Why are some 3090's only hitting ~14k GPU in Timespy bench on 3960x?

[zx128k]

That is a Port Royal result you have there not a Timespy Standard one.

The number of core does not matter on Port Royal, what does matter is the choice of CPU when running SLI.

The 10980XE CPUs are very poor for running SLI due to MESH they use on the cores, JayzTwoCents would have got a better score by about 500 points using a 10900k or even one of the older mainstream gaming CPUs running in a board that has SLI support.

Still that is JayzTwoCents for you.

The Time Spy CPU test does not scale well on processors with 10 or more threads.

page 28 https://s3.amazonaws.com/download-aws.futuremark.com/3dmark-technical-guide.pdf

Port Royal is the same. Your main aim is highest CPU frequency possible and 10 cpu cores/threads. At some point you get more from the higher frequencies of less cores, than adding more cores.

This is how a Core i9-10900K a 10-core/20-thread can get 17k cpu score @ 5.4GHz=5.5GHz. https://www.3dmark.com/spy/14550492 https://www.3dmark.com/spy/14178167

Yet a 24 cores cpu can't get much more. #1 score is just over 17k https://www.3dmark.com/spy/10477583 AMD 3960 @ 4.5GHz 24 cores active (no SMT). All other runs below 17k time spy cpu. Even the 3950 wont get much more than 17k.

 

[Kaapstad]

page 28 https://s3.amazonaws.com/download-aws.futuremark.com/3dmark-technical-guide.pdf

Port Royal is the same. Your main aim is highest CPU frequency possible and 10 cpu cores/threads. At some point you get more from the higher frequencies of less cores, than adding more cores.

This is how a Core i9-10900K a 10-core/20-thread can get 17k cpu score @ 5.4GHz=5.5GHz. https://www.3dmark.com/spy/14550492 https://www.3dmark.com/spy/14178167

Yet a 24 cores cpu can't get much more. #1 score is just over 17k https://www.3dmark.com/spy/10477583 AMD 3960 @ 4.5GHz 24 cores active (no SMT). All other runs below 17k time spy cpu. Even the 3950 wont get much more than 17k.

Sorry you are wrong.

Here is an entry from RSR using an 10980XE and 18 cores but no hyperthreading.

Yes he is using some fancy cooling but that is beside the point, he is using 18 cores.

https://www.3dmark.com/spy/10419319

His CPU score easily demolishes everyone elses in the thread, it is not even close.

I have tried turning cores on and off in the past and I can assure you Timespy supports more than 10.

You will also notice that the above link shows his 10980XE running at 5.3Ghz which is slower than your quoted 10/20 core/thread 10900k at 5.4, the result is not even close.

 

[zx128k]

Sorry you are wrong.

Here is an entry from RSR using an 10980XE and 18 cores but no hyperthreading.

Yes he is using some fancy cooling but that is beside the point, he is using 18 cores.

https://www.3dmark.com/spy/10419319

His CPU score easily demolishes everyone elses in the thread, it is not even close.

I have tried turning cores on and off in the past and I can assure you Timespy supports more than 10.

You will also notice that the above link shows his 10980XE running at 5.3Ghz which is slower than your quoted 10/20 core/thread 10900k at 5.4, the result is not even close.

Not wrong it comes from a top overclocker. Does not matter what cpu you use, 3dmark uses 10 threads in timespy and more greatly reduced returns.

The Time Spy CPU test does not scale well on processors with 10 or more threads.

page 28 of 3d marks manual. Thats why a 5.3GHz 10980XE can get first place on water and beat 5.3GHz 10900k cpus. 10980XE can reach 20k @ 5.2GHz and 18 cores active. https://www.3dmark.com/spy/11288253 A ten core gets 17.5k @ 5.5GHz. https://www.3dmark.com/spy/14601101 So 8 more cores gives only 2.5k score. Which is 2.5k for near double to cores. Thats only for extreme water overclockers. LN2 10900K is 19.7k @ 6.8GHz. https://www.3dmark.com/spy/13290171 https://www.3dmark.com/spy/10859989

The 10980XE bell curve has an average value of 12334, so that is the normal expected performance for the 10980xe but for the 10900k its 13901. The 10900k is on average faster overall. https://www.3dmark.com/search#advanced?test=spy P&cpuId=2612&gpuId=&gpuCount=0&deviceType=ALL&memoryChannels=0&country=&scoreType=physicsScore&hofMode=false&showInvalidResults=false&freeParams=&minGpuCoreClock=&maxGpuCoreClock=&minGpuMemClock=&maxGpuMemClock=&minCpuClock=&maxCpuClock=

I got approx. 14k in timespy with a 10900k which is about average. Thats the 8-pack overclocked bundle 5GHz. Thats with highest LLC and 1.385-1.41volts in cpu-z. Uncore is stock. I won't be messing with it. Just can't be bothered atm for a few hundred/thousand points. Tuning RAM.

The BIOS tells me I could get 5.1GHz SSE, 5GHz avx and uncore 5.1GHz.

