Re: [AMBER] Lite Hash Rate and pmemd.cuda

From: Ross Walker <ross.rosswalker.co.uk>
Date: Thu, 5 Aug 2021 13:18:59 -0400

Hi Sergei,

Here's some benchmarks for the RTX3070TI which is a lite hash rate card. It 'appears' to not impact performance although of course there is always likely to be some kind of corner case and the precedent this sets means that NVIDIA can quite easily just cripple your cards with a driver update at some point. As such while it looks like, for amber at least, the LHR cards are okay to use at least for now it does make one wonder what tricks NVIDIA might try to play down the line. As such if you can avoid them and get uncrippled GeForce cards for similar prices that's probably a good idea. Alternatively try to negotiate in some kind of warranty that allows you to return the cards if NVIDIA later cripples them down the line. That might be tricky though unless you are buying a significant number.

All the best
Ross




JAC_PRODUCTION_NVE - 23,558 atoms PME 4fs
-----------------------------------------

      [0] 1 x GPU: | ns/day = 939.31 seconds/ns = 91.98
      [1] 1 x GPU: | ns/day = 943.52 seconds/ns = 91.57
      [2] 1 x GPU: | ns/day = 923.60 seconds/ns = 93.55
      [3] 1 x GPU: | ns/day = 934.36 seconds/ns = 92.47
Multiple Single GPU Run Performance
      [0] 1 x GPU: | ns/day = 927.07 seconds/ns = 93.20
      [1] 1 x GPU: | ns/day = 939.51 seconds/ns = 91.96
      [2] 1 x GPU: | ns/day = 920.20 seconds/ns = 93.89
      [3] 1 x GPU: | ns/day = 927.48 seconds/ns = 93.16
          2 x GPU: | ns/day = 435.66 seconds/ns = 198.32
          4 x GPU: | ns/day = 379.95 seconds/ns = 227.40
Multiple 2xGPU Run Performance
      [0,1] 2 x GPU: | ns/day = 395.44 seconds/ns = 218.49
      [2,3] 2 x GPU: | ns/day = 434.32 seconds/ns = 198.93

JAC_PRODUCTION_NPT - 23,558 atoms PME 4fs
-----------------------------------------

      [0] 1 x GPU: | ns/day = 922.33 seconds/ns = 93.68
      [1] 1 x GPU: | ns/day = 926.39 seconds/ns = 93.26
      [2] 1 x GPU: | ns/day = 923.15 seconds/ns = 93.59
      [3] 1 x GPU: | ns/day = 915.84 seconds/ns = 94.34
Multiple Single GPU Run Performance
      [0] 1 x GPU: | ns/day = 908.13 seconds/ns = 95.14
      [1] 1 x GPU: | ns/day = 924.14 seconds/ns = 93.49
      [2] 1 x GPU: | ns/day = 915.46 seconds/ns = 94.38
      [3] 1 x GPU: | ns/day = 900.00 seconds/ns = 96.00
          2 x GPU: | ns/day = 445.78 seconds/ns = 193.82
          4 x GPU: | ns/day = 315.41 seconds/ns = 273.93
Multiple 2xGPU Run Performance
      [0,1] 2 x GPU: | ns/day = 363.53 seconds/ns = 237.67
      [2,3] 2 x GPU: | ns/day = 438.87 seconds/ns = 196.87

JAC_PRODUCTION_NVE - 23,558 atoms PME 2fs
-----------------------------------------

      [0] 1 x GPU: | ns/day = 497.26 seconds/ns = 173.75
      [1] 1 x GPU: | ns/day = 499.91 seconds/ns = 172.83
      [2] 1 x GPU: | ns/day = 496.88 seconds/ns = 173.88
      [3] 1 x GPU: | ns/day = 494.40 seconds/ns = 174.76
Multiple Single GPU Run Performance
      [0] 1 x GPU: | ns/day = 496.17 seconds/ns = 174.13
      [1] 1 x GPU: | ns/day = 492.91 seconds/ns = 175.29
      [2] 1 x GPU: | ns/day = 493.96 seconds/ns = 174.91
      [3] 1 x GPU: | ns/day = 484.60 seconds/ns = 178.29
          2 x GPU: | ns/day = 240.15 seconds/ns = 359.77
          4 x GPU: | ns/day = 190.10 seconds/ns = 454.50
Multiple 2xGPU Run Performance
      [0,1] 2 x GPU: | ns/day = 236.44 seconds/ns = 365.42
      [2,3] 2 x GPU: | ns/day = 238.84 seconds/ns = 361.75

