experimenting on sodimm and this GE75 was a really good machine for that (This actually helped in closing the gap between this machine and machines that have a mux switch to bypass optimus)
Was able to hit 3600mhz with tighter timings but had to raise the SysAgent and Dram Voltage to beyond comfortable numbers and it was hard to keep everything under control even with LM and extra fans running directly on the memory, chipset and cpu.
Sam_Oslo said:
This 4600c18 1.48v, 47.6ns is rather my 24/7-setup.
2x8GB G.Skill TridentZ Neo 3600Mhz 16 16 16 36 @ 4600Mhz 18 18 18 18 36 1T 1.48v
View attachment 2537279
Click to expand…
Always loved how those APU’s FCLK can go insanely high, ignoring aida how much better is gaming performance compared to lets say 3800mhz c14?
See less
See more
Reply
Save
Like
Awsan said:
Was able to hit 3600mhz with tighter timings but had to raise the SysAgent and Dram Voltage to beyond comfortable numbers and it was hard to keep everything under control even with LM and extra fans running directly on the memory, chipset and cpu.
View attachment 2537518Click to expand…
what kit/voltage are you running?
I’ve got SODIMM in HTPC (this kit) but AsRock decided to limit DRAM voltage to 1.35v
See less
See more
Reply
Save
Like
The Pook said:
what kit/voltage are you running?
I’ve got SODIMM in HTPC (this kit) but AsRock decided to limit DRAM voltage to 1.
35v
View attachment 2537538
Click to expand…
Its been a while, dont have access to that machine right now but I think it was Samsung S-die and yea sodimm are usually maxed at 1.35v.
The one I posted was 1.2v and +0.125 on the uncore, but it was much simpler with intel I guess compared to AMD.
Reply
Save
Like
Awsan said:
Always loved how those APU’s FCLK can go insanely high, ignoring aida how much better is gaming performance compared to lets say 3800mhz c14?
Click to expand…
Yeah, These APU’s have a totally new IMC and can hit really high FCLKs. This 2300FCLK is actually the sweet spot for my IMC. It can run stable at 2366FCLK too.
I haven’t measured the increase of gameing-performance in a systematic way, but these APUs have a really great single thread performance too. It’s actually topping almost all single thread benching charts, as here in CPUz single thread benching , and here in OCCT single thread benching. This combination of a high FCLK/RAM-perormance and a high single thread performance creates a really fast and snappy user experience for most 24/7 usage scenarios, including gaming.
Reply
Save
Like
storm-chaser said:
Okay we will just continue with AIDA64 for now and in the future can try something else if people are still unhappy. so the comp goes on as is.
Click to expand…
What do you think about making a competition thread about a new (to this forum) benchmark ? I think y-cruncher looks interesting as it relays on both CPU and memory OC, and its a pretty hard stability/stresstest also
Even better its fully incorporated into the benchmate app which makes it even easier to keep it fair and square…
Buildzoid put up some sick Alder Lake numbers in y-cruncher @ DDR5-6732 CL32 here:
Just for fun i also tried a 2. 5b run with 4 active zen3 cores at auto ppt: (to get a feel of the benchmark)
What do you think, do you want to host a new and fresh competition like that ? 😋
See less
See more
Reply
Save
Like
Reactions:
3
domdtxdissar said:
What do you think about making a competition thread about a new (to this forum) benchmark ? I think y-cruncher looks interesting as it relays on both CPU and memory OC, and its a pretty hard stability/stresstest also
Even better its fully incorporated into the benchmate app which makes it even easier to keep it fair and square…
Buildzoid put up some sick Alder Lake numbers in y-cruncher @ DDR5-6732 CL32 here:
Just for fun i also tried a 2.5b run with 4 active zen3 cores at auto ppt: (to get a feel of the benchmark)
View attachment 2537713What do you think, do you want to host a new and fresh competition like that ? 😋
Click to expand.
..
That’s the mark of a true overclocker, always looking for ways to get more. nice work
domdtxdissar said:
What do you think about making a competition thread about a new (to this forum) benchmark ? I think y-cruncher looks interesting as it relays on both CPU and memory OC, and its a pretty hard stability/stresstest also
Even better its fully incorporated into the benchmate app which makes it even easier to keep it fair and square…
Buildzoid put up some sick Alder Lake numbers in y-cruncher @ DDR5-6732 CL32 here:
Just for fun i also tried a 2.5b run with 4 active zen3 cores at auto ppt: (to get a feel of the benchmark)
View attachment 2537713What do you think, do you want to host a new and fresh competition like that ? 😋
Click to expand…
Have you had sufficient time to familiarize yourself with the benchmark? Can you recommend how to best set up the benchmark itself so I can put that in the rules? Or just run all stock?