Just so no one thinks the 8-pack deal is crap. I can get 4000MHz CL16, uncore 48 seems fine in cinebench r20. Using overclockers uk BIOS volltages and the AI OC feature its ~5GHz multi thread avx-2 and 5.3 ghz single thread. This is with the avx offset at 1. SSE is 5.1GHz in 3dmark timespy cpu. Temps are well under 80c is cinebench r20. This is without changing anything but no stress tests and back to defaults afterwards.

 

[Kaapstad]

Not wrong it comes from a top overclocker. Does not matter what cpu you use, 3dmark uses 10 threads in timespy and more greatly reduced returns. page 28 of 3d marks manual. Thats why a 5.3GHz 10980XE can get first place on water and beat 5.3GHz 10900k cpus. 10980XE can reach 20k @ 5.2GHz and 18 cores active. https://www.3dmark.com/spy/11288253 A ten core gets 17.5k @ 5.5GHz. https://www.3dmark.com/spy/14601101 So 8 more cores gives only 2.5k score. Which is 2.5k for near double to cores. Thats only for extreme water overclockers. LN2 10900K is 19.7k @ 6.8GHz. https://www.3dmark.com/spy/13290171 https://www.3dmark.com/spy/10859989

The 10980XE bell curve has an average value of 12334, so that is the normal expected performance for the 10980xe but for the 10900k its 13901. The 10900k is on average faster overall. https://www.3dmark.com/search#advanced?test=spy P&cpuId=2612&gpuId=&gpuCount=0&deviceType=ALL&memoryChannels=0&country=&scoreType=physicsScore&hofMode=false&showInvalidResults=false&freeParams=&minGpuCoreClock=&maxGpuCoreClock=&minGpuMemClock=&maxGpuMemClock=&minCpuClock=&maxCpuClock=

I got approx. 14k in timespy with a 10900k which is about average. Thats the 8-pack overclocked bundle 5GHz. Thats with highest LLC and 1.385-1.41volts in cpu-z. Uncore is stock. I won't be messing with it. Just can't be bothered atm for a few hundred/thousand points. Tuning RAM.

The BIOS tells me I could get 5.1GHz SSE, 5GHz avx and uncore 5.1GHz.

Just so no one thinks the 8-pack deal is crap. I can get 4000MHz CL16, uncore 48 seems fine in cinebench r20. Using overclockers uk BIOS volltages and the AI OC feature its ~5GHz multi thread avx-2 and 5.3 ghz single thread. This is with the avx offset at 1. SSE is 5.1GHz in 3dmark timespy cpu. Temps are well under 80c is cinebench r20. This is without changing anything but no stress tests and back to defaults afterwards.

Stop moving the goal posts lol.

Here is a top overclocker who I think does know what he is doing (Kingpin) and I say it again, he is using 18 cores on a 9980XE.

https://www.3dmark.com/spy/7129027

He must have it right because this is the top Timespy score on the HOF.

 

[zx128k]

Stop moving the goal posts lol.

Here is a top overclocker who I think does know what he is doing (Kingpin) and I say it again, he is using 18 cores on a 9980XE.

https://www.3dmark.com/spy/7129027

He must have it right because this is the top Timespy score on the HOF.

I stand on my previous argument. They are disabling HT to get the core speed to 6GHz on LN2.

You should have linked https://www.3dmark.com/spy/8627324 thats 28 cores right and a higher score (a small amount and just 4 results). https://www.3dmark.com/spy/5715679 9980xe https://www.3dmark.com/spy/7339982 (from the 5th result) Its because even if there are diminished returns after 10 cores, the main goal is higher frequency. That a 28 cores 5.7GHz cpu is more or less the same as an 18 core 6GHz cpu proves my point. Proves 3d mark right when they state that Time spy cpu does not scale well over 10 cores. Frequency matters more than cores, just 3k cpu between a 10 core cpu and a 28 core cpu. 180 approx. between an 18 core and a 28 core. This is on LN2 only.

18 core 9980xe 6,007 MHz CPU Score 23 404 https://www.3dmark.com/spy/5715679
28 core Intel Xeon W-3175X Processor 5,714 MHz CPU Score 23 581 https://www.3dmark.com/spy/8627324
18 core 10980xe 5,917 MHz CPU Score 22 852 https://www.3dmark.com/spy/10859989
18 cores i9-7980XE 5,606 MHz CPU Score 22 508 https://www.3dmark.com/spy/5566885
16 cores i9-7960X frequency unknown CPU Score 19 413 https://www.3dmark.com/spy/2350571
10/20 core/threads 10900k 6,803 MHz CPU Score 19 707 https://www.3dmark.com/spy/13290171
8/16 core/threads 9900k 6,503 MHz CPU Score 16 437 https://www.3dmark.com/spy/6273503

This is of course different for the extreme version on timespy's cpu tests. There more cores are better. The Intel Xeon W-3175X Processor and AMD Ryzen Threadripper 3990X are top. https://www.3dmark.com/hall-of-fame-2/timespy+cpu+score+extreme+preset/version+1.0

 

[Kaapstad]

I stand on my previous argument. They are disabling HT to get the core speed to 6GHz on LN2.

