What controls Turbo Core in Xeons?

Page 44 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

sciff

Member
Mar 6, 2017
136
52
71
Wrong argument, since I know it's possible to run without microcode patch/update. On my single-CPU board I have no issues with it at all, even though rendering in Corona rises the CPU's TDP up to 240 W and even a little higher.

I was asking if your mobo can provide the same power so as to run Corona at 3.4-3.5 GHz, not 2.8-2.9. But apparently you don't know that. I was wondering if you could test it for me. Believe me, I'm not the only one who is interested in the answer...
 

kjboughton

Senior member
Dec 19, 2007
330
118
116
Oh, there will be testing. These babies are going under water. I should have something to report this weekend or early next week.

If you want, start to prepare what you need to have tested and I will begin preps.
 

txgy

Member
Apr 24, 2017
29
1
41
I am right in the middle of my v3 build and was researching OC methods when I can across this thread.

Current WS is a dual E5-2697 v2 that I've pushed to a BCLK of 112 for a top turbo of 3.36Ghz for all cores.
I've managed to POST and boot into Windows at 114 BCLK; 115MHz or more causes a BIOS corruption or POST loop failure that requires me to swap out a whole new BIOS chip and so I've become quite adapt at modding BIOS' and the such...

That v2 build is under a custom watercooling loop. The v3 build in my sig (in process) will be almost exactly the same except for Haswell over IVY-B.
I'm curious about the stability issues when using such high BCLK.
RAM,video card, HDD, SSD, Sound Card, NIC, do they work good ???
 

kjboughton

Senior member
Dec 19, 2007
330
118
116
Wrong argument, since I know it's possible to run without microcode patch/update. On my single-CPU board I have no issues with it at all, even though rendering in Corona rises the CPU's TDP up to 240 W and even a little higher.

I was asking if your mobo can provide the same power so as to run Corona at 3.4-3.5 GHz, not 2.8-2.9. But apparently you don't know that. I was wondering if you could test it for me. Believe me, I'm not the only one who is interested in the answer...

how about this?
https://corona-renderer.com/blog/corona-1-3-benchmark/

My dual IVY-B system took top E5-2697v2 (x2) with 0:59 (freecableguy). Actual all-core frequency was 3.33GHz and not the 2.97GHz reported.
https://corona-renderer.com/benchmark/?all

-FCG
 
Last edited:

kjboughton

Senior member
Dec 19, 2007
330
118
116
Would be helpful if someone could add definitions and/or explanations to the following EFI driver option files:

v3.efi :

v3x2_50_vcc.efi :
v3x2_50_39_vcc.efi :

v3x2_cp.efi :
v3x2_cp27.efi :
v3x2_cp39.efi :

v3x2_cup.efi :
v3x2_cup27.efi :
v3x2_cup39.efi :
 

Zladimir

Member
Apr 14, 2011
34
3
71
On my single-CPU board I have no issues with it at all, even though rendering in Corona rises the CPU's TDP up to 240 W and even a little higher.

Im curious how you managed such high TDP ? My CPU is throttling on its pre-defined TDP level 0 and 1. Tries to overwrite it, even with efi, didnt succeed.
How did you proceed to manage it ?
 

sciff

Member
Mar 6, 2017
136
52
71
Zladimir, on my motherboard combined with my CPU, if you don't load µCU (microcode update) one way or another (from the efi file itself or from Windows/VMWare microcode updater), then AVX/AVX2 load doesn't cause turbo boost frequencies to drop. My CPU works at the same 3.4 GHz in Corona Benchmark instead of 2.9 GHz, hence this crazy TDP (also it heats up quite well)
 

Zladimir

Member
Apr 14, 2011
34
3
71
Zladimir, on my motherboard combined with my CPU, if you don't load µCU (microcode update) one way or another (from the efi file itself or from Windows/VMWare microcode updater), then AVX/AVX2 load doesn't cause turbo boost frequencies to drop. My CPU works at the same 3.4 GHz in Corona Benchmark instead of 2.9 GHz, hence this crazy TDP (also it heats up quite well)

No, it depends on the efi file. The recent files somehow fake the power readings and by this the card isnt throttling in TDP, which it should. Its true that AVX2 unload (AVX1 doesnt reduce frequency by design) isnt active in pre8 mc, but it gets still throttled by higher TDP. And there is no known way to deactivate or modify Intel's predefined TDP and current throttling. All change which is possible (cTDP up, TDP1+2 temporary limits) is also settled by Intel.
However to fake the real readings is an very interesting thing and I ve no clue what msr got written to do so. All was done by @randir, but sadly he hasnt been active since 11 weeks.
All what is known is a little bit information from him like

Use at your own risk! Ensure that your CPU has adequate cooling and monitor temperature if you decide to stress-test it! This breaks all in-system power consumption monitors - do not rely on them!