See less
See more
Reply
Save
Like
storm-chaser said:
Have you had sufficient time to familiarize yourself with the benchmark? Can you recommend how to best set up the benchmark itself so I can put that in the rules? Or just run all stock?
View attachment 2537979
Click to expand.
..
I haven’t played around with it other than my 4 core testrun, but i think it would be best to run with the «PI-2.5B» option instead of «PI-1B».
As i think the «PI-1B» will run too fast on todays high corecount cpus, maybe sub ~20 seconds on some of the hardware on this forum.. (?)
Saw bscool got a 74sec score @ «2.5B» option running on a 12900k with highly tuned memory.
Think that is a pretty good length for this benchmark.. ofcourse for those with only 4 cores it will take longer, but with a sub 20sec «1B» submission on a «high core count cpu» you would risk getting somewhat unstable systems completing the benchmark if we use the «1B» option.
In regards to «memory allocator» and «parallel framework» i have no idea whats best, maybe keep them at auto ?
Think the benchmate y-cruncher window’s together with memsettings (zentimings/mem tweak2 etc) need to be required in the screenshot. CPU-Z and hwinfo etc can be included as a bonus (like i did on my 4 core testrun below)
See less
See more
Reply
Save
Like
Reactions:
2
I would do the y-cruncher if storm-chaser puts it up.
Reply
Save
Like
I will not put 400 watts through my processor
I will not put 400 watts through my processor
I will not put 400 watts through my processor
Screw it, let’s do this thing!
Reply
Save
Like
Reactions:
1
JSHamlet234 said:
I will not put 400 watts through my processor
I will not put 400 watts through my processor
I will not put 400 watts through my processorScrew it, let’s do this thing!
Click to expand…
5 min
Reply
Save
Like
ez
See less
See more
Reply
Save
Like
Reactions:
1
any way to get negative latency with aida64?
The Pook said:
Click to expand…
If you guys want we can start another thread for maxmemm2, but doing latency only, since this aida test may be suspect depending on your configuration.
Reply
Save
Like
Hello,
i made it now, not a stable, but benchstable, lol, sorry
Merry christmas to everyone!
See less
See more
Reply
Save
Like
Reactions:
1
storm-chaser said:
If you guys want we can start another thread for maxmemm2, but doing latency only, since this aida test may be suspect depending on your configuration.
Click to expand…
I’m offended
are you saying my score isn’t legit 😟
Reply
Save
Like
The Pook said:
I’m offended
are you saying my score isn’t legit 😟
Click to expand…
On the contrary, I’m saying some platforms in general are affected by this bug. Your particular scores are bang on point.
Only trying to level the playing field so we can get accurate and trustworthy results for yall. And we should be able to do that with maxmemm.
And we can still keep this thread open as well
Reply
Save
Like
Sam_Oslo said:
These APU’s have a totally new IMC and can hit really high FCLKs. This 2300FCLK is actually the sweet spot for my IMC. It can run stable at 2366FCLK too.
Click to expand…
its not because its a new IMC, its the design..its not based on chiplets architecture but a monolithic die approach which would and can yield higher memory and IF clock OC (since they all sit under 1 die)..unlike the chiplets design for the non APU SKU’s..
Reply
Save
Like
61 — 80 of 95 Posts
- This is an older thread, you may not receive a response, and could
be reviving an old thread. Please consider creating a new thread.
Top
Aida64 system stability test — how to use it
Aida64 is a program with many functions to detect computer properties, test to show system health, stability, overclocking resistance and the like. The application is optimal for Aida 64 stability test systems in systems with low throughput.
Contents:
This procedure involves a load on each element. At the same time, possible component failures are investigated in order to take timely measures to prevent critical moments.
Preparing the system
If you have a small computer, then before testing it is necessary:
- Check for overheating at standard load. Temperatures in the range of 40 to 45 degrees are normal.
- If the temperature is higher, then the tests should be abandoned or studies should be carried out carefully.
