not impressed with compute performance (tho not surprised), but i AM impressed by its gaming performance!
not impressed with compute performance (tho not surprised), but i AM impressed by its gaming performance!
Updated the charts, Radeon RX 5600 XT is our present champion
Here he/she is, compared to some former champions:
Model/Type GPU Fab
(nm)Shaders TMUs ROPs Compute
UnitsGFLOP SP GFLOP DP GFLOP HP TBP Gflop(SP)
/WattGflop(DP)
/WattGflop(HP)
/Watt Radeon HD 5970 Hemlock XT
=2x Cypress XT 40 2x 1600 2x 80 2x 2 20 4640 928.0 - 294 15.78 3.16 - Radeon HD 7990 Malta XT
=2x Tahiti XT2 28 2x 2048 2x 128 2x 32 32 8192 2048.0 - 375 21.85 5.46 - Radeon RX 5600 XT 1x Navi 10 XLE 7 2304 144 64 36 8066 504.0 16131 160 50.41 3.15 100.81
Last edited by Dirk Broer; 03-16-2020 at 11:11 PM.
can't remember. probably both.
has the 5700/5700XT/5600XT gotten any better at compute since launch?
Here is a recent article on the competition between GPU's https://www.tomshardware.com/feature...e=notification
Challenge me, or correct me, but don't ask me to die quietly.
…Pursuit is always hard, capturing is really not the focus, it’s the hunt ...
And then someone posts this ideal solution, ARM server involved: https://www.anandtech.com/show/15621...e=notification
Challenge me, or correct me, but don't ask me to die quietly.
…Pursuit is always hard, capturing is really not the focus, it’s the hunt ...
Erh.....where's the GPU in the Cavium ThunderX3 server? It is a fantastic ARM CPU, 96 Cores/384 Threads, ARMv8.3 instruction set, but does the ThunderX3 server have PCIe slots?
Edit: I found a Cavium ThunderX2 server with a nVidia GT 710.
BOINC-wise the only caveat is the screaming lack of 64-bit Linux ARM applications.
With the few projects that have such apps you can sure make a big dent using an dual ThunderX3 machine.....768 threads pumping WUs direction project server.
You would need 128 Odroid-N2's to do the same, or 192 Raspberry Pi 4's!
Last edited by Dirk Broer; 03-18-2020 at 02:42 AM.
And here are the new 6000-series:
What would the 6600XT be capable of, I wonder?
Model/Type GPU Fab
(nm)Shaders TMUs ROPs Compute
UnitsGFLOP SP GFLOP DP GFLOP HP TBP Gflop(SP)
/WattGflop(DP)
/WattGflop(HP)
/Watt Radeon RX 6400 1x Navi 24XL 6 768 48 32 12 3,565 222.8 7,130 53 67.26 4.20 134.52 Radeon RX 6500XT 1x Navi 24XT 6 1024 64 32 16 5,765 360.3 11,530 113 51.02 3.19 102.04 Radeon RX 6600 1x Navi 23XL 7 1792 112 64 28 8,928 558.0 17,860 132 67.64 4.23 135.30 Radeon RX 6600XT 1x Navi 23XT 7 2048 128 64 32 10,605 662.8 21,209 160 66.28 4.14 132.56 Radeon RX 6700 1x Navi 22 XTL 7 2304 144 64 36 10,370 648 20,740 180 57.61 3.60 115.22 Radeon RX 6700XT 1x Navi 22 XT 7 2560 192 64 40 11,520 720 23,040 200 57.60 3.60 115.20 Radeon RX 6800 1x Navi 21 XL 7 3840 240 96 60 16,166 1,010 32,333 250 64.66 4.04 129.33 Radeon RX 6800XT 1x Navi 21 XT 7 4608 288 128 72 20,736 1,296 41,472 300 69.12 4.32 138.24 Radeon RX 6900XT 1x Navi 21 XTX 7 5120 320 128 80 23,040 1,440 46,080 300 76.80 4.80 153.60
And here are the competitors from the green camp:
If it is single precision you are after, nVidia is the card. For double or half precision: go AMD. nVidia is also the brand to go for in case of Tensor-based computation at the moment.
Model/Type GPU Fab
(nm)Shaders TMUs ROPs Compute
UnitsGFLOP SP GFLOP DP GFLOP HP TBP Gflop(SP)
/WattGflop(DP)
/WattGflop(HP)
/Watt GeForce RTX 3050 GA106-150 8 2560 80 32 20 9,098 142 9,098 130 69.98 1.09 69.98 GeForce RTX 3060 GA106-300-A1 8 3584 112 64 28 12,740 199 12,740 170 74.94 1.17 74.94 GeForce RTX 3060 Ti GA104-200-A1 8 4864 152 80 38 16,197 253 16,197 200 80.99 1.27 80.99 GeForce RTX 3070 GA104-300-A1 8 5888 184 96 46 20,372 318 20,372 220 92.60 1.45 92.60 GeForce RTX 3080 GA102-200-K1-A1 8 8704 272 96 68 29,768 465 29,768 320 93.03 1.45 93.03 GeForce RTX 3090 GA102-300-A1 8 10496 328 112 82 35,686 558 35,686 350 101.96 1.59 101.96
Last edited by Dirk Broer; 12-12-2022 at 09:29 AM.
Great work on the charts. It is interesting and helpful to see the performance per watt for out next investments.