Question Raptor Lake - Official Thread

[Hulk]

Ladies and Gentlemen. I have scoured the internet for all available information and testing regarding the 13900K and 7950X. After a very comprehensive Borg-like assimilation of the data I have come to a conclusion....

The 13900K and the 7950X are very evenly matched in performance. The 7950X however, is more efficient.

Assimilate the one that performs better in the majority of applications you use, take note of power if you will be running full load (DC) for hours on end, and factor price into this buying equation. You can't go wrong either way.

 

[In2Photos]

Ladies and Gentlemen. I have scoured the internet for all available information and testing regarding the 13900K and 7950X. After a very comprehensive Borg-like assimilation of the data I have come to a conclusion....

The 13900K and the 7950X are very evenly matched in performance. The 7950X however, is more efficient.

Assimilate the one that performs better in the majority of applications you use, take note of power if you will be running full load (DC) for hours on end, and factor price into this buying equation. You can't go wrong either way.

Valiant effort sir, but the attempt is likely futile! 😂

 

[Hitman928]

It crossed my mind today that Ashes of Singularity is no more part of latest gaming benchmarking suites, at least i dont recall seeing any of the new CPUs and GPUs being tested in it. What´s up, is it too dated at this point to be even useful as a performance benchmark, even if nobody really plays it anymore? Cause it was never ultra-popular game, got nowhere near Supreme Commander, but it was highly threaded and whatnot, yet CPUs/GPUs used to be tested in it a lot.

I think it's just old at this point. Most reviewers like to refresh their test suites from time to time and I'm sure it's been long enough that everyone has cycled that game out.

 

[Hulk]

Valiant effort sir, but the attempt is likely futile! 😂

Touche!

 

[Henry swagger]

Ladies and Gentlemen. I have scoured the internet for all available information and testing regarding the 13900K and 7950X. After a very comprehensive Borg-like assimilation of the data I have come to a conclusion....

The 13900K and the 7950X are very evenly matched in performance. The 7950X however, is more efficient.

Assimilate the one that performs better in the majority of applications you use, take note of power if you will be running full load (DC) for hours on end, and factor price into this buying equation. You can't go wrong either way.

Ladies and Gentlemen. I have scoured the internet for all available information and testing regarding the 13900K and 7950X. After a very comprehensive Borg-like assimilation of the data I have come to a conclusion....

The 13900K and the 7950X are very evenly matched in performance. The 7950X however, is more efficient.

Assimilate the one that performs better in the majority of applications you use, take note of power if you will be running full load (DC) for hours on end, and factor price into this buying equation. You can't go wrong either way.

Techpowerup doesn't agree with you

 

[Markfw]

Techpowerup doesn't agree with you

Most everyone else does agree with him. Whats your point ? Trolling again ?

 

[Hulk]

Techpowerup doesn't agree with you

You are wrong.

Techpowerup conclusion. Just as I stated. Thank you and good night.

• Solid performance improvements in both gaming and applications
• Faster than all Ryzens in gaming
• Matches Ryzen 9 7950X in applications
• • Very high power usage
  • 5.8 GHz boost on only two cores
  • Demanding cooling requirements / high temperatures

 

[Herald]

Ladies and Gentlemen. I have scoured the internet for all available information and testing regarding the 13900K and 7950X. After a very comprehensive Borg-like assimilation of the data I have come to a conclusion....

The 13900K and the 7950X are very evenly matched in performance. The 7950X however, is more efficient.

Assimilate the one that performs better in the majority of applications you use, take note of power if you will be running full load (DC) for hours on end, and factor price into this buying equation. You can't go wrong either way.

The 7950x is indeed more efficient at heavy mt scenarios, but at iso wattage the difference is small, 7 to 15%.

In mixed workloads the 13900k is in fact more efficient, by a lot.

In gaming the 13900k is faster. Considerably. Heck a 12900k is faster.

 

[Carfax83]

@Herald, here are my memory benchmarks at DDR5 7800. Lowering the tRAS to 36 really helped me to break the 60ns barrier, though not by much. Before, I had it at 92. I didn't know I could set it that low.

