GPU Crunching on a budget: nVidia

Using your nVidia or AMD Graphics card for BOINC computation.
User avatar
Dirk Broer
Corsair
Corsair
Posts: 1962
Joined: Thu Feb 20, 2014 11:24 pm
Location: Leiden, South Holland, Netherlands
Contact:

#1 GPU Crunching on a budget: nVidia

Post by Dirk Broer »

Kepler, Maxwell and Pascal chips compared for the 'budget card' segment:


Kepler, Maxwell, Pascal and Turing chips compared for the 'budget card' segment:
ModelVideo ChipCUDA coresGFLOP SPGFLOP DPTDP in Watt
GTX 650GK 107-4503848123465
GTX 650 TiGK 106-220768142559110
GTX 650 Ti BoostGK 106-240768150563134
GTX 750GM 107-30051211113555
GTX 750 TiGM 107-40064013054160
GTX 950GM 206-25076815724990
GTX 1050GP 107-30064017335475
GTX 1050 TiGP 107-40076819816275
GTX 1650TU 117-30089629849375
GTX 1650 TiTU 117-4001024353311075
Image
User avatar
scole of TSBT
Boinc Major General
Boinc Major General
Posts: 5980
Joined: Mon Feb 03, 2014 2:38 pm
Location: Goldsboro, (Eastern) North Carolina, USA

#2 Re: GPU Crunching on a budget: nVidia

Post by scole of TSBT »

Just a note though. If you plan to run more than one GTX 1050 / 1050 ti, you get more GFLOPs/watt and more GFLOPs/$ from a single GTX 1060 than two GTX 1050 /1050 ti. The GFLOPs/watt goes up as the GPU model goes up.
Image
User avatar
Alez
[ TSBT's Pirate ]
[ TSBT's Pirate ]
Posts: 10363
Joined: Thu Oct 04, 2012 1:22 pm
Location: roaming the planet

#3 Re: GPU Crunching on a budget: nVidia

Post by Alez »

Efficiency is becoming more and more important to me. The leccy bill was getting crazy. Most of my older cards are shutdown and I try to use the others only through the night on cheap rate and all day through winter rather than use the electric heaters.
Image
The best form of help from above is a sniper on the rooftop....
User avatar
Dirk Broer
Corsair
Corsair
Posts: 1962
Joined: Thu Feb 20, 2014 11:24 pm
Location: Leiden, South Holland, Netherlands
Contact:

#4 Re: GPU Crunching on a budget: nVidia

Post by Dirk Broer »

nVidia just seem to have released a GT1030 specially against the Radeon RX 550. What is so special about the Radeon RX 500 series?
Let's start with the RX 550 (original can be seen here, you might have to register with AMD Users)
Radeon RX 500 Series
Model TypeGPUFab (nm)ShadersTMUROPGFLOP SPGFLOP DPGFLOP HPTBPGflop (SP) WattGflop (DP) WattGflop (HP) Watt
Radeon RX 540 Lexa145123216115171.911515023.021.4423.02
Radeon RX 550Polaris 12145123216121175.712115024.221.5124.22
Radeon RX 560Polaris 1114102464162406150.4 24068030.011.8830.01
Radeon RX 570Polaris 20142048128324784300.0478415031.892.0031.89
Radeon RX 580Polaris 20142304144325792362.0579218531.312.1331.31
Radeon Pro Duo Polaris2x Polaris 201446082886411456716.01145625045.822.8645.82
The RX 550 is a very capable card in view of its power needs, a mere 50 Watt. Those willing to spend a little more on purchase and power bills can choose for the even more capable RX 560
The TBP (Typical Board Power) of the RX 560 is actually 60-80 Watt, so the GFLOPS/Watt may be even better yet.
What does nVidia offer?
Geforce 10 Series
Model TypeGPUFab (nm)ShadersTMUROPGFLOP SPGFLOP DPGFLOP HPTBPGflop (SP) WattGflop (DP) WattGflop (HP) Watt
GeForce GT 1030GP108-300-A11638424895230153031.731.000.50
GeForce GTX 1050GP107-300-A1146404032173354277523.110.720.36
GeForce GTX 1050 TiGP107-400-A1147684832198162317526.410.830.41
GeForce GTX 1060 3GBGP106-300-A1161152724834701085412028.920.900.45
GeForce GTX 1060 6GBGP106-400/410-A1161280804838551206012032.131.000.50
GeForce GTX 1070GP104-200-A11619201206457831819015038.561.210.60
GeForce GTX 1080GP104-400/410-A116256016064822825712818045.711.430.71
GeForce GTX 1080 TiGP102-350-K1-A1163584224881060933216625042.441.330.66
nVidia Titan XGP102-400-A1163584224961015731715925040.631.270.64
nVidia Titan XpGP102-450-A11638402409612150*380*190*25048.611.520.76
*=Normal values unknown yet. For the Titan Xp I only know of the 'Turbo boost' values, at the moment.