Restrictions are due to the fact that during the procedure the processor may experience an increased load, as a result of which the operating temperature rises in some cases up to 90 degrees or more. Such frames are dangerous for the components of the device, both main and nearby.
Selecting elements
To start the System Stability Test in Aida64, you need to perform the following list of actions:
- Open the Service section in the top menu.
- In the drop-down box, find System Stability Test.
- In a separate open window, there are two charts, as well as several items to choose from. Let’s take a closer look at these sections:
- Stress CPU — by checking this item, we subject only the processor to additional load.
- Stress FPU – when checked, the load will be carried out on the cooler.
- Stress cache – cache testing.
- Stress system memory – RAM stress test.
- Stress local disk — check when the hard disk is under load.
- Stress GPU — stress load on the video card.
All items can be selected, but this will lead to disastrous consequences if the system is weak. In this case, an emergency overload of the personal computer will occur. Several parameters are displayed on the graphs at once, which significantly slows down the operation of the system as a whole.
Graphs in testing
Two graphs are located at the bottom and show the following parameters:
- The first one shows the current temperature of the processor. It is possible to monitor the average value for the entire device or for individual cores, as well as display both values on one graph.
- The second graph monitors the processor load. The CPU Throttling item should not be greater than zero during normal operation. Otherwise, the problem must be sought precisely in the operation of the processor.
A special menu is located above these graphs and allows you to open other graphic windows showing characteristics such as voltage and frequency. The Statistics section contains an overview of each component.
Start testing
In order to start the test, you need to mark the elements that you need.
- We mark the necessary components.
- Press the Start button. The process itself will take about half an hour.
- During the procedure, in the window opposite the element, you can see the detected problems and errors.
- The temperature curves must be constantly monitored. When changing to critical values, it is necessary to stop the tests.
- To end, press the Stop button. The results are saved with the Save button.
Note that the presence of 5 or more errors indicates that the operation is not stable and it is necessary to correct errors in the device. Each error has its own name in accordance with the belonging to the component.
3.7 11 votes
Article rating
Aida64 test — how to test a video card and processor
Diagnostic programs appeared in wide access for a reason. A user testing his computer can pursue different goals: from simple interest to the prevention of possible breakdowns. Some of the lovers of technological innovations, thanks to the tests, want to make sure that the choice of this or that equipment is correct.
Contents:
How to make testing safe
There are many different ways to diagnose a computer: from completely harmless measurements of temperature, data transfer speed to stress tests. The latter represent a sequential load on the object under test up to a critical indicator. When testing the processor, the frequency response and bus voltage do not increase. The temperature is very sensitive to the stress test and the cooling device must work perfectly to prevent fatal overheating of the equipment.
How to test the processor
Aida 64 is an excellent application for monitoring and testing computer components. Before proceeding, it is necessary to measure the current temperature of the processor. Ignoring this procedure can lead not only to an emergency reboot, but to have sad consequences in the form of burnt components.
If the temperature is 45°C or lower at rest, you can start testing. You can determine the temperature in section “ Sensors ” category “ Computer ”. The no-load diagnostics is in the section “ Tests ” (everything that starts on the CPU belongs to the category we need).
How to check the video card
Similar to checking the CPU temperature, you need to make sure that the graphics processor is cool enough. The “Computer” category, “Sensors” section will help us with this. Here we pay attention to the values opposite “GPU” or opposite the line indicating the name of your video card (if it is not built-in). The norm is still the same — 45 ° C. With an increased indicator of this value, the test should also be postponed and the cause of overheating should be eliminated. To diagnose, do the following:
- Go to “ Service ”;
- Select item “ GPGPU Test ”;
- Mark devices required for testing;
- Press “ Start Benchmark ”;
- After going through the entire list, the results can be saved.
How to do a stress test for a video card and a processor
After we are convinced of the temperature stability and see the values sufficient for a load test, we can start to conduct it. For this:
- Run Aida64;
- In the horizontal menu, select “ Service ”;
- Press the item “ System stability test ”;
- Mark the necessary objects for loading:
- Stress CPU – central processing unit;
- Stress FPU — cooler;
- Stress cache — cache;
- Stress system memory — random access memory;
- Stress local disk — hard disk;
- Stress GPU — graphics card.
- Press “ Start ”.
This process usually takes about half an hour. In case of unstable operation of the system with overheating, an appropriate notification will be issued. In critical cases, the system will reboot. After passing, it is possible to save the report.