I haven't messed with the tREFI that much yet either, but when I looked in the UEFI it was at 7000 or thereabouts, which is much higher than yours. When I have time, I'll start fiddling with it later. I think my motherboard uses less aggressive sub timings than yours at default.

Updated to a new BIOS which recently released and it seems to have made my system more stable. On my previous BIOS, I could never have run at these settings.

 

[Herald]

@Herald, here are my memory benchmarks at DDR5 7800. Lowering the tRAS to 36 really helped me to break the 60ns barrier, though not by much. Before, I had it at 92. I didn't know I could set it that low.

I haven't messed with the tREFI that much yet either, but when I looked in the UEFI it was at 7000 or thereabouts, which is much higher than yours. When I have time, I'll start fiddling with it later. I think my motherboard uses less aggressive sub timings than yours at default.

Updated to a new BIOS which recently released and it seems to have made my system more stable. On my previous BIOS, I could never have run at these settings.

Trefi works the exact opposite way of any other timing. Increasing trefi makes your ram faster, but also hotter. It basically reduces the amount of time out your ram gets for every refresh.

7000 is normal stock trefi, set it to 65535, should be perfectly stable. Higher than that and you might have temp issues and instability if not actively cooling the ram

 

[Carfax83]

Trefi works the exact opposite way of any other timing. Increasing trefi makes your ram faster, but also hotter. It basically reduces the amount of time out your ram gets for every refresh.

7000 is normal stock trefi, set it to 65535, should be perfectly stable. Higher than that and you might have temp issues and instability if not actively cooling the ram

OK I'm going to try 65535 and report back.

 

[Herald]

OK I'm going to try 65535 and report back.

Thats an old run i had before i was done tuning, have dropped half a ns from the above

 

[Carfax83]

@Herald OK here is the new upload. I didn't use 65535 however, as I still want to test stability to make sure my system can handle it. So I tested 32768 and here are the results. It's a nice improvement I must say. Can't believe this one setting can have such a dramatic effect on performance

BTW, what do you use to monitor RAM temps? HWinfo64 has a section for RAM called "SPD Hub Temperature" and it shows my RAM in the lower 30s, but I don't know how accurate that is. I doubt I will have temperature issues though as the Fractal Design Torrent is an airflow monster and pushes a ton of air directly onto the DIMMs.

Thanks for the help though man, I really appreciate the advice.

 

[Herald]

@Herald OK here is the new upload. I didn't use 65535 however, as I still want to test stability to make sure my system can handle it. So I tested 32768 and here are the results. It's a nice improvement I must say. Can't believe this one setting can have such a dramatic effect on performance

BTW, what do you use to monitor RAM temps? HWinfo64 has a section for RAM called "SPD Hub Temperature" and it shows my RAM in the lower 30s, but I don't know how accurate that is. I doubt I will have temperature issues though as the Fractal Design Torrent is an airflow monster and pushes a ton of air directly onto the DIMMs.

Thanks for the help though man, I really appreciate the advice.

Im using hwinfo. It is accurate in a way, dimms have a thermal probe in there. Now whether the thermal pobe is positioned where it should be, which is on the actual ram chips, i dont know, but generally when stress testing you want to keep that probe below 60c.

Did you test it with something like tm5?

 

[Herald]

@Carfax83 This is the area you were talking about in CP2077, tom's dinner. Only ALD and RPL can consistently hold 100 fps, and even these 2 struggle in front of the restaurant. Zen 4 does decently (still a good 20+ fps behind) but everything else (the 3d included) struggles to hold 50. My 3700x tuned with 3800c15 tuned ram is dropping to the 25-30ies

E cores also help a lot in this area, no idea why. Turning them off noticeably drops performance

This is a completely stock CPU btw

 

[Markfw]

@Carfax83 This is the area you were talking about in CP2077, tom's dinner. Only ALD and RPL can consistently hold 100 fps, and even these 2 struggle in front of the restaurant. Zen 4 does decently (still a good 20+ fps behind) but everything else (the 3d included) struggles to hold 50. My 3700x tuned with 3800c15 tuned ram is dropping to the 25-30ies

E cores also help a lot in this area, no idea why. Turning them off noticeably drops performance

This is a completely stock CPU btw

in both cases (e-cores disabled and only 8 cores on 3700x) you are at a low core count. What is the 7950x in this benchmark with 6000+ ram and the same video card ? I bet its a lot faster.

