2019/07/20 12:44:09
jasoncodispoti
I have been running BIOS v1.18 on the X299 Dark for a couple of weeks now without issue. However today I was messing around with HandBrake doing some video encoding and discovered something a bit odd with the CPU frequency that only seems to be impacting BIOS v1.18.
 
PC Build Information: 
Mother Board, EVGA X299 Dark
CPU, Intel i7-9800x
 
Issue: 
What I found was that when running HandBrake that the CPU ratio and subsequently the CPU frequency were not remaining static  even though they were set to x43 (4.3GHz) with no AVX offset configured in the BIOS. A whole bunch of testing later I determined that if you set a CPU ratio of say x43 with no AVX offset and leave all of the voltage settings in the BIOS on AUTO, everything operates as normal with no issues. However if you go and change VCORE voltage to override and set it to say 1.25v and you start using HandBrake you will find that the CPU ratio and subsequently the CPU frequency no longer remain set to x43 instead it is changing (frequently) between x41 and x43. I have confirmed and retested multiple times that this is only an issue with BIOS v1.18 and is not appear to be an issue with BIOS v1.17. 
 
Really I am looking for an EVGA Tech to help work through this with me, to figure out the issue. Let me know any additional information that you need... 
 
Thanks in advance! 
 
Update 7/21 #1:
I have been doing some additional testing and believe that this is occurring due to some type of imposed power limit with AVX-512 instructions. When running HandBrake with only AVX2 instructions there is no drop in CPU frequency. However if I use HandBrake with AVX-512 instructions any core that is executing said instructions will start to down-clock below the set CPU ratio/frequency in the BIOS. In addition it appears that if a static VCORE voltage higher than 1.90v is set than the cores will start to throttle down. For example setting a CPU ratio of x43 @ 1.190v results in no CPU throttling however setting a CPU ratio of x43 @ 1.195v results in throttling. I am not sure what exactly is throttling it, however from my research online this is normal behavior for properly implemented AVX-512? Which would imply that BIOS v1.17 was not functioning correctly when it came to AVX-512? I am not really sure at this point what the issue is, but it does appear that at least with my CPU/Mother Board combination and this new BIOS that some type of LIMIT has been added that is impacting AVX-512. 
 
Update 7/21 #2: 
About 3 weeks ago I upgraded the BIOS on my EVGA X299 Dark to BIOS v1.18. I have been running it without issue since the upgrade however recently I was using my PC to do some video encoding and observed some odd behavior with the CPU ratio and subsequently the CPU frequency.​
 
Issue Description: 
I noticed while using HandBrake to encode a video that even though I had the CPU ratio set to x43 (4.3GHz) in the BIOS with no AVX2 or AVX-512 offset configured. That the CPU was not running at 4.3GHz instead it was operating around 4.0GHz.
 
PC Build:
Motherboard, EVGA X299 Dark
CPU, Intel i7-9800x
Memory, Corsair Dominator Platinum 32GB DDR4 3200MHz
GPU, EVGA RTX 2080ti FTW3 Ultra Gaming Hydro Cooper
Power Supply, SuperNOVA 1000 P2
 
Initial Trouble Shooting:
 Initially I thought the issue was related to thermals however HWInfo64 was reporting that all temperatures were normal; CPU=58C and VRM=52C.
Since the most recent change to the PC was the update to the BIOS that I had preformed a few weeks back. I decided to switch to one of the alternative BIOS chips on the motherboard and see if the same issue was present with BIOS v1.17, it was not it appears that this issue is isolated to BIOS v1.18.
 
