RapydMark CPU benchmark

[igor_kavinski]

RapydMark for Windows (frikiscape.com)

Synthetic highly parallel benchmark.




My company's dual socket Ivy Bridge-E beat Comet Lake in total time, only due to more cores. Should be fun to see how 5950X and i9-12900K fare in this benchmark. My guess is that 5950X will win.

Update: 5950X did win! But i9-12900K put up a good fight despite fewer cores.

Performance comparison chart so far thanks to Makaveli:



Public view-only link for the sheet: RapydMark64 - Google Drive

CCD comparison by Det0x:

1 CCD -> 8 cores / 8 threads = 368.309 seconds
1 CCD -> 8 cores / 16 threads = 230.797 seconds
2 CCD's -> 16 cores / 32 threads = 135.978 seconds

Interesting HT/no-HT comparison:

User​	CPU​	Cores​	Threads​	Time​
JoeRambo​	i9-12900K​	8​	16​	212.656​
JoeRambo​	5800X​	8​	16​	263.424​
Det0x​	5950X​	8​	16​	230.797​
Det0x​	5950X​	8​	8​	368.309​
JoeRambo​	i9-12900K​	8​	8​	328.88​

 

[AdamK47]

Ran it on the new 12900KS. Stable clocks.

 

[nicalandia]

Quick, someone test 5800X3D

 

[Markfw]

There are a lot more tests when I ran it, and I did the cathedral test

 

[aclos3]

Edit: to add a 2600X (below).

One more 5900X. PBO on with memory at 3600Mhz.


2600X:

 

[aclos3]

And my old file server: E3-1230 V2. Similar to an Ivy Bridge i7-3770, I think.

 

[Markfw]

Accidentally did medium stress

 

[AdamK47]

Ran it on the new 12900KS. Stable clocks.

Still had room for improvement.
Upped the L3/Uncore to a stable 4.4GHz. Tweaked memory sub-timings.

 

[ZGR]

Stock 5800X3D -20mV offset with xmp profile.

 

[burninatortech4]

A little low TDP Zen 3 action

 

[PJVol]

Auto PBO2

 

[JoeRambo]

No 5800X3D tests yet? I suspect it will completely break some of the subtests. If someone does it it, please also post both screens of subtests. Thanks!

 

[igor_kavinski]

No 5800X3D tests yet? I suspect it will completely break some of the subtests. If someone does it it, please also post both screens of subtests. Thanks!

ZGR did and yes, the cache seems to be helping in some of the tests. Updated the chart in the first post.

Page 4 - RapydMark CPU benchmark

Page 4 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[Det0x]

No 5800X3D tests yet? I suspect it will completely break some of the subtests. If someone does it it, please also post both screens of subtests. Thanks!

ZGR did and yes, the cache seems to be helping in some of the tests. Updated the chart in the first post.

Page 4 - RapydMark CPU benchmark

Page 4 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Here are a ~23 seconds faster 5800x3d, still at stock ~4450mhz MT clock.


At 1900:3900 speeds I'm getting ~241s time

 

[igor_kavinski]

Sorry, folks. I'm totally useless with charts. Here's what I could manage in Libre Calc.

 

[AdamK47]

Here are a ~23 seconds faster 5800x3d, still at stock ~4450mhz MT clock.
View attachment 61552

At 1900:3900 speeds I'm getting ~241s time
View attachment 61553

How many seconds of Prime95 would that last?

 

[Det0x]

How many seconds of Prime95 would that last?

Why do you want me to run prime95 and how long do you want me to run it ?

If i didn't make it clear, im running stock CPU clocks..

*edit*
Since no response i just ran 30min small FFT's

.... And 10min large FFT's

Have 1hour 10 cycles testmem5 with 1usmus cfg with same settings here

 

[Makaveli]

I've updated my original spreadsheet. My copy only has quad cores and up i've removed the dual core chips as there scores are slower and they mess with the graph.

 

[ZGR]

Definitely seems like this benchmark LOVES fast RAM with tight latency. My lazy stock XMP profile vs tuned RAM is absolutely a worthwhile time investment! I am curious to see if I can get tighter timings, but I did do feel like I got lucky getting this kit to work so easily for 64GB.

Gonna be interesting to see how far this benchmark scales with higher and higher speed DDR5 in the future.

 

[PJVol]

I've updated my original spreadsheet

I think the 5700G score in the chart is not fair to this little workhorse, so here is mine:
(Auto PBO2, 4000CL15)

Spoiler: report

Tests report generated by RapydMark v1.2a (64 bits) https://www.frikiscape.com/rapydmark/
=====================================
Date: 05-17-2022 (m-d-y)
Time: 13:27:04
CPU: AMD Ryzen 7 5700G with Radeon Graphics (8 Cores)
Threads available: 16
GPU: AMD Radeon(TM) Graphics
Resolution: 3440 x 1440
Operating System: Windows 10 Pro Build 19044
System memory: 14229 MB
Stress level: High
System ID: 01BB150B
=====================================
Results of all tests:

Total time (all tests): 213.704 seconds

-------------------------------------
Test Type: Matrix multiplication
Iterations: 400M | Performance: 165,152,768 it/s (16th) | Time: 2.422 s

-------------------------------------
Test Type: MulDivAddSub (int,double,int64_t)
Iterations: 400M | Performance: 32,819,168 it/s (16th) | Time: 12.188 s

-------------------------------------
Test Type: Trigonometric functions
Iterations: 400M | Performance: 48,024,972 it/s (16th) | Time: 8.329 s

-------------------------------------
Test Type: Floating point operations per second
Iterations: 16K | Performance: 2,745 it/s 504.200 GFLOPS (16th) | Time: 5.828 s