If you use any of the older efis and make a cold reset, you will notice your CPU will throttle by TDP, too.
I just prefer to use my own voltage offsets. Thats why I'd like to know how to get my power readings broken, too. ;)


However, did anyone experienced with 0x61E ? As by design this MSR is writable. System hangs when I try to write. Not sure why. Surely Im missing here something.
 

sciff

Member
Mar 6, 2017
136
52
71
No, it depends on the efi file. The recent files somehow fake the power readings and by this the card isnt throttling in TDP, which it should. Its true that AVX2 unload (AVX1 doesnt reduce frequency by design) isnt active in pre8 mc, but it gets still throttled by higher TDP. And there is no known way to deactivate or modify Intel's predefined TDP and current throttling. All change which is possible (cTDP up, TDP1+2 temporary limits) is also settled by Intel.
However to fake the real readings is an very interesting thing and I ve no clue what msr got written to do so. All was done by @randir, but sadly he hasnt been active since 11 weeks.
Frankly, I don't observe these fake power readings. I have made measurements on my system using HWiNFO monitoring and a socket power meter. I run Corona Benchmark with and without µCU, also with and without the use of EFI exploit. You can see the results in the following table:

0_15d376_6a7061ab_orig.png

Also you can download the screen/power meter photos from this link: https://yadi.sk/d/Ck9hs_tM3LXtKS. A couple of examples:






Do you still doubt that not using a microcode update allows 240 W power consumption on E5-2696 V3?
 

kjboughton

Senior member
Dec 19, 2007
330
118
116
I don't believe that telemetry reading. Can you see what is reported with AIDA64 or CoreTemp 1.9? That's a board sample point (Asus EC: All In...) and most definitely not coming from the CPU PCU.

I reiterate... the highest power I have seen with my 2969v3 pair is approximately 145W per package... funny that. That's during a Corona 1.3 bench run which we know makes heavy use of AVX2. Can you report back what you find with a tool other than HWInf64?
 

kjboughton

Senior member
Dec 19, 2007
330
118
116
In fact, I believe the additional "MSR hackery" done as part of v3x2_50_vcc.efi breaks the PCU power monitoring. HWInfo64 shows cores in C6 states and what is reported by PCU does not change regardless of load or idle. This may be part of what is done to fool the system to pushing to higher a higher all core multi.
 

sciff

Member
Mar 6, 2017
136
52
71
I don't believe that telemetry reading. Can you see what is reported with AIDA64 or CoreTemp 1.9? That's a board sample point (Asus EC: All In...) and most definitely not coming from the CPU PCU.

I reiterate... the highest power I have seen with my 2969v3 pair is approximately 145W per package... funny that. That's during a Corona 1.3 bench run which we know makes heavy use of AVX2. Can you report back what you find with a tool other than HWInf64?

I don't like other tools like AIDA64 because they show inaccurate results. I only can show you old pictures of ASUS OC utility back when I still used v3x2_cup.efi.


Why don't socket power meter readings satisfy you? Take a look at my photos in the archive more carefully. Of couse, those (socket power readings) are for the whole PC. But they allow you to see that under load the unlocked CPU indeed has a massive TDP.

145 W is specification TDP. If your CPUs don't breach this limit, then they are not really unlocked.

What's the all-core frequency when you run Corona Benchmark?
 

kjboughton

Senior member
Dec 19, 2007
330
118
116
Holding all-core turbo max of 3.4GHz regardless of workload type (AVX/AVX2, non-AVX).
This is using v3x2_50_vcc.efi and no ucode. CPUs are watercooled with high-end custom loop.
The funny thing is, I have to do a warm-reboot after the initial cold boot to move up from all-core max of 3.2GHz to 3.4GHz. There are some funny interactions going on that are well worth exploring...

Cine R15 is 5K; Corona bench is down to 35 secs.
 

sciff

Member
Mar 6, 2017
136
52
71
Holding all-core turbo max of 3.4GHz regardless of workload type (AVX/AVX2, non-AVX).

That's just great! Could you provide some screenshots?

Have you figured out a reliable way of determining the real TDP of each CPU?

This is using v3x2_50_vcc.efi and no ucode. CPUs are watercooled with high-end custom loop.
The funny thing is, I have to do a warm-reboot after the initial cold boot to move up from all-core max of 3.2GHz to 3.4GHz. There are some funny interactions going on that are well worth exploring...

Cine R15 is 5K; Corona bench is down to 35 secs.

Cool! It's a pity you don't show these results.