Test #1 Prime95:
BIOS Settings, v1.18 all settings are set to default except for the following; CPU Multiplier Control = Manual Ratio Limit x43, AVX2 Negative Offset = 0, AVX3 Negative Offset = 0, VCORE = Override @ 1.250v. TJMax value was manually set to 95C since the motherboard does not auto detect the correct value (bug that I have previously reported with BIOS v1.17).
Through a great deal of testing I have found that this issue only presents when certain loads are applied to the CPU. It has to be an AVX-512 load and it had to either be a load applied by HandBrake when encoding h.264 video or you can use Prime95 v29.8 Torture Test Large FFT.
For this test I used Prime95.
While running the test the CPU ratio dropped to x40 which resulted in a final CPU frequency of 4.0GHz. CPU temperature stayed around 60C and VRM temp stayed around 54C. Using a Kill-A-Watt meter I was able to determine that the PC as a whole was pulling about 338w at the wall. Idle power consumption was around 189w at the wall which would put CPU power consumption at an estimated 149w at the wall.
 
Test #2 Prime95:
I repeated the same test from “Test #1 Prime95” listed above with the same settings\parameters however for this test I lowered VCORE voltage to 1.200v.
While running the test the CPU ratio dropped to x42 which resulted in a final CPU frequency of 4.2GHz. CPU temperature stayed around 56C and VRM temp stayed around 50C. Using a Kill-A-Watt meter I was able to determine that the PC as a whole was pulling about 323w at the wall. Idle power consumption was around 189w at the wall which would put CPU power consumption at an estimated 134w at the wall.
Take note that in this test compared to “Test #1 Prime95” that there was a slight droppage in estimated CPU power consumption. Also make note that the CPU ratio increased to x42 instead of the previous x40. I believe that the slight drop in wattage is what allowed the CPU to clock higher in this test.
 
Test #3 Prime95: 
I repeated the same test from “Test #2 Prime95” listed above with the same settings\parameters however for this test I lowered VCORE voltage to 1.180v.
While running the test the CPU ratio remained at x43 which resulted in a final CPU frequency of 4.3GHz. This is the frequency that the CPU should be operating at per the settings configured in the BIOS. CPU temperature stayed around 63C and VRM temp stayed around 50C. Using a Kill-A-Watt meter I was able to determine that the PC as a whole was pulling about 318w at the wall. Idle power consumption was around 189w at the wall which would put CPU power consumption at an estimated 129w at the wall.
Take note that in this test power consumption was once again lower than the previous test and as a result the CPU ratio remained at the configured setting in the BIOS of x43 resulting in a final CPU frequency of 4.3GHz.

Note:
While not specifically mentioned above CPU C States were also disabled in the BIOS and the High Performance power plan was selected within Windows, with all power savings features disabled. 
 
It appears that there is some type of power throttling that is taking place when running AVX-512 instructions with BIOS v1.18. I have confirmed multiple times by repeating the above tests with BIOS v1.17 that this DID NOT occur with BIOS v1.17. I am able to reproduce very similar results when using Handbrake while encoding H.264 video files.
 
It is also worth noting that at one point while testing and using the Intel Extreme Tuning application it did indicate that there was “Current Limit Throttling” taking place on the CPU. I am not sure how reliable that error is since the reporting of it is not always and HWINfo64 did not indicate that there was an issue. However from my testing that I have done thus far it would appear that this is the issue… some sort of power limit that has been put in place in BIOS v1.18.
I was provided a BETA BIOS to resolve an issue that I found in BIOS v1.17 and it appears to function as normal.
Please let me know if there is anything that you need from me or if you need me to do additional testing.
 
Thanks!


​​
2019/07/20 13:14:27
bcavnaugh
Do you see the same under Bios 1.17 or the BETA BIOS EVGA Gave you?
2019/07/20 16:30:46
jollydet
Are you hitting TJmax? No issues on 1.18 with 7960x @ 4.8 no avx offset.
2019/07/20 19:08:28
jasoncodispoti
bcavnaugh
Do you see the same under Bios 1.17 or the BETA BIOS EVGA Gave you?




BIOS v1.17 does not have the issue, I didnt try the BETA BIOS. 
2019/07/20 19:10:45
jasoncodispoti
jollydet
Are you hitting TJmax? No issues on 1.18 with 7960x @ 4.8 no avx offset.




