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​

 

[Markfw]

That's still a great score compared to your 5950X.

I got tied of messing with it. Put the expo memory in, set to default, and BAM working perfectly 6000 CL30 38 38 96. 32 gig was $280. I have 2 more kits coming on the 19th.

 

[tamz_msc]

i7-11370H 51 W PL2, 30 W PL1, Tau = 28 s, Best Performance mode in Windows, Ultimate Thermal profile, all in an Inspiron 14 5410:

 

[Det0x]

managed to squeeze in under 110sec

Handtuned 7950x

 

[nicalandia]

managed to squeeze in under 110sec

Handtuned 7950x
View attachment 69670

Thanks for sharing

 

[igor_kavinski]

managed to squeeze in under 110sec

I think this benchmark prefers Intel architecture (maybe compiled with Intel optimizations?)



The 12900KS is less than a second behind the 5950X despite being core deficient. I think the 13900K with DDR5-6400 CL32 could beat the 7950X. And if it doesn't, that would be a surprise.

 

[Carfax83]

The 12900KS is less than a second behind the 5950X despite being core deficient. I think the 13900K with DDR5-6400 CL32 could beat the 7950X. And if it doesn't, that would be a surprise.

Here is my run with underclocked 13900KF at 5.2ghz on P cores and 4.3ghz on E cores with DDR5 7600 CL36 memory. I think something is wrong with the matrix multiplication test as it's much slower on ADL and RPL than it is on Zen, even though my CPU is underclocked.

Memory frequency doesn't seem to make much of a difference at all.

 

[ZGR]

Here is my run with underclocked 13900KF at 5.2ghz on P cores and 4.3ghz on E cores with DDR5 7600 CL36 memory. I think something is wrong with the matrix multiplication test as it's much slower on ADL and RPL than it is on Zen, even though my CPU is underclocked.

Memory frequency doesn't seem to make much of a difference at all.

There is a lot of run to run variance on a few of the benchmarks which is a bummer. Fast and tight memory seemed to scale well on the 5800X3D since my XMP 3600 kit got wrecked by Det0x’s kit. Pretty cool that you are seeing different results with 7600.

I can’t get your picture to load for me, but it could be on my end.

 

[Carfax83]

There is a lot of run to run variance on a few of the benchmarks which is a bummer. Fast and tight memory seemed to scale well on the 5800X3D since my XMP 3600 kit got wrecked by Det0x’s kit. Pretty cool that you are seeing different results with 7600.

Det0x is really good at tuning memory sub timings, but for matrix multiplication, that should be influenced primarily by the core microarchitecture and cache of the CPU I would think.

I can’t get your picture to load for me, but it could be on my end.

Weird, I checked on my phone and the picture was viewable. The score was 133.8s.

 

[lopri]

7700X and DDR5-6400 (Hynix IC)


7700X and DDR5-6400 (Samsung IC)


7700X and DDR5-6200 (64GB, Hynix + Samsung)


Samsung dies BSOD'ed during the run, and upon reboot I ran it on Safe mode. With 4 sticks it won't boot at 6400 but it's stable-ish at 6200. Needs more testing but realistically a 24/7 setup with 4 sticks would be 6000. Voltages jump bigly from 6000 to 6200, and to 6400 (for 2 sticks).

 

[lopri]

Edit: Never mind!

 

[igor_kavinski]

Samsung dies BSOD'ed during the run, and upon reboot I ran it on Safe mode.

I think this is the first ever instance of someone getting a BSOD with this test.

 

[lopri]

Do I get a medal for that

 

[igor_kavinski]

Do I get a medal for that

No. You get something better.

BSOD Achievement Trophy unlocked!

 

[Carfax83]

No idea how @Det0x is so much ahead of everyone else, even other Zen 4 owners What is his secret?

Anyway, broke the 130 second barrier by a frog hair. Got 129.928 seconds at 5.4ghz compared to my earlier score which was done at 5.2ghz.

 

[Det0x]

No idea how @Det0x is so much ahead of everyone else, even other Zen 4 owners What is his secret?

I just spend alot of time optimizing and tweaking my system, and that's why I have multiple world #1 rankings on hwbot (if you know what that site is..)
I uploaded lots of new scores just yesterday, take a look if your interested (i can see igor is here and i know he likes geekbench scores)
*edit*
Mistakenly mixed igor with nicalandia

 

[Markfw]

I just spend alot of time optimizing and tweaking my system, and that's why I have multiple world #1 rankings on hwbot (if you know what that site is..)
I uploaded lots of new scores just yesterday, take a look if your interested (i can see @igor_kavinski is here and i know he likes geekbench scores)

I am too lazy, and also need the most efficiency, so I set mine to CO = -25 and temp = 80 and it draws about 142 watts, and still has like 96% of the speed of the default setup at almost half the power.