-------------------------------------
Test Type: Find prime numbers SQR method
Iterations: 200M | Performance: 17,628,910 it/s (16th) | Time: 11.345 s

-------------------------------------
Test Type: Fibonachi sequence (92)
Iterations: 4000M | Performance: 377,002,848 it/s (16th) | Time: 10.610 s

-------------------------------------
Test Type: Pi approximation Taylor series method
Iterations: 800 | Performance: 115 it/s (16th) | Time: 6.954 s

-------------------------------------
Test Type: Greatest Common Divisor (GCD) Euclid's algorithm
Iterations: 1600M | Performance: 182,502,560 it/s (16th) | Time: 8.767 s

-------------------------------------
Test Type: Fill Memory 256 MB
Iterations: 60K | Performance: 3,509 it/s (16th) | Time: 17.095 s

-------------------------------------
Test Type: Copy Memory 256 MB
Iterations: 60K | Performance: 3,146 it/s (16th) | Time: 19.066 s

-------------------------------------
Test Type: Memory cache performance (max 4 threads)
Iterations: 40 | Performance: 12 it/s (4th) | Time: 12.404 s
Last result: SizeMem= 8 kB Iterations= 65536 Time= 4 ms
SizeMem= 16 kB Iterations= 32768 Time= 4 ms
SizeMem= 32 kB Iterations= 16384 Time= 3 ms
SizeMem= 64 kB Iterations= 8192 Time= 4 ms
SizeMem= 128 kB Iterations= 4096 Time= 4 ms
SizeMem= 256 kB Iterations= 2048 Time= 5 ms
SizeMem= 512 kB Iterations= 1024 Time= 19 ms
SizeMem= 1 MB Iterations= 512 Time= 32 ms
SizeMem= 2 MB Iterations= 256 Time= 29 ms
SizeMem= 4 MB Iterations= 128 Time= 26 ms
SizeMem= 8 MB Iterations= 64 Time= 23 ms
SizeMem= 16 MB Iterations= 32 Time= 26 ms
SizeMem= 32 MB Iterations= 16 Time= 28 ms
SizeMem= 64 MB Iterations= 8 Time= 38 ms
SizeMem= 128 MB Iterations= 4 Time= 57 ms


-------------------------------------
Test Type: Files Encrypt AES256 (10 MB)
Iterations: 20K | Performance: 3,820 it/s (16th) | Time: 5.235 s

-------------------------------------
Test Type: Memory Encrypt/Decrypt AES256 (1 MB)
Iterations: 8K | Performance: 723 it/s (16th) | Time: 11.063 s

-------------------------------------
Test Type: Memory Encrypt/Decrypt Blowfish448 (1 MB)
Iterations: 8K | Performance: 1,204 it/s (16th) | Time: 6.641 s

-------------------------------------
Test Type: Memory Compress/Decompress 1 MB (Gzip)
Iterations: 8K | Performance: 990 it/s (16th) | Time: 8.079 s

-------------------------------------
Test Type: Files Compress 10 MB (Gzip)
Iterations: 4K | Performance: 506 it/s (16th) | Time: 7.892 s

-------------------------------------
Test Type: RSA Keys Generator (2048 bits)
Iterations: 2K | Performance: 242 it/s (16th) | Time: 8.250 s

-------------------------------------
Test Type: Hash SHA1/256/512 MD4/5 CRC32 RIPEMD160 (512 bytes)
Iterations: 1M | Performance: 70,791 it/s (16th) | Time: 14.126 s

-------------------------------------
Test Type: Random Numbers Generator (2048 bits)
Iterations: 200M | Performance: 16,160,310 it/s (16th) | Time: 12.376 s

-------------------------------------
Test Type: Bitcoin Address Generator
Iterations: 200K | Performance: 15,401 it/s (16th) | Time: 12.986 s

-------------------------------------
Test Type: Distances between points on Earth (Haversine)
Iterations: 1000M | Performance: 83,001,328 it/s (16th) | Time: 12.048 s

-------------------------------------
Test Type: Cathedral 3D Test
Iterations: 1 | Performance: 0 | Time: 0

 

[JoeRambo]

(Auto PBO2, 4000CL15)

I wonder what is the memory latency for this beast, the score is awesome. If You could also post Aida64 memory latency and Intel Memory Latency checker results, it would be great.

213s is the same as 5Ghz 8C ADL with tuned memory, kinda shows the heavy tax IOD imposes on ZEN3.

 

[PJVol]

If You could also post Aida64 memory latency and Intel Memory Latency checker results, it would be great.

Sure (sorry, don't know what intel checker is, and tbh don't want to)

 

[JoeRambo]

Sure (sorry, don't know what intel checker is, and tbh don't want to)

Thanks for Aida results, 50ns is awesome for non extreme (read non @Det0x) memory setup.

And MLC is good tool to show memory subsystem performance not in idle, but in varying load conditions:

Intel® Memory Latency Checker v3.11

Intel® Memory Latency Checker (Intel® MLC) is a tool used to measure memory latencies and b/w, and how they change with increasing load on the system.

www.intel.com

need to be run from elevated CMD to copy the results.

 

[PJVol]

And MLC is good tool to show memory subsystem performance not in idle, but in varying load conditions:

Unfortunately they (intel) decided to block me by geo IP, so as I said before, i'm not keen to deal with it anyway, but will gladly provide the other info/results by means of available for everyone tools.

 

[igor_kavinski]

And MLC is good tool to show memory subsystem performance not in idle, but in varying load conditions

Can that tool disable AMD's hardware prefetchers too?

 

[Det0x]

Quick look at the results, seems like there is a large difference in "fill memory" numbers.. Is this down to 16GB vs 32GB ?

vs