Discussion Rudi_Float_Bench v0.02a

[Hail The Brain Slug]

866 on avx256 binary, 3290 with avx512 binary on 9800X3D

 

[Markfw]

8347 !

Now let me try 128 threads !

 

[Markfw]

8321 with 128 threads, and only 1/2 the CPUs were used.

 

[igor_kavinski]

8321 with 128 threads, and only 1/2 the CPUs were used.

Still trying to figure out how to work on the other half.

 

[igor_kavinski]

@Markfw , if possible, can you run v0.02 AVX-512 on a Genoa CPU similar (in core counts) to your Turin? Need to see if the benchmark exposes the half width and full width difference between the AVX-512 implementations of Genoa and Turin.

 

[igor_kavinski]

An interesting find: Core Memory Isolation turned on reduces the score.

Tested on Core i7-10750H gaming laptop.

With the nasty option turned on: 594
With it turned off: 725

I'm guessing older CPUs take a greater hit.

There was also a difference in frequency behavior.

With the isolation crap, the speed would start at around 3.5 GHz all core.
With it off, it started at 4 GHz all core but around the 30 second mark, the fans kick in and speed reduces to 3.5 GHz.

Any guesses why Core Memory Isolation impacts the all core speed? And why this benchmark, not being very memory intensive, sees a hit?

 

[Makaveli]

 

[igor_kavinski]

View attachment 120240

This with or without Core Memory Isolation?

 

[Makaveli]

This with or without Core Memory Isolation?

Its off.

 

[igor_kavinski]

9950X3D Stock (no PBO anything), DDR5-7200C34 FCLK 2133 UCLK 1800 Windows Ultimate Power Plan

 

[igor_kavinski]

Chinese KX-7000 with a score of 2: http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=threads/zhaoxins-zx-f-kx-7000-kh-40000-and-beyond.2564975/post-41433827

Well, at least we got a score. I wasn't even sure it would finish on it!

 

[aigomorla]

Stock settings, because i can't seem to overclock my threadripper without borking something, and need more practice.

 

[gregory_x86]

AMD FX-8120 Stock with turbo 3,1-4,0 GHz


AMD FX-8120 @3,0 GHz


AMD FX-8300 Stock with turbo 3,3-4,2 GHz


AMD FX-8300 @3,0 GHz



AMD Ryzen 7 1700 Stock with PBO 3,0-3,75 GHz



AMD Ryzen 7 1700 @3,0 GHz


AMD Ryzen 7 1700 no SMT @3,0 GHz




AMD Athlon 5370 Stock 2,2 GHz (Jaguar cores)




Program doesnt work on AMD Phenom due the lack AVX instruction.
The RAM latency tool works.

 

[igor_kavinski]

Stock settings, because i can't seem to overclock my threadripper without borking something, and need more practice.

View attachment 122278

Seems the benchmark is calculating AVX-512 capability properly. If we divide Markfw's score by 2, we get half AVX-512 capability of about 4000 and 48 threads score should be 3000 something. But then Zen 5 has more IPC so that's why your score is lower than 3000.

By the way, doesn't Ryzen Master work for you? I tried it and really liked the experience. You could at least try some PBO and light RAM overclocking with it.

 

[igor_kavinski]

AMD FX-8120 Stock with turbo 3,1-4,0 GHz
View attachment 122269

AMD FX-8120 @3,0 GHz
View attachment 122270

AMD FX-8300 Stock with turbo 3,3-4,2 GHz
View attachment 122271

AMD FX-8300 @3,0 GHz
View attachment 122272


AMD Ryzen 7 1700 Stock with PBO 3,0-3,75 GHz
View attachment 122273


AMD Ryzen 7 1700 @3,0 GHz
View attachment 122274

AMD Ryzen 7 1700 no SMT @3,0 GHz

View attachment 122275


AMD Athlon 5370 Stock 2,2 GHz (Jaguar cores)

View attachment 122276


Program doesnt work on AMD Phenom due the lack AVX instruction.
The RAM latency tool works.

View attachment 122279

Thanks for all these scores!

 

[aigomorla]

Seems the benchmark is calculating AVX-512 capability properly. If we divide Markfw's score by 2, we get half AVX-512 capability of about 4000 and 48 threads score should be 3000 something. But then Zen 5 has more IPC so that's why your score is lower than 3000.

By the way, doesn't Ryzen Master work for you? I tried it and really liked the experience. You could at least try some PBO and light RAM overclocking with it.

nope... i tried.
I blame ASUS for making the SAGE series board so complicated, that it would give even gigabyte owners a headache.
I went though so many guides... mostly they are done on the gigabyte platform, and well, ASUS has nothing they are talking about except a select few.

And there is no data sheet on what "Stock Voltage" is for any of the things i can tune, that i don't know whats safe or not except my CPU temps.

I would love a cheat sheet on what manual "Stock" voltages are for this cpu across every category.
I dont think even AMD wants us to know.

Whats worse is with my water PBO likes / wants to boost my clock to 5.6ghz single core sometimes at the cost of 1.51V VID. Im like Oh no you dont, and had to hard set voltages manually.

 

[gregory_x86]

AMD Ryzen 9 5950X Stock with PBO @3,4-5,05 GHz

 

[igor_kavinski]

AMD's model numbers in the past were bonkers! I never could have guessed that the FX-8120 is a higher performing, higher wattage part with a different architecture codename!

 

[igor_kavinski]

AMD Ryzen 9 5950X Stock with PBO @3,4-5,05 GHz

View attachment 122286

This is so funny! Your 5950X is better at AVX2 than my thermally throttling 9950X3D!

 

[gregory_x86]

AMD's model numbers in the past were bonkers! I never could have guessed that the FX-8120 is a higher performing, higher wattage part with a different architecture codename!

Something is wrong with my Piledriver AMD cpu. Please look at FPU flops in Aida. That is worse than Bulldozer. Both are set on 3,0 GHz.


But in Cinebench R23 better is Piledriver.

 

[igor_kavinski]

Whats worse is with my water PBO likes / wants to boost my clock to 5.6ghz single core sometimes at the cost of 1.51V VID. Im like Oh no you dont, and had to hard set voltages manually.

After bolting down the voltages, any change in boost behavior? Still hitting 5.6 GHz?

 

[igor_kavinski]

Something is wrong with my Piledriver AMD cpu. Please look at FPU flops in Aida. That is worse than Bulldozer. Both are set on 3,0 GHz.

It's the same thing I noticed in rudi_bench too. Maybe CB R23 uses different instructions that favor FX-8300. But AIDA FPU and rudi_bench favor FX-8120.

 

[gregory_x86]

It's the same thing I noticed in rudi_bench too. Maybe CB R23 uses different instructions that favor FX-8300. But AIDA FPU and rudi_bench favor FX-8120.

Theoretical FP32 for AMD Bulldozer FX-8120 3,0GHz * 4 FPU cores * 8/cycle *2(FMA4)=192 gflops
For Piledriver FX-8300 it should be excacly the same result, I guess.

 

[igor_kavinski]

For Piledriver FX-8300 it should be excacly the same result, I guess.

Maybe FX-8300 has hotspot issues? Try using a better thermal paste and bigger heatsink on it.

 

[gregory_x86]

Maybe FX-8300 has hotspot issues? Try using a better thermal paste and bigger heatsink on it.

I don't think so. Cooler is good. I mount and unmont it many times. I tried both cpus also with set 2,0 GHz and always 8120 has better score in FP32, but was slower in CPU-Z, Cinebench etc. I need another one to compare.