JAC_PRODUCTION_NPT - 23,558 atoms PME 2fs
-----------------------------------------

      [0] 1 x GPU: | ns/day = 474.98 seconds/ns = 181.90
      [1] 1 x GPU: | ns/day = 478.73 seconds/ns = 180.48
      [2] 1 x GPU: | ns/day = 475.36 seconds/ns = 181.76
      [3] 1 x GPU: | ns/day = 471.81 seconds/ns = 183.12
Multiple Single GPU Run Performance
      [0] 1 x GPU: | ns/day = 474.63 seconds/ns = 182.04
      [1] 1 x GPU: | ns/day = 468.83 seconds/ns = 184.29
      [2] 1 x GPU: | ns/day = 464.10 seconds/ns = 186.17
      [3] 1 x GPU: | ns/day = 466.09 seconds/ns = 185.37
          2 x GPU: | ns/day = 219.39 seconds/ns = 393.82
          4 x GPU: | ns/day = 197.98 seconds/ns = 436.41
Multiple 2xGPU Run Performance
      [0,1] 2 x GPU: | ns/day = 204.83 seconds/ns = 421.81
      [2,3] 2 x GPU: | ns/day = 187.12 seconds/ns = 461.74

FACTOR_IX_PRODUCTION_NVE - 90,906 atoms PME
-------------------------------------------

      [0] 1 x GPU: | ns/day = 188.93 seconds/ns = 457.31
      [1] 1 x GPU: | ns/day = 186.65 seconds/ns = 462.89
      [2] 1 x GPU: | ns/day = 187.20 seconds/ns = 461.53
      [3] 1 x GPU: | ns/day = 186.74 seconds/ns = 462.69
Multiple Single GPU Run Performance
      [0] 1 x GPU: | ns/day = 187.30 seconds/ns = 461.29
      [1] 1 x GPU: | ns/day = 186.50 seconds/ns = 463.28
      [2] 1 x GPU: | ns/day = 187.12 seconds/ns = 461.72
      [3] 1 x GPU: | ns/day = 187.50 seconds/ns = 460.80
          2 x GPU: | ns/day = 56.64 seconds/ns = 1525.52
          4 x GPU: | ns/day = 49.22 seconds/ns = 1755.55
Multiple 2xGPU Run Performance
      [0,1] 2 x GPU: | ns/day = 58.76 seconds/ns = 1470.42
      [2,3] 2 x GPU: | ns/day = 58.04 seconds/ns = 1488.54

FACTOR_IX_PRODUCTION_NPT - 90,906 atoms PME
-------------------------------------------

      [0] 1 x GPU: | ns/day = 181.21 seconds/ns = 476.81
      [1] 1 x GPU: | ns/day = 180.13 seconds/ns = 479.65
      [2] 1 x GPU: | ns/day = 177.59 seconds/ns = 486.51
      [3] 1 x GPU: | ns/day = 177.62 seconds/ns = 486.42
Multiple Single GPU Run Performance
      [0] 1 x GPU: | ns/day = 176.93 seconds/ns = 488.33
      [1] 1 x GPU: | ns/day = 178.19 seconds/ns = 484.87
      [2] 1 x GPU: | ns/day = 177.84 seconds/ns = 485.84
      [3] 1 x GPU: | ns/day = 177.36 seconds/ns = 487.16
          2 x GPU: | ns/day = 55.28 seconds/ns = 1562.97
          4 x GPU: | ns/day = 46.79 seconds/ns = 1846.68
Multiple 2xGPU Run Performance
      [0,1] 2 x GPU: | ns/day = 52.69 seconds/ns = 1639.65
      [2,3] 2 x GPU: | ns/day = 53.45 seconds/ns = 1616.50

