Discussion Raptor Lake Build Thread

[DooKey]

Those that have a Raptor Lake build should chime in with what they have. Post your build and system benchmarks if you have any.

My build:

13700KF
MSI Z690 Force WiFi
32gb G.Skill 7200 CL34 DDR5
2tb Samsung 970 EVO NVMe
Thermaltake ToughPower GF3 1000w ATX 3.0 PSU
Gigabyte RTX 4090 Gaming OC
Noctua NH-U14S HSF

Great system that runs cool, quiet and fast. Upgraded a 12700K to this configuration.

Edit: new BIOS out that allows my MB to run DDR5 at 7200


Benches:
CPU at stock, HSF tower setting in BIOS gives me a 28922 CB23 score. (score updated later in thread)

 

[Carfax83]

have another 7950x and its running windows. 1.14 by Ryzen master, also 4.8 ghz, also 32 cores@100%, and its set to 85c.

@Carfax83 I updated this.

That's very good! Mine was at 1.18v but I'd get some slight clock speed regression after several minutes at full load, ie from 5.2ghz to 5.1 but not on all the P cores, just some of them.

Now that's gotten colder where I live, I don't think that would happen anymore.

All the E cores would drop from 4.3ghz to 4.1ghz though. In normal usage like gaming and browsing, both the P cores and E cores operated at max speeds at all times.

 

[Tech Junky]

linux

Open a terminal and type "sensors"

 

[Carfax83]

@Carfax83 what were your cb23 scores ? Now I am curious. Once you post, I will post in the Zen 4 builders thread, to keep this thread clean.

Cool, I'll do it once I get home.

 

[DooKey]

CPU at stock, HSF tower setting in BIOS gives me a 28922 CB23 score.

 

[Carfax83]

CPU at stock, HSF tower setting in BIOS gives me a 28922 CB23 score.

What is HSF tower setting?

 

[Carfax83]

OK here are my scores at 5.2ghz 180w and 1.18v.

Multithreaded:

Single threaded:

 

[MarkPost]

I just used Cinebench and a few games plus regular browsing. Cinebench tests CPU stability on at a high power draw, while gaming and browsing at moderate to low power draw. All of them are important, especially because my CPU is undervolted.

Whenever you undervolt a CPU, you can't just rely on Cinebench to test stability. In the past I've used Prime95 and other apps to test stability and while it passed, I got crashes during simple things like idling, browsing and gaming. That's because the CPU was undervolted and the voltage would drop too low during idle and low power states.

So far I haven't had a single CPU related crash yet since I dialed in my settings. During the initial testing phase I did though. I had a memory related crash two days ago while gaming when I tested my RAM at 7800mhz without adjusting any voltages. Switching back to DDR5 7600mhz fixed that problem.

I see. I noted something odd in the screenshot you posted while running cinebench:



CPU usage is only 76% so power consumption is lower than when CPU is running at 100% load (logically). I never saw that behavior running cinebench :?

 

[DooKey]

What is HSF tower setting?

It's a setting in the MSI BIOS that allows you to select what type cooling you have and then it sets the max power for you.

 

[Carfax83]

I see. I noted something odd in the screenshot you posted while running cinebench:

CPU usage is only 76% so power consumption is lower than when CPU is running at 100% load (logically). I never saw that behavior running cinebench :?

I had throttled a bit, likely due to being so undervolted. The voltage probably couldn't sustain the max power draw.

I guess I could easily remedy that if I wanted to, but what's the use? 100mhz on the P core difference and 200mhz on the E core in a workload that I'll never use my rig for other than for testing purposes.

In my daily use, I haven't seen any downclocking at all and that's much more important to me.

 

[Carfax83]

It's a setting in the MSI BIOS that allows you to select what type cooling you have and then it sets the max power for you.

That's actually a great idea on their part. Because most people will just assume their $45 heatsink and fan combo is going to handle Raptor Lake at default settings.

@Markfw I uploaded screenshots of my single runs at 5.2ghz for ST and MT.

 

[Markfw]

That's actually a great idea on their part. Because most people will just assume their $45 heatsink and fan combo is going to handle Raptor Lake at default settings.

@Markfw I uploaded screenshots of my single runs at 5.2ghz for ST and MT.

Here is a link to mine. http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=threads/zen-4-builders-thread.2605910/post-40873867

36k something. Looks like yours is faster at 180 watt. Maybe I should try something less ? Is there any benchmark that will stress for 10 minutes or more ?

 

[Carfax83]

Here is a link to mine. http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=threads/zen-4-builders-thread.2605910/post-40873867

36k something. Looks like yours is faster at 180 watt. Maybe I should try something less ? Is there any benchmark that will stress for 10 minutes or more ?

OK @Markfw and @MarkPost, I made a mistake so my earlier screenshot with the HWinfo stuff was not accurate. I used print screen to capture that image, and I just did a 10 minute run where my CPU was pulling the full 220w throughout the run. The 180w comes from the CPU downclocking a bit due to the Cinebench loop ending and then beginning again. The downclock occurs very quickly and is similar to the voltage spike at the beginning of the run which is also anomalous.

The average voltage throughout the 10 minute run was around 1.196v.

Here is a much more accurate depiction using the standard 10 minute test (30 passes) taken directly AFTER the test:

And here is a screen grab DURING the run itself. The package power now shows 220w and not 182w.