 

[Timmah!]

@Carfax83 This is the area you were talking about in CP2077, tom's dinner. Only ALD and RPL can consistently hold 100 fps, and even these 2 struggle in front of the restaurant. Zen 4 does decently (still a good 20+ fps behind) but everything else (the 3d included) struggles to hold 50. My 3700x tuned with 3800c15 tuned ram is dropping to the 25-30ies

E cores also help a lot in this area, no idea why. Turning them off noticeably drops performance

This is a completely stock CPU btw

If everything but RPL/ADL and Zen4 struggles, its safe to say reason is RAM.

 

[Herald]

If everything but RPL/ADL and Zen4 struggles, its safe to say reason is RAM.

Not so sure, id like to see ald / rpl with ddr4. I know memory bandwidth plays a big role in heavy RT games (like spiderman) but this specific scene in cyberpunk is tapping out everything, it's single thread bound (when turning off ecores some cores hit 98%) but it also needs CPU resources and fast ram.

 

[Herald]

in both cases (e-cores disabled and only 8 cores on 3700x) you are at a low core count. What is the 7950x in this benchmark with 6000+ ram and the same video card ? I bet its a lot faster.

See for yourself , 5:20

 

[Markfw]

See for yourself , 5:20

I was arguing the 50 fps, and I never saw it in that. I saw 70 something once, but most of the time over 100.

 

[Herald]

I was arguing the 50 fps, and I never saw it in that. I saw 70 something once, but most of the time over 100.

I said the 3d struggles to hold 50, not zen 4.


The whole time he was on that small road with the npcs it was between 70 and 90.

 

[Carfax83]

Im using hwinfo. It is accurate in a way, dimms have a thermal probe in there. Now whether the thermal pobe is positioned where it should be, which is on the actual ram chips, i dont know, but generally when stress testing you want to keep that probe below 60c.

Did you test it with something like tm5?

To be honest, I test with games. There's been too many times when I've used memtest or some other memory stressor, passed it and then when I play a game, the game crashes to desktop or something. To me, games are the best memory stress tests because they very utilize very large amounts of memory.

My favorite game benchmarks are BL3 and Spider-Man Remastered. From my experience, they will crash the system very quickly if the memory can't handle the overclock and timings.

I'm pleased to report that so far, everything is 100% stable

 

[Carfax83]

@Carfax83 This is the area you were talking about in CP2077, tom's dinner. Only ALD and RPL can consistently hold 100 fps, and even these 2 struggle in front of the restaurant. Zen 4 does decently (still a good 20+ fps behind) but everything else (the 3d included) struggles to hold 50. My 3700x tuned with 3800c15 tuned ram is dropping to the 25-30ies

E cores also help a lot in this area, no idea why. Turning them off noticeably drops performance

This is a completely stock CPU btw

Yeah, that's definitely one of the areas I see tested a lot when it comes to CBP 2077. With RT on and crowd density set to high, it really gives CPUs a great workout. Those E cores are being used it seems, I wonder what the game is using them for?

 

[Carfax83]

If everything but RPL/ADL and Zen4 struggles, its safe to say reason is RAM.

Not just RAM, but cache bandwidth as well. RPL has a significant performance advantage over ADL despite using the same core microarchitecture and memory frequency, so the only other possible reason is the cache bandwidth.

RPL has about 85% greater read bandwidth for the L3 cache compared to ADL for instance.

 

[Herald]

To be honest, I test with games. There's been too many times when I've used memtest or some other memory stressor, passed it and then when I play a game, the game crashes to desktop or something. To me, games are the best memory stress tests because they very utilize very large amounts of memory.

My favorite game benchmarks are BL3 and Spider-Man Remastered. From my experience, they will crash the system very quickly if the memory can't handle the overclock and timings.

I'm pleased to report that so far, everything is 100% stable

You should definitely test with TM5, 10cycles of 1usmus or whatever config you prefer. Especially when you are playing with TREFI, if you are not stable it can cause windows corruption and you will need to format.