CELLULOSE_PRODUCTION_NVE - 408,609 atoms PME
--------------------------------------------

      [0] 1 x GPU: | ns/day = 40.67 seconds/ns = 2124.32
      [1] 1 x GPU: | ns/day = 41.02 seconds/ns = 2106.15
      [2] 1 x GPU: | ns/day = 40.59 seconds/ns = 2128.36
      [3] 1 x GPU: | ns/day = 40.53 seconds/ns = 2131.86
Multiple Single GPU Run Performance
      [0] 1 x GPU: | ns/day = 40.67 seconds/ns = 2124.31
      [1] 1 x GPU: | ns/day = 41.01 seconds/ns = 2106.89
      [2] 1 x GPU: | ns/day = 40.56 seconds/ns = 2130.24
      [3] 1 x GPU: | ns/day = 40.35 seconds/ns = 2141.46
          2 x GPU: | ns/day = 12.75 seconds/ns = 6775.57
          4 x GPU: | ns/day = 10.55 seconds/ns = 8186.05
Multiple 2xGPU Run Performance
      [0,1] 2 x GPU: | ns/day = 10.49 seconds/ns = 8233.40
      [2,3] 2 x GPU: | ns/day = 10.49 seconds/ns = 8234.06

CELLULOSE_PRODUCTION_NPT - 408,609 atoms PME
--------------------------------------------

      [0] 1 x GPU: | ns/day = 39.06 seconds/ns = 2212.04
      [1] 1 x GPU: | ns/day = 39.19 seconds/ns = 2204.43
      [2] 1 x GPU: | ns/day = 39.03 seconds/ns = 2213.80
      [3] 1 x GPU: | ns/day = 38.91 seconds/ns = 2220.36
Multiple Single GPU Run Performance
      [0] 1 x GPU: | ns/day = 39.05 seconds/ns = 2212.81
      [1] 1 x GPU: | ns/day = 38.97 seconds/ns = 2217.12
      [2] 1 x GPU: | ns/day = 38.72 seconds/ns = 2231.27
      [3] 1 x GPU: | ns/day = 38.48 seconds/ns = 2245.09
          2 x GPU: | ns/day = 12.84 seconds/ns = 6728.77
          4 x GPU: | ns/day = 8.55 seconds/ns = 10108.03
Multiple 2xGPU Run Performance
      [0,1] 2 x GPU: | ns/day = 11.75 seconds/ns = 7355.96
      [2,3] 2 x GPU: | ns/day = 12.40 seconds/ns = 6967.04

STMV_PRODUCTION_NPT - 1,067,095 atoms PME
-----------------------------------------

      [0] 1 x GPU: | ns/day = 26.15 seconds/ns = 3304.38
      [1] 1 x GPU: | ns/day = 26.16 seconds/ns = 3303.25
      [2] 1 x GPU: | ns/day = 26.05 seconds/ns = 3316.30
      [3] 1 x GPU: | ns/day = 25.95 seconds/ns = 3329.97
Multiple Single GPU Run Performance
      [0] 1 x GPU: | ns/day = 26.00 seconds/ns = 3323.55
      [1] 1 x GPU: | ns/day = 26.03 seconds/ns = 3319.23
      [2] 1 x GPU: | ns/day = 25.92 seconds/ns = 3333.50
      [3] 1 x GPU: | ns/day = 25.83 seconds/ns = 3345.52
          2 x GPU: | ns/day = 8.22 seconds/ns = 10514.60
          4 x GPU: | ns/day = 6.00 seconds/ns = 14403.88
Multiple 2xGPU Run Performance
      [0,1] 2 x GPU: | ns/day = 8.09 seconds/ns = 10679.47
      [2,3] 2 x GPU: | ns/day = 8.17 seconds/ns = 10579.67