You should have linked https://www.3dmark.com/spy/8627324 thats 28 cores right and a higher score (a small amount and just 4 results). https://www.3dmark.com/spy/5715679 9980xe https://www.3dmark.com/spy/7339982 (from the 5th result) Its because even if there are diminished returns after 10 cores, the main goal is higher frequency. That a 28 cores 5.7GHz cpu is more or less the same as an 18 core 6GHz cpu proves my point. Proves 3d mark right when they state that Time spy cpu does not scale well over 10 cores. Frequency matters more than cores, just 3k cpu between a 10 core cpu and a 28 core cpu. 180 approx. between an 18 core and a 28 core. This is on LN2 only.

18 core 9980xe 6,007 MHz CPU Score 23 404 https://www.3dmark.com/spy/5715679
28 core Intel Xeon W-3175X Processor 5,714 MHz CPU Score 23 581 https://www.3dmark.com/spy/8627324
18 core 10980xe 5,917 MHz CPU Score 22 852 https://www.3dmark.com/spy/10859989
18 cores i9-7980XE 5,606 MHz CPU Score 22 508 https://www.3dmark.com/spy/5566885
16 cores i9-7960X frequency unknown CPU Score 19 413 https://www.3dmark.com/spy/2350571
10/20 core/threads 10900k 6,803 MHz CPU Score 19 707 https://www.3dmark.com/spy/13290171
8/16 core/threads 9900k 6,503 MHz CPU Score 16 437 https://www.3dmark.com/spy/6273503

This is of course different for the extreme version on timespy's cpu tests. There more cores are better. The Intel Xeon W-3175X Processor and AMD Ryzen Threadripper 3990X are top. https://www.3dmark.com/hall-of-fame-2/timespy+cpu+score+extreme+preset/version+1.0

You are still moving the goal posts

Here is your original comment which is clearly wrong.

With the 10980XE you need to limit the cores to 10. Take the frequency as high as you can. https://www.3dmark.com/pr/419756

Timespy Standard definitely uses more than 10 physical cores as I have pointed out several times.

You got it wrong, move on.

I own a 10980XE system and also a 7980XE system so have had plenty of opportunity to test out how many cores Timespy Standard uses.

 

[zx128k]

You are still moving the goal posts

Here is your original comment which is clearly wrong.



Timespy Standard definitely uses more than 10 physical cores as I have pointed out several times.

You got it wrong, move on.

I own a 10980XE system and also a 7980XE system so have had plenty of opportunity to test out how many cores Timespy Standard uses.

Timespy is optimised for 8 cores, when you go past 10 the performance dies because there is not enough threads to use the extra cores.
Thats why people are disabling HT and SMT. You dont need them and they reduce the maximum frequency you can reach.

CPU Average score

W-3175X Average score: 11530 https://www.3dmark.com/search#advan...ck=&maxGpuMemClock=&minCpuClock=&maxCpuClock=
10980XE Average score: 12345 https://www.3dmark.com/search#advan...ck=&maxGpuMemClock=&minCpuClock=&maxCpuClock=
9980xe Average score: 11491 https://www.3dmark.com/search#advan...ck=&maxGpuMemClock=&minCpuClock=&maxCpuClock=
i9-7980XE Average score: 11664 https://www.3dmark.com/search#advan...ck=&maxGpuMemClock=&minCpuClock=&maxCpuClock=
10900k Average score: 13905 https://www.3dmark.com/search#advan...ck=&maxGpuMemClock=&minCpuClock=&maxCpuClock=
9900k Average score: 10983 https://www.3dmark.com/search#advan...ck=&maxGpuMemClock=&minCpuClock=&maxCpuClock=

10 cores has the highest average score. So on average the 10900k is the fastest.

Ryzen cpu's the cores get faster as the core count increases. You want frequency first, if you then get more cores thats okay. So the fastest cpu should be the 3950x and the every cpu in order.
AMD Ryzen 9 3950X Average score: 13371 https://www.3dmark.com/search#advan...ck=&maxGpuMemClock=&minCpuClock=&maxCpuClock=
AMD Ryzen 9 3900XT Average score: 12327 https://www.3dmark.com/search#advan...ck=&maxGpuMemClock=&minCpuClock=&maxCpuClock=
AMD Ryzen 9 3900X Average score: 12190 https://www.3dmark.com/search#advan...ck=&maxGpuMemClock=&minCpuClock=&maxCpuClock=
AMD Ryzen 9 3900 Average score: 11197 https://www.3dmark.com/search#advan...ck=&maxGpuMemClock=&minCpuClock=&maxCpuClock=
AMD Ryzen 7 3800XT Average score: 10189 https://www.3dmark.com/search#advan...ck=&maxGpuMemClock=&minCpuClock=&maxCpuClock=
AMD Ryzen 7 3800X Average score: 9918 https://www.3dmark.com/search#advan...ck=&maxGpuMemClock=&minCpuClock=&maxCpuClock=
AMD Ryzen 7 3700X Average score: 9675 https://www.3dmark.com/search#advan...ck=&maxGpuMemClock=&minCpuClock=&maxCpuClock=

The 3950x has double the cores as the 3800xt but the performance sees a tiny increase. This is very diferent from the outcome in timespy extreme cpu.