No CPU temps are like 60c when this occurs, however it sure does seem like it some kind of throttling that is taking place. Not all work loads seem to cause it, Prime95 for example does not cause the issue, but video encoding with HandBrake does. 
2019/07/20 19:24:25
bcavnaugh
jasoncodispoti
No CPU temps are like 60c when this occurs, however it sure does seem like it some kind of throttling that is taking place. Not all work loads seem to cause it, Prime95 for example does not cause the issue, but video encoding with HandBrake does. 

AVX-512 with a 0 Settings can cause this as well as AVX2 can cause this with a 0 Setting in the Motherboard Bios.
Set AVX2=2 and AVX3=3 and retest with your video encoding with HandBrake.
2019/07/21 06:21:32
jasoncodispoti
Update: 
 
I have been doing some additional testing and believe that this is occurring due to some type of imposed power limit with AVX-512 instructions. When running HandBrake with only AVX2 instructions there is no drop in CPU frequency. However if I use HandBrake with AVX-512 instructions any core that is executing said instructions will start to down-clock below the set CPU ratio/frequency in the BIOS. In addition it appears that if a static VCORE voltage higher than 1.90v is set than the cores will start to throttle down. For example setting a CPU ratio of x43 @ 1.190v results in no CPU throttling however setting a CPU ratio of x43 @ 1.195v results in throttling. I am not sure what exactly is throttling it, however from my research online this is normal behavior for properly implemented AVX-512? Which would imply that BIOS v1.17 was not functioning correctly when it came to AVX-512? I am not really sure at this point what the issue is, but it does appear that at least with my CPU/Mother Board combination and this new BIOS that some type of LIMIT has been added that is impacting AVX-512. 
2019/07/21 11:27:18
bcavnaugh
So what did you set AVX2 and AVX3 in the Bios to?
2019/07/21 13:49:33
jasoncodispoti
bcavnaugh
So what did you set AVX2 and AVX3 in the Bios to?


Has no impact, throttling still occurs... tried using AVX2 and AVX3 offset of 3.

Appears that there is some kind of power limit being hit, Intel Extreme Tuning software will occasionally indicate this as being the case as well.

Also I am now able to reproduce the issue in Prime95.

Something changed in the BIOS version...
2019/07/21 14:24:45
bcavnaugh
jasoncodispoti
bcavnaugh
So what did you set AVX2 and AVX3 in the Bios to?


Has no impact, throttling still occurs... tried using AVX2 and AVX3 offset of 3.

Appears that there is some kind of power limit being hit, Intel Extreme Tuning software will occasionally indicate this as being the case as well.

Also I am now able to reproduce the issue in Prime95.

Something changed in the BIOS version...

Prime95 is not using AVX-512 unless you tell it to.
Prime95 29.5 now supports AVX-512
https://linustechtips.com/main/topic/985047-prime95-295-now-supports-avx-512-windows-ver-available/
 
Prime95 version 29.6/29.7/29.8
Windows 64-bit: ftp://mersenne.org/gimps/p95v298b5.win64.zip
What's New
1)  Support added for AVX-512 FFTs.
2)  FMA3 FFTs now have slightly higher FFT crossover points.  Soft crossovers are
    no longer used by default.  See undoc.txt.
3)  Torture test dialog box options now based on cache sizes.  Options for performing
    a weaker torture test are available.  Torture tests that use all RAM are now more
    stressful.  In-place vs. not in-place memory accesses now displayed on screen.
    On machines with more than 4GB of memory, blend defaults to 1/16th of RAM.
4)  Add & subtract operations for AVX-512 FFTs are now multithreaded.  This should
    improve performance for P-1 and ECM when using multiple threads.
5)  Benchmark results are now written to results.bench.txt.
6)  JSON results are now available for all work performed.  JSON results are
    written to results.json.txt.
7)  PRP tests with Gerbicz error checking are more immune to hardware errors. 
 
I run Prime95 but not for testing but for finding Prime Numbers.

Use My Existing Forum Account

Use My Social Media Account