 

[Carfax83]

I just spend alot of time optimizing and tweaking my system, and that's why I have multiple world #1 rankings on hwbot (if you know what that site is..)
I uploaded lots of new scores just yesterday, take a look if your interested (i can see igor is here and i know he likes geekbench scores)
*edit*
Mistakenly mixed igor with nicalandia

Yeah I've definitely heard of hwbot. I knew you were a skilled tweaker and optimizer based on what I saw you do with your memory benchmarks in the Zen 4 thread, but I didn't know you were that elite to where you are setting World records! Well, that explains it then

You should start a YouTube channel, as not many YouTubers offer specialized knowledge like that. The only one I can think of is der8auer.

 

[igor_kavinski]

Core i5-1235U on HP ProBook 450 15.6 inch G9 Notebook

From left to right:

Stock - Max DC performance BIOS option enabled - HT off with Max DC enabled



Max DC improved performance a tiny bit but turning off the supposedly "weakest" two HT threads tanks the performance by 33.1247% !!!

 

[LightningZ71]

Not that surprising when the benchmark is heavily influenced by memory performance as it's main constraint per thread in many of the sub-benches. HT has one big advantage in that it can keep a processor busy during memory stalls. It also doesn't help that the max boost clock of the E cores is so low (3.3Ghz on a good day). They're hardly bringing much to the table. This is all aside from the issues that this bench has with variance between nearly identical systems.

 

[igor_kavinski]

@Det0x @Carfax83

Guys, a request, if you don't mind please.

Could you run Rapydmark at different times of the day and note the variance in scores?

I'm trying to figure out if it's just my system (12700K+Z790) that performs inconsistently or if this is just how Rapydmark is (though, that would make all past scores kinda invalid if they could all be significantly higher or lower).

I've been getting times from 245 seconds to 216 seconds on different days and as far as I know, there hasn't been much fluctuation in the ambient temperature so it's not like the CPU should be boosting considerably higher due to lower ambient temps. In fact, I'm pretty sure that some of the worst scores were when it was particularly freakin' cold at night. Now it's daytime with 23C temp and I got less than 216 seconds on stress level High.

Any suggestions for a good easy to install and easy to run free memory intensive multicore benchmark other than CBR23 that performs consistently and finishes in less than 5 minutes?

 

[BoomerD]

 

[Det0x]

@Det0x @Carfax83

Guys, a request, if you don't mind please.

Could you run Rapydmark at different times of the day and note the variance in scores?

I'm trying to figure out if it's just my system (12700K+Z790) that performs inconsistently or if this is just how Rapydmark is (though, that would make all past scores kinda invalid if they could all be significantly higher or lower).

I've been getting times from 245 seconds to 216 seconds on different days and as far as I know, there hasn't been much fluctuation in the ambient temperature so it's not like the CPU should be boosting considerably higher due to lower ambient temps. In fact, I'm pretty sure that some of the worst scores were when it was particularly freakin' cold at night. Now it's daytime with 23C temp and I got less than 216 seconds on stress level High.

Any suggestions for a good easy to install and easy to run free memory intensive multicore benchmark other than CBR23 that performs consistently and finishes in less than 5 minutes?

Try to run at static cpu clockspeed and/or reduce corecount to a minimum.
You could also try to disconnect from internet / check background windows prosess before running the benchmark

 

[Sulaco]

Totally stock, haven't fiddled with anything [yet]

 

[igor_kavinski]

Try to run at static cpu clockspeed and/or reduce corecount to a minimum.
You could also try to disconnect from internet / check background windows prosess before running the benchmark

If I put all core multiplier of 50 in the BIOS (which should be fine for a 12700K I think), the idle temp of all the cores shoots past 90C and the score is lower due to throttling. By leaving the cores at stock setting, idle temps stay below 50 and I get better scores. The PC is disconnected from internet and I don't really want any Windows updates until I'm done testing the system properly. Nothing else runs while Rapydmark is running.

 

[igor_kavinski]

Totally stock, haven't fiddled with anything [yet]

View attachment 75831

Good score. Try going for the tightest memory timings. That helps significantly in this benchmark. I think it could go below 120 seconds if properly tuned.