TRPCAGE_PRODUCTION - 304 atoms GB
---------------------------------

      [0] 1 x GPU: | ns/day = 1264.75 seconds/ns = 68.31
      [1] 1 x GPU: | ns/day = 1270.72 seconds/ns = 67.99
      [2] 1 x GPU: | ns/day = 1255.50 seconds/ns = 68.82
      [3] 1 x GPU: | ns/day = 1246.19 seconds/ns = 69.33
Multiple Single GPU Run Performance
      [0] 1 x GPU: | ns/day = 1271.26 seconds/ns = 67.96
      [1] 1 x GPU: | ns/day = 1258.77 seconds/ns = 68.64
      [2] 1 x GPU: | ns/day = 1257.58 seconds/ns = 68.70
      [3] 1 x GPU: | ns/day = 1261.87 seconds/ns = 68.47

MYOGLOBIN_PRODUCTION - 2,492 atoms GB
-------------------------------------

      [0] 1 x GPU: | ns/day = 529.42 seconds/ns = 163.20
      [1] 1 x GPU: | ns/day = 524.94 seconds/ns = 164.59
      [2] 1 x GPU: | ns/day = 524.63 seconds/ns = 164.69
      [3] 1 x GPU: | ns/day = 525.48 seconds/ns = 164.42
Multiple Single GPU Run Performance
      [0] 1 x GPU: | ns/day = 537.55 seconds/ns = 160.73
      [1] 1 x GPU: | ns/day = 528.69 seconds/ns = 163.42
      [2] 1 x GPU: | ns/day = 523.77 seconds/ns = 164.96
      [3] 1 x GPU: | ns/day = 520.45 seconds/ns = 166.01
          2 x GPU: | ns/day = 391.87 seconds/ns = 220.48
          4 x GPU: | ns/day = 355.39 seconds/ns = 243.12
Multiple 2xGPU Run Performance
      [0,1] 2 x GPU: | ns/day = 387.97 seconds/ns = 222.70
      [2,3] 2 x GPU: | ns/day = 356.11 seconds/ns = 242.62

NUCLEOSOME_PRODUCTION - 25,095 atoms GB
---------------------------------------

      [0] 1 x GPU: | ns/day = 12.77 seconds/ns = 6766.15
      [1] 1 x GPU: | ns/day = 12.85 seconds/ns = 6721.95
      [2] 1 x GPU: | ns/day = 12.82 seconds/ns = 6739.20
      [3] 1 x GPU: | ns/day = 12.66 seconds/ns = 6823.50
Multiple Single GPU Run Performance
      [0] 1 x GPU: | ns/day = 12.80 seconds/ns = 6747.39
      [1] 1 x GPU: | ns/day = 12.81 seconds/ns = 6743.28
      [2] 1 x GPU: | ns/day = 12.76 seconds/ns = 6773.32
      [3] 1 x GPU: | ns/day = 12.65 seconds/ns = 6832.69
          2 x GPU: | ns/day = 22.55 seconds/ns = 3832.11
          4 x GPU: | ns/day = 32.79 seconds/ns = 2634.68
Multiple 2xGPU Run Performance
      [0,1] 2 x GPU: | ns/day = 21.24 seconds/ns = 4067.55
      [2,3] 2 x GPU: | ns/day = 21.06 seconds/ns = 4102.63




> On Aug 5, 2021, at 5:07 AM, Sergei Izmailov <sergei.a.izmailov.gmail.com> wrote:
>
> Hi, there!
>
> I want to reiterate a question from Alessandro Contini (
> http://archive.ambermd.org/202105/0171.html).
> It's been 2.5 month since the original question post, so there could be an
> update on a relatively new topic.
>
> Do LHR cards deliver the same performance as non-LHR versions on
> pmemd.cuda?
> Did anyone benchmark them?
>
> Our lab is about to purchase a new GPU-equipped server and miscalculation
> might be quite costly :(
>
> Sincerely,
> Sergei Izmailov
> --
> Biomolecular NMR lab.
> Saint Petersburg State University
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber



_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Aug 05 2021 - 10:30:02 PDT
Custom Search