AMD maximum timespy cpu scores.

AMD Ryzen 9 3950X Max Boost ClockUp to 4.7GHz 16 cores, https://www.3dmark.com/spy/12681488 Highest all core frequency 4.7GHz CPU Score 17 173
AMD Ryzen 9 3900XT Max Boost ClockUp to 4.7GHz 12 cores https://www.3dmark.com/spy/14168711 Highest all core frequency 4.651GHz CPU Score 15 413
AMD Ryzen 9 3900X Max Boost ClockUp to 4.6GHz 12 cores https://www.3dmark.com/spy/8467964 Highest all core frequency 4.7GHz CPU Score 15 406 RAM from two different manufactures
AMD Ryzen 9 3900 Max Boost ClockUp to 4.3GHz 12 cores https://www.3dmark.com/spy/14257733 Highest all core frequency 4.4GHz CPU Score 13 885
AMD Ryzen 9 3800XT Max Boost ClockUp to 4.7GHz 8 cores https://www.3dmark.com/spy/14231409 Highest all core frequency 4.7GHz CPU Score 12 156
AMD Ryzen 9 3800X Max Boost ClockUp to 4.5GHz 8 cores https://www.3dmark.com/spy/8798682 Highest all core frequency 4.55GHz CPU Score 12 227
AMD Ryzen 9 3700X Max Boost ClockUp to 4.4GHz 8 cores https://www.3dmark.com/spy/8418587 Highest all core frequency 4.3GHz CPU Score 13 988
The 10900k is the fastest cpu on average at 10 cores. 3950X with its 16 cores and 4.7GHz boost is next.

The pattern is easy to see the 10900k reaches the highest average score at 10 cores. Timespy extreme is different, there you can use all the thread available. 3DMark tells you to use the extreme version for higher core count cpu's. Tells you more than 10 cores will see very diminished returns.

Ryzen is a little different, RAM has a big impact and thus more important. You can see that with the 3700x vs the 3900. 8 vs 12 cores. 4.3GHz vs 4.4GHz. Why the 3700x beats the 3800xt, 3900 at maximum overclock and is as fast as the average 10900k.

For 10 cores the 3950x has the fastest all core overclock and with the right RAM tightening should be the fastest Ryzen cpu. 3900X also reaches 4.7Ghz but two different RAM kits wont be good. The difference between a 12 core at 4.4GHz and one at 4.6GHz seems big but RAM overclocking increases performance massively. This is how a 3700x can match a 3900, this is with 100Mhz slower multi core speeds and four less cores.

RAM wise the 3700x is correctly overclocked which is why this score is 14k. An 8 core cpu faster than the average 10900k. The issue is how well the rest of the CPU's are overclocked RAM wise. Without the same setup its hard to see which core count is optimal once you overclock for Ryzen. Note you just get 1425 points for 4 extra cores and 350Mhz speed increase. Most of the performance is at 8 cores.

 

[Kaapstad]

Timespy is optimised for 8 cores, when you go past 10 the performance dies because there is not enough threads to use the extra cores.
Thats why people are disabling HT and SMT. You dont need them and they reduce the maximum frequency you can reach.

CPU Average score

W-3175X Average score: 11530 https://www.3dmark.com/search#advanced?test=spy P&cpuId=2429&gpuId=&gpuCount=0&deviceType=ALL&memoryChannels=0&country=&scoreType=physicsScore&hofMode=false&showInvalidResults=false&freeParams=&minGpuCoreClock=&maxGpuCoreClock=&minGpuMemClock=&maxGpuMemClock=&minCpuClock=&maxCpuClock=
10980XE Average score: 12345 https://www.3dmark.com/search#advanced?test=spy P&cpuId=2550&gpuId=&gpuCount=0&deviceType=ALL&memoryChannels=0&country=&scoreType=physicsScore&hofMode=false&showInvalidResults=false&freeParams=&minGpuCoreClock=&maxGpuCoreClock=&minGpuMemClock=&maxGpuMemClock=&minCpuClock=&maxCpuClock=
9980xe Average score: 11491 https://www.3dmark.com/search#advanced?test=spy P&cpuId=2419&gpuId=&gpuCount=0&deviceType=ALL&memoryChannels=0&country=&scoreType=physicsScore&hofMode=false&showInvalidResults=false&freeParams=&minGpuCoreClock=&maxGpuCoreClock=&minGpuMemClock=&maxGpuMemClock=&minCpuClock=&maxCpuClock=
i9-7980XE Average score: 11664 https://www.3dmark.com/search#advanced?test=spy P&cpuId=2267&gpuId=&gpuCount=0&deviceType=ALL&memoryChannels=0&country=&scoreType=physicsScore&hofMode=false&showInvalidResults=false&freeParams=&minGpuCoreClock=&maxGpuCoreClock=&minGpuMemClock=&maxGpuMemClock=&minCpuClock=&maxCpuClock=
10900k Average score: 13905 https://www.3dmark.com/search#advanced?test=spy P&cpuId=2612&gpuId=&gpuCount=0&deviceType=ALL&memoryChannels=0&country=&scoreType=physicsScore&hofMode=false&showInvalidResults=false&freeParams=&minGpuCoreClock=&maxGpuCoreClock=&minGpuMemClock=&maxGpuMemClock=&minCpuClock=&maxCpuClock=
9900k Average score: 10983 https://www.3dmark.com/search#advanced?test=spy P&cpuId=2402&gpuId=&gpuCount=0&deviceType=ALL&memoryChannels=0&country=&scoreType=physicsScore&hofMode=false&showInvalidResults=false&freeParams=&minGpuCoreClock=&maxGpuCoreClock=&minGpuMemClock=&maxGpuMemClock=&minCpuClock=&maxCpuClock=

