Question Zen 4 builders thread

[Markfw]

I know its early, but as someone pointed out, I better get my ram now before the scalpers are all over it.

To go with a 7950x, what do you think of this ram:

https://www.newegg.com/g-skill-32gb-288-pin-ddr5-sdram/p/N82E16820374369?Item=N82E16820374369

(edited for my best recommendation, 12/5/2023)

 

[AdamK47]

What I think is happening is that the negative CO adjustment still follows the scale where there is less voltage for a given clock speed. With the increased voltage offset it still allows the same clocks on the curve. It's just that all of the adjustments now have the added voltage. As long as it's within the temperature and power envelopes, it will still have the same expected clocks for the negative CO. At least that's the jist of it after consulting Dr. Google.

 

[Markfw]

What I think is happening is that the negative CO adjustment still follows the scale where there is less voltage for a given clock speed. With the increased voltage offset it still allows the same clocks on the curve. It's just that all of the adjustments now have the added voltage. As long as it's within the temperature and power envelopes, it will still have the same expected clocks for the negative offset. At least that's the jist of it after consulting Dr. Google.

I have mine both set to -25 and the performance differentce is very negligible. I am happy at the 142 watt the same as the 5950x, and it still outperforms them by 40% or something, a BIG difference. I think it will take a lot of tweaking before we figure out the perfect settings, but -30 was mentioned by a web "expert" and I tried 25, which was the default, and I am thrilled. Keep playing, maybe you can find the perfect setting !

 

[AdamK47]

I have mine both set to -25 and the performance differentce is very negligible. I am happy at the 142 watt the same as the 5950x, and it still outperforms them by 40% or something, a BIG difference. I think it will take a lot of tweaking before we figure out the perfect settings, but -30 was mentioned by a web "expert" and I tried 25, which was the default, and I am thrilled. Keep playing, maybe you can find the perfect setting !

I get system reboots at -20 no matter what. This is with just the good cores. Looks like I have an unlucky chip.

 

[Markfw]

I get system reboots at -20 no matter what. This is with just the good cores. Looks like I have an unlucky chip.

we don't know what the motherboards do and the bios. I have 2 that act the same way. Don't think that yours is a bad chip, too many variables.

Here is a test. What do you bench on these at your best setup ?




So I only get 36120, what do you get on these 2 ?

 

[AdamK47]

It's not much, but it adds up with all of the other little tweaks that can be done.

Edit: Right is -5 CO with no voltage offset. Left is -12CO with +0.0250V offset. Both stable settings.

 

[AdamK47]

we don't know what the motherboards do and the bios. I have 2 that act the same way. Don't think that yours is a bad chip, too many variables.

Here is a test. What do you bench on these at your best setup ?

So I only get 36120, what do you get on these 2 ?

CPU-Z Non-AVX


CPU-Z 512-AVX


Cine R23

 

[Markfw]

So on both you are beating mine. So your setup is just fine !!!!

 

[AdamK47]

So on both you are beating mine. So your setup is just fine !!!!

You're right. Motherboard makers have their own secret recipe per board and BIOS version. The CO and PBO extend on to that.

 

[Markfw]

You're right. Motherboard makers have their own secret recipe per board and BIOS version. The CO and PBO extend on to that.

And , bottom line, you are beating me ! So whatever your bios is doing is working ! I just want the 142 watts. I don't care if I am the fastest. Some quick googling, and the 5950x is about 28k. I am doing 36k on the same or lower wattage ? (those results are overclocked).

I just tried mine. I do 16 k on this box(a 5950x on a 470 chipset), and 21k is the 5950x stock per the web. So 36k on the same wattage ? I am happy.

 

[Chaotic42]

Put in the latest BIOS for my motherboard, the ROG CROSSHAIR X670E HERO, enabled the AI tweaks, and got 755.9356 GFlops in linpack, up from 730.0836

 

[alexruiz]

Has any board maker released a UEFI with an AGESA newer than ComboAM5 1.0.0.3A?
Gigabyte released theirs 2 weeks ago, but the previous 3 versions were separated by only a few days.

 

[Det0x]

Has any board maker released a UEFI with an AGESA newer than ComboAM5 1.0.0.3A?
Gigabyte released theirs 2 weeks ago, but the previous 3 versions were separated by only a few days.

Asus are at 1.0.0.3C

Alittle better memory OC, alittle worse cpu core boosting..

 

[alexruiz]

Asus are at 1.0.0.3C

Alittle better memory OC, alittle worse cpu core boosting..

Thank you, that spreadsheet is incredibly useful!

 

[Det0x]

Asus have released a new testbios with changeable tREFI

It can now be change from stock value which is 12076, all the way up to maximum 65535...
This is the numbers i can achieve with this change, together with true T1:

Latency optimized:


Bandwidth optimized:


Maximum stability for my system:

• 25 cycles testmem5 1usmus cfg
• ~1 hour Y-cruncher all tests enabled

 

[Chaotic42]

Assuming all of my sensors, including the LED on the motherboard are telling the truth, I seem to be limited to about 40C, even when pushing 230W. I wonder what it could do at this mystical 95C I keep hearing about

 

[mv2devnull]

Assuming all of my sensors, ...

Which tool creates that output?

 

[Chaotic42]

Which tool creates that output?

The chart thing is s-tui: https://github.com/amanusk/s-tui
The text output of wattage is powerstat: https://github.com/ColinIanKing/powerstat
I also use nvtop to monitor my GPU

The whole thing is running in MobaXTerm since the system is down in the basement

 

[igor_kavinski]

Bandwidth optimized:
View attachment 69834

Just a little more tweaking needed to break the 100,000 MB/s barrier for Write speed

 

[AdamK47]

It's not much, but it adds up with all of the other little tweaks that can be done.

