I've been benching a 7980XE on HWBOT lately. Motherboard started out as R6Apex until it broke. The shortage of replacements drove me to settle for something else for the recent benches. OS has been mostly Windows 7 64 Pro, which doesn't make use of all the goodies in the i9's, but competitive bench scores are better than with Win 10 -- except for those benches the benefit from AVX-512 and those 3D benches using DX12. RAM tuning worked up to 4200MHz on the R6A. The substitute tunes up to ddr4-4000 CL 14.
With water cooling, I can get this one sample to 4.9GHz core clock. With dry ice, it regularly gets to 5.15GHz, which is where I do the competitive benching. Scale your expectations downward from the benchmark scores proportional to the cooling you will use. Needless to say, HCC processors do well with multicore/multithread tests and suffer from slow core clocks in single core tests. A few dry ice benchmarks that might include some familiar to you:
Cinebench r15 - 5265
GPUPI 100m - 3.369 sec
GPUPI for CPU 1b - 1:09
HWBOT x265 1080p - 171.947 fps
HWBOT x265 4K - 41.1 fps
y-Cruncher 25m - 0.436 sec
y-Cruncher 1b - 22.6 sec
Geekbench 3 Multicore - 84023
Geekbench 4 Multicore - 63080
Aida 64 Memory Read Bandwidth - 126400 (on the R6A. Bandwidth would be higher on a motherboard with more RAM sockets.)
The high-core-count i9's have FIVRs for core, mesh etc voltages. Competitive overclockers who use LN2 have found that the FIVRs stop working around -100C. I have only used dry ice down to -65C and had no cold bug trouble of that sort.
With water cooling, I can get this one sample to 4.9GHz core clock. With dry ice, it regularly gets to 5.15GHz, which is where I do the competitive benching. Scale your expectations downward from the benchmark scores proportional to the cooling you will use. Needless to say, HCC processors do well with multicore/multithread tests and suffer from slow core clocks in single core tests. A few dry ice benchmarks that might include some familiar to you:
Cinebench r15 - 5265
GPUPI 100m - 3.369 sec
GPUPI for CPU 1b - 1:09
HWBOT x265 1080p - 171.947 fps
HWBOT x265 4K - 41.1 fps
y-Cruncher 25m - 0.436 sec
y-Cruncher 1b - 22.6 sec
Geekbench 3 Multicore - 84023
Geekbench 4 Multicore - 63080
Aida 64 Memory Read Bandwidth - 126400 (on the R6A. Bandwidth would be higher on a motherboard with more RAM sockets.)
The high-core-count i9's have FIVRs for core, mesh etc voltages. Competitive overclockers who use LN2 have found that the FIVRs stop working around -100C. I have only used dry ice down to -65C and had no cold bug trouble of that sort.
feedmeink: I agree with all you said. AVX takes a larger setback than in 4- or 6-core i7, minimum 3x for AVX and 5x for AVX512. I looked at y-Cruncher (quite sensitive to memory tuning, CPU speed and stabilty, and to the use of AVX) scores. On water, with AVX at 4.5GHz compared to very little AVX using Sandybridge workload at 4.9GHz -- with AVX ran in half the time even with slower clock. The setback earns its keep to retain the efficiency of AVX instructions.
I also agree with what you say about core volts, but please tell me that 1.7v is a typo. I'd prefer 1.17v. Under dry ice, I have had the courage to go to 1.325Vcore to get over 5GHz clock under benchmark loads.
I didn't mention that my RAM tuning was done with 4400 C19 G.Skill sticks. Some of the blame for limiting those to 4200 or 4000 has to fall at the feet of the motherboard as shown by the difference between motherboards and that my 7820K also tuned around 4200 on the R6A with the same RAMs. Nonetheless, the HCC CPUs don't seem able to get to the 4400 that 8700K, 8600K, 7700K have been able to do.
My testbench is equipped with a water chiller. I try to do longer tuning sessions with water temperature just above the dew point to eliminate condensation. The chiller has a tough time keeping up with the total power thrown off by the 7980XE in multicore work. It handles a 6950K or 7820K well enough, so I know it's capability is as designed for a 15C temperature drop at around 120watts, but not at 225watts. I agree that no air cooling nor any AIO water loop would be able to cool this many cores.
I also agree with what you say about core volts, but please tell me that 1.7v is a typo. I'd prefer 1.17v. Under dry ice, I have had the courage to go to 1.325Vcore to get over 5GHz clock under benchmark loads.
I didn't mention that my RAM tuning was done with 4400 C19 G.Skill sticks. Some of the blame for limiting those to 4200 or 4000 has to fall at the feet of the motherboard as shown by the difference between motherboards and that my 7820K also tuned around 4200 on the R6A with the same RAMs. Nonetheless, the HCC CPUs don't seem able to get to the 4400 that 8700K, 8600K, 7700K have been able to do.
My testbench is equipped with a water chiller. I try to do longer tuning sessions with water temperature just above the dew point to eliminate condensation. The chiller has a tough time keeping up with the total power thrown off by the 7980XE in multicore work. It handles a 6950K or 7820K well enough, so I know it's capability is as designed for a 15C temperature drop at around 120watts, but not at 225watts. I agree that no air cooling nor any AIO water loop would be able to cool this many cores.