10 cores has the highest average score. So on average the 10900k is the fastest.

Ryzen cpu's the cores get faster as the core count increases. You want frequency first, if you then get more cores thats okay. So the fastest cpu should be the 3950x and the every cpu in order.
AMD Ryzen 9 3950X Average score: 13371 https://www.3dmark.com/search#advanced?test=spy P&cpuId=2546&gpuId=&gpuCount=0&deviceType=ALL&memoryChannels=0&country=&scoreType=physicsScore&hofMode=false&showInvalidResults=false&freeParams=&minGpuCoreClock=&maxGpuCoreClock=&minGpuMemClock=&maxGpuMemClock=&minCpuClock=&maxCpuClock=
AMD Ryzen 9 3900XT Average score: 12327 https://www.3dmark.com/search#advanced?test=spy P&cpuId=2695&gpuId=&gpuCount=0&deviceType=ALL&memoryChannels=0&country=&scoreType=physicsScore&hofMode=false&showInvalidResults=false&freeParams=&minGpuCoreClock=&maxGpuCoreClock=&minGpuMemClock=&maxGpuMemClock=&minCpuClock=&maxCpuClock=
AMD Ryzen 9 3900X Average score: 12190 https://www.3dmark.com/search#advanced?test=spy P&cpuId=2477&gpuId=&gpuCount=0&deviceType=ALL&memoryChannels=0&country=&scoreType=physicsScore&hofMode=false&showInvalidResults=false&freeParams=&minGpuCoreClock=&maxGpuCoreClock=&minGpuMemClock=&maxGpuMemClock=&minCpuClock=&maxCpuClock=
AMD Ryzen 9 3900 Average score: 11197 https://www.3dmark.com/search#advanced?test=spy P&cpuId=2578&gpuId=&gpuCount=0&deviceType=ALL&memoryChannels=0&country=&scoreType=physicsScore&hofMode=false&showInvalidResults=false&freeParams=&minGpuCoreClock=&maxGpuCoreClock=&minGpuMemClock=&maxGpuMemClock=&minCpuClock=&maxCpuClock=
AMD Ryzen 7 3800XT Average score: 10189 https://www.3dmark.com/search#advanced?test=spy P&cpuId=2694&gpuId=&gpuCount=0&deviceType=ALL&memoryChannels=0&country=&scoreType=physicsScore&hofMode=false&showInvalidResults=false&freeParams=&minGpuCoreClock=&maxGpuCoreClock=&minGpuMemClock=&maxGpuMemClock=&minCpuClock=&maxCpuClock=
AMD Ryzen 7 3800X Average score: 9918 https://www.3dmark.com/search#advanced?test=spy P&cpuId=2478&gpuId=&gpuCount=0&deviceType=ALL&memoryChannels=0&country=&scoreType=physicsScore&hofMode=false&showInvalidResults=false&freeParams=&minGpuCoreClock=&maxGpuCoreClock=&minGpuMemClock=&maxGpuMemClock=&minCpuClock=&maxCpuClock=
AMD Ryzen 7 3700X Average score: 9675 https://www.3dmark.com/search#advanced?test=spy P&cpuId=2479&gpuId=&gpuCount=0&deviceType=ALL&memoryChannels=0&country=&scoreType=physicsScore&hofMode=false&showInvalidResults=false&freeParams=&minGpuCoreClock=&maxGpuCoreClock=&minGpuMemClock=&maxGpuMemClock=&minCpuClock=&maxCpuClock=

The 3950x has double the cores as the 3800xt but the performance sees a tiny increase. This is very diferent from the outcome in timespy extreme cpu.

AMD maximum timespy cpu scores.