Edit: Right is -5 CO with no voltage offset. Left is -12CO with +0.0250V offset. Both stable settings.

View attachment 69816

And the march of progress continues. Tried the new 0803 BIOS for the Crosshair Extreme. I cannot enable PBO in any form without reboots, so single thread boost speeds are lower. With CO I can increase it much higher than before. -20 now with the new BIOS vs -12 on 0801 BIOS. Memory performance also improved. Tripled tREFI to 35000 and set ADDR_CMD_MODE to "Buf".

CPU score in Timespy went up.

 

[Det0x]

Very interesting article:

TLDR: Enabled "MEDIUM LOAD BOOSTIT" is the way forward for those with highend asus motherboard

https://skatterbencher.com/2022/10/26/update-on-ryzen-7000-c-state-boost-limit/

 

[ericgandt]

If you are not using the iGPU, disable it at the UEFI
There is no reason to install AMD graphics drivers, so uninstall it.
Make sure you are installing chipset drivers. A lot of people overlook them, but they are extremely important. Grab them from AMD's website

Right sadly on my Asus MB I can not find a BIOS value to disable the GPU yet.

 

[eek2121]

I wanted to add some benchmark results from the the 7950X set to 105WTDP/142W PPT using AMD ECO mode. All results were done in the FormD T1 v2 Case, Phanteks Glacier One 240MP AIO with Noctua slim NF-A12x15 fans (Full size fans do not fit in the case), 64gb of DDR5-6000, the latest build of Windows 11 (which may have some performance issues as outlined elsewhere, so YMMV), anEVGA RTX 3090 FTW3 Ultra. The only BIOS changes that were made from defaults is the CPU was set to ECO 105 mode, and the UCLK was set to FCLK. The machine tends to run a bit warm due to no case fans outside the two fans mounted on the radiator. However it wasn't bad.

The numbers below represent what you would see if you went with an extremely small ITX case and set your CPU to a 105W TDP via AMD's ECO mode.

Geekbench 5:

ASUS System Product Name - Geekbench Browser

Benchmark results for an ASUS System Product Name with an AMD Ryzen 9 7950X processor.

browser.geekbench.com

Single Core: 2153
Multicore: 23035

CPU-Z:

Version 19.01.64 AVX-512 (beta):

Single Core: 992.9
Multi Core: 18598.8

7Zip x64 19.6102:

(Note that these may not be the official binaries. These binaries were provided by nanazip, a front end redesign effort, I may test the official distro later if there is reason to believe the results are materially different)

Command: 7z.exe b 6 -md23 -mx9 -mm=LZMA
1T CPU Freq (MHz): 5398 5389 5390 5383 5383 5389 5396
16T CPU Freq (MHz): 1509% 5290 1475% 5268

Compression MIPS: 181,929
Decompression MIPS: 254,226
Combined: 218,077 MIPS

Cinebench r23:

Single Core: 1,913
Multicore, 5 runs with a 30 seconds between each:

33,271
33,932
34,077
34,122
33,953

(Note: This result is curious, I wonder if cache was retaining data between runs?)

10 Minute Multicore Stress: 33,487

Rapydmark High: 158.628 s

Handbrake 1.5.1
LG 4K Demo:
Base preset: H.265 MKV 2160p60
encoded 1806 frames in 111.16s (16.25 fps), 11820.19 kb/s, Avg QP:29.09

Blender 3.3.1:

Barbershop - 10:46.77
Classroom - 02:25:58:
According to HWInfo64 (only open during a separate run, peak/average were the same for the duration of the run)
Average effective clocks: 4.47 GHz
Peak/Average power consumption: 134W
Peak/Average tdie temps: 93.9C

(Steam open for these next few, so some overhead here)

3DMark CPU Profile:

Max Threads: 13540
16 Threads: 13,648
8 Threads: 7794
4 Threads: 4,140
2 Threads: 2,113
1 Thread: 1,065

Civ6 AI Test - Turn Time: 7.17
Civ6 Expansion AI Test - Turn Time 22.6

I may post more benchmarks later, but next I want to test throttling temps to 85C instead. I hope to cap it off by testing with stock TDP settings.

EDIT: Added 7zip and CPU-Z AVX-512

 

[moinmoin]

next I want to test throttling temps to 85C

Would you be willing to test much lower temps limits?

Personally I'd be most interested if there is a way to limit the max frequency a core reaches without setting a static all core frequency (goal being always running cores at their most efficient frequency), so I'm imagining something like limiting the temps to something low like 45°C or even lower may have the desired effect. But so far I'm not aware of anybody doing tests like that so even this 45°C point is a shot in the dark. Thanks in advance!

 

[eek2121]

Would you be willing to test much lower temps limits?

Personally I'd be most interested if there is a way to limit the max frequency a core reaches without setting a static all core frequency (goal being always running cores at their most efficient frequency), so I'm imagining something like limiting the temps to something low like 45°C or even lower may have the desired effect. But so far I'm not aware of anybody doing tests like that so even this 45°C point is a shot in the dark. Thanks in advance!

I believe you can limit the max frequency. I may also test that, however I want to test a target temp first. I also need to find my Kill-A-Watt to measure system power. I do wish I had access to SPEC 2017.

 

[eek2121]

You're right. Motherboard makers have their own secret recipe per board and BIOS version. The CO and PBO extend on to that.

Actually, most of the code is written by AMD themselves. From my understanding, board makers usually slap a UI on top and map to internal functions provided by AMD. That was why AMD eventually exposed AMD PBS/CBS/Overclocking, because board makers did not do a good job with the settings there. (note there is a bit more to it then that, but most of the code comes from AMD and vendors upstream from AMD)