 

[Carfax83]

Here is a link to mine. http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=threads/zen-4-builders-thread.2605910/post-40873867

36k something. Looks like yours is faster at 180 watt. Maybe I should try something less ? Is there any benchmark that will stress for 10 minutes or more ?

Yeah I just ran the full 10 minute test and updated my results. The earlier figures I gave were incorrect. To do the full 10 minute run, click on File in the upper left corner of the CBR23 window, then select Advanced Benchmark. Once you do that, it will loop for 10 minutes and then generate a score.

My final score was 38539, which is slightly less than my single run score of 39233.

 

[Markfw]

I can't see whats the score ? BTW, mine never changes under load. The 1,14v is under 100% load indefinitely.(WCG) and the CB 23 was after 10 minutes.

 

[Carfax83]

I can't see whats the score ? BTW, mine never changes under load. The 1,14v is under 100% load indefinitely.(WCG) and the CB 23 was after 10 minutes.

The score was 38,539 at 220w and 1.196v for 10 minutes.

 

[Markfw]

The score was 38,539 at 220w and 1.196v for 10 minutes.

Well, its a little higher score and a little higher voltage and wattage. But I am happy with my performance at this level. I would say at the same wattage they are very close in performance. For at least 10 minutes. A one hour run on all cores would be interesting, and I would have loved to have done an AVX-512 run, but the newer Raptor lake does not support that (my old 12700F with no e-cores does !)

Thanks for the benches !

 

[Carfax83]

Well, its a little higher score and a little higher voltage and wattage. But I am happy with my performance at this level. I would say at the same wattage they are very close in performance. For at least 10 minutes. A one hour run on all cores would be interesting, and I would have loved to have done an AVX-512 run, but the newer Raptor lake does not support that (my old 12700F with no e-cores does !)

Thanks for the benches !

No problem. BTW, I don't think CBR23 supports AVX-512. For vector extensions, AVX and AVX2 are supported.

And this is the stock performance for both the 13900K and 7950x in CBR23. The stock performance isn't much higher than my single run score, which was 39233 so I'm not really losing that much from the underclock and undervolt.

 

[MarkPost]

OK @Markfw and @MarkPost, I made a mistake so my earlier screenshot with the HWinfo stuff was not accurate. I used print screen to capture that image, and I just did a 10 minute run where my CPU was pulling the full 220w throughout the run. The 180w comes from the CPU downclocking a bit due to the Cinebench loop ending and then beginning again. The downclock occurs very quickly and is similar to the voltage spike at the beginning of the run which is also anomalous.

The average voltage throughout the 10 minute run was around 1.196v.

Here is a much more accurate depiction using the standard 10 minute test (30 passes) taken directly AFTER the test:

Spoiler

And here is a screen grab DURING the run itself. The package power now shows 220w and not 182w.

well thats not what is observed in your previous screenshot:

Spoiler

CB iteration is fully running there, its not between two iterations...

 

[MarkPost]

No problem. BTW, I don't think CBR23 supports AVX-512. For vector extensions, AVX and AVX2 are supported.

And this is the stock performance for both the 13900K and 7950x in CBR23. The stock performance isn't much higher than my single run score, which was 39233 so I'm not really losing that much from the underclock and undervolt.

Spoiler

thats not stock perfomance on the 13900K part. Thats with the unlimited power thing enabled by default

 

[Carfax83]

well thats not what is observed in your previous screenshot:

CB iteration is fully running there, its not between two iterations...

I see your point and I agree that it's strange, because on my other runs including the full 10-minute ones I didn't observe that behavior at all.

I have no real explanation for it to be honest. Could not enough voltage cause that to occur?

*Edit* I now have an answer. I just did a Cinebench run for 5 minutes at 5.5ghz with the PL1 and PL2 set to 253w, and I noticed that right after the CPU finishes rendering and cycles to do it again, it temporarily drops power consumption until it raises again. Apparently, some latency is involved in how quickly it can ramp up again and that is why it showed 182w in that image even when it was processing, because it can sometimes take some time to fully ramp back up.

 

[Carfax83]

I did a run at 5.5ghz with stock PL1 and PL2 253w using the same negative offset -0.100v for over 5 minutes. Ran very well and the package power maxed out at 88c which is still very tolerable. I think I'm going to stick with my 5.2ghz settings though.

 

[MarkPost]

so undervolted it downclocks from 5.5 to 5-2-5.3 at 253W (and we dont know if its really stable). too much power hungry for my taste

 

[Carfax83]

so undervolted it downclocks from 5.5 to 5-2-5.3 at 253W (and we dont know if its really stable). too much power hungry for my taste

It would probably need 1.2xx volts to stop it from downclocking, but even at this low voltage of 1.137v it ran perfectly fine.

 

[lopri]

Had to return Arctic Liquid Freezer 420 because the cold plate did not fit the socket. Running 13700K with Mugen 5 right now, and I do lose out some speed at multicore benchmarks.

Edit: it's running on DDR4

 

[lopri]

Multi-core benchmarks take it instantly to 100C/260W. I do not think I can blame lower scores to the motherboard. The chip does "Max Turbo" at 5386 MHz, which is fleeting at best, but it seem capable of 5287 MHz on all cores indefinitely, at least in this benchmark. Prime95 brings it down to 4.8~4.9 GHz.

I am looking for a smaller AIO, a 2-fan version (either 260mm or 280mm) with a wide socket compatibility. Any recommendation?