AMD Ryzen 9 3950X Max Boost ClockUp to 4.7GHz 16 cores, https://www.3dmark.com/spy/12681488 Highest all core frequency 4.7GHz CPU Score 17 173
AMD Ryzen 9 3900XT Max Boost ClockUp to 4.7GHz 12 cores https://www.3dmark.com/spy/14168711 Highest all core frequency 4.651GHz CPU Score 15 413
AMD Ryzen 9 3900X Max Boost ClockUp to 4.6GHz 12 cores https://www.3dmark.com/spy/8467964 Highest all core frequency 4.7GHz CPU Score 15 406 RAM from two different manufactures
AMD Ryzen 9 3900 Max Boost ClockUp to 4.3GHz 12 cores https://www.3dmark.com/spy/14257733 Highest all core frequency 4.4GHz CPU Score 13 885
AMD Ryzen 9 3800XT Max Boost ClockUp to 4.7GHz 8 cores https://www.3dmark.com/spy/14231409 Highest all core frequency 4.7GHz CPU Score 12 156
AMD Ryzen 9 3800X Max Boost ClockUp to 4.5GHz 8 cores https://www.3dmark.com/spy/8798682 Highest all core frequency 4.55GHz CPU Score 12 227
AMD Ryzen 9 3700X Max Boost ClockUp to 4.4GHz 8 cores https://www.3dmark.com/spy/8418587 Highest all core frequency 4.3GHz CPU Score 13 988
The 10900k is the fastest cpu on average at 10 cores. 3950X with its 16 cores and 4.7GHz boost is next.

The pattern is easy to see the 10900k reaches the highest average score at 10 cores. Timespy extreme is different, there you can use all the thread available. 3DMark tells you to use the extreme version for higher core count cpu's. Tells you more than 10 cores will see very diminished returns.

Ryzen is a little different, RAM has a big impact and thus more important. You can see that with the 3700x vs the 3900. 8 vs 12 cores. 4.3GHz vs 4.4GHz. Why the 3700x beats the 3800xt, 3900 at maximum overclock and is as fast as the average 10900k.

For 10 cores the 3950x has the fastest all core overclock and with the right RAM tightening should be the fastest Ryzen cpu. 3900X also reaches 4.7Ghz but two different RAM kits wont be good. The difference between a 12 core at 4.4GHz and one at 4.6GHz seems big but RAM overclocking increases performance massively. This is how a 3700x can match a 3900, this is with 100Mhz slower multi core speeds and four less cores.

RAM wise the 3700x is correctly overclocked which is why this score is 14k. An 8 core cpu faster than the average 10900k. The issue is how well the rest of the CPU's are overclocked RAM wise. Without the same setup its hard to see which core count is optimal once you overclock for Ryzen. Note you just get 1425 points for 4 extra cores and 350Mhz speed increase. Most of the performance is at 8 cores.

With the 10980XE you need to limit the cores to 10. Take the frequency as high as you can. https://www.3dmark.com/pr/419756

You got it wrong, admit it and move on.

The best scores to be had on a 10980XE are using 18 cores and no hyperthreading.

I have tested this many times and so have plenty of other people.

 

[southy1978]

This was my result, 2950x stock speeds, Ryzen balanced power profile, 3090fe stock speeds, 32gb Corsair 3200mhz ram. Old windows 10 install. https://www.3dmark.com/spy/14705382

 

[ketma]

That's bizarre how you get that gpu score....when everything is my rig is faster.

I've given up with this bench mark and looking at the actual inbuilt game benches.

 

[zx128k]

You got it wrong, admit it and move on.

The best scores to be had on a 10980XE are using 18 cores and no hyperthreading.

I have tested this many times and so have plenty of other people.

The best scores are with the Intel Xeon W-3175X Processor but only for 0.1% that overclock for high frequency. You have to remember that 18 cores and 28 core CPU have quad-six channel Memory. They have far better memory bandwidth which affects the score.

The Intel Xeon W-3175X Processor is only 7099 on average. It hits 5,813 MHz. It has 6 memory channels. This is the reason for better performance. i9-10980XE clocks higher (5,917 MHz) but only has 4 channel memory. Yet is slower. The amount between the two, the W-3175X (23 581 cpu points) and the 10980XE (22 852 cpu points) are about 729 points. Explain why 10 extra cores are only about 729 points more. Or to put it another way, have negligible affect on performance.

There are only 21 faster W-3175X than the 10900k. https://www.3dmark.com/search#advanced?test=spy P&cpuId=2429&gpuId=&gpuCount=0&deviceType=ALL&memoryChannels=0&country=&scoreType=physicsScore&hofMode=false&showInvalidResults=false&freeParams=&minGpuCoreClock=&maxGpuCoreClock=&minGpuMemClock=&maxGpuMemClock=&minCpuClock=&maxCpuClock= Score 18k or above. Average score: 11530 If 19707 for the 10900k then 9 faster cpus in the whole world.

There are only 72 faster 10980XE than the 10900k https://www.3dmark.com/search#advanced?test=spy P&cpuId=2550&gpuId=&gpuCount=0&deviceType=ALL&memoryChannels=0&country=&scoreType=physicsScore&hofMode=false&showInvalidResults=false&freeParams=&minGpuCoreClock=&maxGpuCoreClock=&minGpuMemClock=&maxGpuMemClock=&minCpuClock=&maxCpuClock= Score 18k or above. Average score: 12338 If 19707 for the 10900k then 29 faster cpus in the whole world.

