Based on the 32 bit FP TFLOP the 4080 Super should whip out a little/a lot more production when holding the CPU load even. Ignoring the possibility of being able to run 6x vs 5x on the 3080 ti.
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
you really wont know until you try it. looking at any one spec isnt very meaningful when often there are other things that are the bottleneck. Einstein historically has been pretty memory bound, so the GPU VRAM bandwidth tended to matter a lot. the 4080S might be constrained by the smaller memory bus. even though the 4080S has faster memory, it has lower overall memory bandwidth than the 3080ti.
plus it depends how far down the optimization/tuning rabbit hole one person decides to go down. the most I was able to get out of my 3080Ti with tuning and a "slow" CPU was about 3M ppd, power limited to 300W. and that was already a lot better than most people could get with that card.
On paper: 3080ti 10,240
)
On paper:
3080ti
10,240 cores
12 GB VRAM
384 bit
912.4 GB/s
Base Clock 1365 MHz
Boost Clock 1665 MHz
Memory Clock 1188 MHz
19 Gbps effective
FP32 (float) 34.10 TFLOPS
FP64 (double) 532.8 GFLOPS (1:64)
4080
9,728 cores
16 GB VRAM
256 bit
716.8 GB/s
Base Clock 2205 MHz
Boost Clock 2505 MHz
Memory Clock 1400 MHz
22.4 Gbps effective
FP32 (float) 48.74 TFLOPS
FP64 (double) 761.5 GFLOPS (1:64)
In reality for E@H? Not sure. I think you might be able to run 1 more concurrent task on the 4080, if your CPU can keep up with the GPU.
I forgot I ordered the 4080
)
I forgot I ordered the 4080 Super. It just arrived with everything except the m/b.
It looks like they shipped the m/b separately.
Phil
I thought I was wrong once, but I was mistaken.
Got it. Then, 4080
)
Got it. Then,
4080 --> 4080 Super
9,728 cores --> 10,204
16 GB VRAM --> No change
256 bit --> No change
716.8 GB/s --> 736.3 GB/s
Base Clock 2205 MHz --> 2295 MHz
Boost Clock 2505 MHz --> 2550 MHz
Memory Clock 1400 MHz --> 1438 MHz
22.4 Gbps effective --> 23 Gbps effective
FP32 (float) 48.74 TFLOPS --> 52.22 TFLOPS
FP64 (double) 761.5 GFLOPS --> 816.0 GFLOPS
Based on the 32 bit FP TFLOP
)
Based on the 32 bit FP TFLOP the 4080 Super should whip out a little/a lot more production when holding the CPU load even. Ignoring the possibility of being able to run 6x vs 5x on the 3080 ti.
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
you really wont know until
)
you really wont know until you try it. looking at any one spec isnt very meaningful when often there are other things that are the bottleneck. Einstein historically has been pretty memory bound, so the GPU VRAM bandwidth tended to matter a lot. the 4080S might be constrained by the smaller memory bus. even though the 4080S has faster memory, it has lower overall memory bandwidth than the 3080ti.
plus it depends how far down the optimization/tuning rabbit hole one person decides to go down. the most I was able to get out of my 3080Ti with tuning and a "slow" CPU was about 3M ppd, power limited to 300W. and that was already a lot better than most people could get with that card.
_________________________________________________________________________