Like the 3700x above you go from 9675-13 988 just more or less tuning the memory. 4,313 points. That gets even better when you have more memory channels. For the most extreme overclocks only because thats the only time I could be wrong. When you go from 10 cores to 28 cores and get only 3,145(18 cores)-3974(28 cores) more points over the highest cored 10900k. Are the extra cores really being used? Remember this applies to 40 or so cpus. We are 100% arguing the outliers. Outliers cannot recipient a whole distribution. When you do a bell curve you basically ignore the rare events or outliers. See these example bell curves https://image.slidesharecdn.com/mon...ve-dashboard-design-111-638.jpg?cb=1503940674 https://st4.ning.com/topology/rest/1.0/file/get/2808287757?profile=RESIZE_480x480

For the W-3175X, 10980XE and 10900k. Each bell curve is done by 3dmark. You just have to look at the results. On average for all W-3175X, 10980XE and 10900k cpus based on real world data. Looking at their bell curves the 10900k is on average faster. Thus you dont need more than ten cores is basically mathematically true, reguardless of what happens with the outliners or rare events.

 

[southy1978]

So I would try uninstalling your nvidia drivers using ddu, take the card out and try reseating it & reinstall fresh drivers. Make sure you have the latest chipset drivers & bios for your mobo.
Are you using any monitoring program like afterburner? Just to see utilisation and if it’s hitting a power limit etc

 

[ketma]

Thanks, but every other bench reports correct scores/performance, just this crappy bench that doesn't, BIOS, chipset, etc all up-to-date.
As I mentioned it seemed to get a better score when running with 12 cores which implies there is an underlying issue with TR/Intel high core cpu's.
But I'm not going to run this cpu in game mode, it was purchased as a work horse.

 

[Kaapstad]

The best scores are with the Intel Xeon W-3175X Processor but only for 0.1% that overclock for high frequency. You have to remember that 18 cores and 28 core CPU have quad-six channel Memory. They have far better memory bandwidth which affects the score.

The Intel Xeon W-3175X Processor is only 7099 on average. It hits 5,813 MHz. It has 6 memory channels. This is the reason for better performance. i9-10980XE clocks higher (5,917 MHz) but only has 4 channel memory. Yet is slower. The amount between the two, the W-3175X (23 581 cpu points) and the 10980XE (22 852 cpu points) are about 729 points. Explain why 10 extra cores are only about 729 points more. Or to put it another way, have negligible affect on performance.

There are only 21 faster W-3175X than the 10900k. https://www.3dmark.com/search#advanced?test=spy P&cpuId=2429&gpuId=&gpuCount=0&deviceType=ALL&memoryChannels=0&country=&scoreType=physicsScore&hofMode=false&showInvalidResults=false&freeParams=&minGpuCoreClock=&maxGpuCoreClock=&minGpuMemClock=&maxGpuMemClock=&minCpuClock=&maxCpuClock= Score 18k or above. Average score: 11530 If 19707 for the 10900k then 9 faster cpus in the whole world.

There are only 72 faster 10980XE than the 10900k https://www.3dmark.com/search#advanced?test=spy P&cpuId=2550&gpuId=&gpuCount=0&deviceType=ALL&memoryChannels=0&country=&scoreType=physicsScore&hofMode=false&showInvalidResults=false&freeParams=&minGpuCoreClock=&maxGpuCoreClock=&minGpuMemClock=&maxGpuMemClock=&minCpuClock=&maxCpuClock= Score 18k or above. Average score: 12338 If 19707 for the 10900k then 29 faster cpus in the whole world.

Like the 3700x above you go from 9675-13 988 just more or less tuning the memory. 4,313 points. That gets even better when you have more memory channels. For the most extreme overclocks only because thats the only time I could be wrong. When you go from 10 cores to 28 cores and get only 3,145(18 cores)-3974(28 cores) more points over the highest cored 10900k. Are the extra cores really being used? Remember this applies to 40 or so cpus. We are 100% arguing the outliers. Outliers cannot recipient a whole distribution. When you do a bell curve you basically ignore the rare events or outliers. See these example bell curves https://image.slidesharecdn.com/mon...ve-dashboard-design-111-638.jpg?cb=1503940674 https://st4.ning.com/topology/rest/1.0/file/get/2808287757?profile=RESIZE_480x480

For the W-3175X, 10980XE and 10900k. Each bell curve is done by 3dmark. You just have to look at the results. On average for all W-3175X, 10980XE and 10900k cpus based on real world data. Looking at their bell curves the 10900k is on average faster. Thus you dont need more than ten cores is basically mathematically true, reguardless of what happens with the outliners or rare events.

I will say it for the last time,

This is what you quoted and got wrong.

​

With the 10980XE you need to limit the cores to 10.
Take the frequency as high as you can.
https://www.3dmark.com/pr/419756​

The correct way to use a 10980XE is with all 18 cores and the hyperthreading turned off.

No amount of extra waffle is going to change the fact you got it wrong.

 

[Kaapstad]

Thanks, but every other bench reports correct scores/performance, just this crappy bench that doesn't, BIOS, chipset, etc all up-to-date.
As I mentioned it seemed to get a better score when running with 12 cores which implies there is an underlying issue with TR/Intel high core cpu's.
But I'm not going to run this cpu in game mode, it was purchased as a work horse.

For your CPU it is best to forget this bench and try something like Timespy Extreme where you will get a much bigger boost to the score from your processor.

Timespy Extreme will also stretch your GPU a lot more too and avoid bottlenecks you get with the standard version.

 

[UnderToker]

The ridiculous power draw
The ridiculous pricing on the 3090
The total lack of stock
The performance of Big Navi
The nvidia cancellation of a number of Nvidia SKU’s - the 3080 20gb cancelled
Ngreedia now buying up a LOT of TSMC 7nm stock
It all points to one thing - a much better 7nm Ampere refresh in early 2021 pulling about 300w maximum
Why would a one buy the current 8nm ampere node ?
No way on earth - you'll all be sick in February - just 16 weeks and everyone will be offloading them
Mark my words well

 

[ketma]

^ Why are you posting garbage like this in this thread? I won't be marking your words with anything, and I won't be off-loading anything.

 

[ketma]

For your CPU it is best to forget this bench and try something like Timespy Extreme where you will get a much bigger boost to the score from your processor.

Timespy Extreme will also stretch your GPU a lot more too and avoid bottlenecks you get with the standard version.

Thanks Kaapstad, I'll look into trying that bench.

 

[zx128k]

I will say it for the last time,

This is what you quoted and got wrong.

​

The correct way to use a 10980XE is with all 18 cores and the hyperthreading turned off.

No amount of extra waffle is going to change the fact you got it wrong.

That is correct way to do it and the right mathematical outcome. You see in ever produced item there are outliners. Most of the CPU's will be around the mean or average. There are approx. 40 cpus in the upper 99th percentile that would never, for these cpu's be truely representative of how the W-3175X, 10980XE or 10900k would perform. Most cpu's are not running on LN2. Thus, LN2 CPU's are outliners.

Outliers are a simple concept—they are values that are notably different from other data points, and they can cause problems in statistical procedures. It’s easy to see how a single outlier can distort reality. For a cpu to be faster it most be faster overall builds. If the outliners are the ones that are faster, then the outlier can distort reality. Making you think that cpu is the fastest, then normally that is not the case.

We will generally define outliers as samples that are exceptionally far from the mainstream of the data. For an approximately normal data set, the values within one standard deviation of the mean account for about 68% of the set; while within two standard deviations account for about 95%; and within three standard deviations account for about 99.7%. Now 20 or so cpu's in a data set wont even fit within three standard deviations. Thus extreme overclockering wont affect which cpu is fastest, they would likely be outside three standard deviations and thus outliers. So we do the bell curves and find which CPU is on average faster. We can't pick one CPU example as faster because thats called cherry picking. Taking cherry picked data leads to a false conclusion.

So this is the bell curve for the 10900k, we see that the highest score is indeed 19707. We note the average which is were most of the data will be around. This is 13909 points, all the other CPU's ill have to beat this value.

This is the bell curve of the 10980xe, note the highest score but note the average score 12340. The orange line is the 10900k, note the much higher average performance.

This is the bell curve of the W-3175X, note the highest score but note the average score 5970. The orange line is the 10900k, note the much higher average performance.

Thus a 10900k is the fastest CPU. By stating an outliers is faster thus this data above is wrong. Is basically stating that a very rare event explains normal life. That that outlier is representative of the whole. The reason why the outliers are rare is that most people can't reason that level of performance. So by the bell curve the 10900k data shows that most people reach a higher average performance. This makes the 10900k the faster cpu. There is approx. 40 (this is counting all three cpu's together) or so cpus out of the 100,000's that form this data that are outliers. They do not change the fact that if you pick a 10900k you are nearly certain to get a faster processor. Thus outside of rare events, you just need a 10 core 10900k is compleley true. I was completely perpared to argue about the outliers but I am not wrong in what I was saying.

 

[Kaapstad]

Don't care about graphs, the name of the game is to get as much performance as possible and on a 10980XE that is using 18 physical cores.

I don't care that the 10980XE is slightly less efficient than a 10900k because the extra cores more than make up for it.

A 10900k is for gaming.

A 10980XE or W-3175X is for heavy duty work.

I did once look at building a 10900k for gaming but the single core performance was not up to the job compared to my Ryzen PC. Before you say it remember single core performance is not just about clockspeed.

My little 3950X PC can run games that would flatten any intel based PC. If you don't believe me try playing "Master of Orion" on max settings in a huge galaxy about 400 turns in, it is a very sad sight on an intel setup.