Considering the fact that most projects only ask for Single precision, the GT 1030 offers real value. But when we take double precision performance (e.g. MilkyWay) into account, nVidia is fighting a lost battle as they have sacrificed that feature to boost the sales of their professional (Quadro/Tesla) cards. I do not know yet of the effects of the half-precision performance, but nVidia seems to have made the wrong choice there too by offering Half-precision at 1/64th of the Single precision while AMD offers it at the same GFLOPS as the single precision. There ought to be projects where AMD cards smash the nVidia cards, performance-wise.
Image
User avatar
Alez
[ TSBT's Pirate ]
[ TSBT's Pirate ]
Posts: 10363
Joined: Thu Oct 04, 2012 1:22 pm
Location: roaming the planet

#5 Re: GPU Crunching on a budget: nVidia

Post by Alez »

Dirk Broer wrote: Wed May 10, 2017 3:55 pm .... There ought to be projects where AMD cards smash the nVidia cards, performance-wise.
Dirk, I really appreciate the effort you put into producing these tables and it hacks me off that they are not working properly for you. I intend installing a new version of advanced bbcode on the server which may allow you to use tables but no promises. In the mean time I have started rebuilding some of the tables. I intend to complete this once I am finished updating and enhancing the WCG forum.
In the mean time bare with me and let me know if I have missed any tables once I am finished in the Website problems forum.
Alez
Image
The best form of help from above is a sniper on the rooftop....
User avatar
Dirk Broer
Corsair
Corsair
Posts: 1962
Joined: Thu Feb 20, 2014 11:24 pm
Location: Leiden, South Holland, Netherlands
Contact:

#6 Re: GPU Crunching on a budget: nVidia

Post by Dirk Broer »

If your update solves the problem, I can rebuild those tables myself too. I really like how they look under the new version, tartan and all.
Image
User avatar
Alez
[ TSBT's Pirate ]
[ TSBT's Pirate ]
Posts: 10363
Joined: Thu Oct 04, 2012 1:22 pm
Location: roaming the planet

#7 Re: GPU Crunching on a budget: nVidia

Post by Alez »

Dirk Broer wrote: Sat May 13, 2017 1:09 am If your update solves the problem, I can rebuild those tables myself too. I really like how they look under the new version, tartan and all.
I've rebuilt the tables for you and left them in a more human friendly view which you can see if you edit them. Biggest difference between this board and the AMD board is that this board does not understand
it uses [tr=textleft][/tr] and you only have to declare it once. Also declaring [thread ] and tbody is different but not hugely. Anyway, let me know as I mentioned before of any tables I have missed and I will rebuild them for you.
Image
The best form of help from above is a sniper on the rooftop....
User avatar
Dirk Broer
Corsair
Corsair
Posts: 1962
Joined: Thu Feb 20, 2014 11:24 pm
Location: Leiden, South Holland, Netherlands
Contact:

#8 Re: GPU Crunching on a budget: nVidia

Post by Dirk Broer »

Thanks!
It's looking good and I think I can build them myself now too :dance:
Image
User avatar
Dirk Broer
Corsair
Corsair
Posts: 1962
Joined: Thu Feb 20, 2014 11:24 pm
Location: Leiden, South Holland, Netherlands
Contact:

#9 Re: GPU Crunching on a budget: nVidia

Post by Dirk Broer »

Updated and expanded:
Dirk Broer wrote: Wed Oct 26, 2016 11:36 pm Tesla, Fermi, Kepler, Maxwell, Pascal, Turing and Ampère chips compared for the 'budget card' segment:
ModelVideo ChipCUDA versionVideo Speed MHzRAM speed MHzRAM bandwidth GB/sCUDA coresSM/CU countGFLOP SPGFLOP DPTDP in Watt
GTS 250GT 92-4281.1702200064.0012816387-150
GTS 450GK 106-2502.1783360857.73192460150106
GTX 550 TiGF 116-4002.1900410498.50192469158116
GTX 650GK 107-4503.01058500080.0038428123465
GTX 650 TiGK 106-2203.0928540086.407684142559110
GTX 650 Ti BoostGK 106-2403.010326008144.207684150563134
GTX 750GM 107-3005.01085501280.19512411113555
GTX 750 TiGM 107-4005.01085540086.40640513054160
GTX 950GM 206-2505.211886612105.80768615724990
GTX 1050GP 107-3006.114557008112.10640617335475
GTX 1050 TiGP 107-4006.113927008112.10768619816275
GTX 1650 DDR5TU 117-3007.516658000128.008961429849375
GTX 1650 DDR6TU116-150-KA7.5159012000192.008961428498975
GTX 1650 SuperTU116-250-KA7.5172512000192.001280204416138100
RTX 3050GA107-300-A18.5172514000224.00230418794912490
Those 'budget' GTX 1650 cards do between 150 and 210 Euro's though... and 1650 Ti's were nowhere to be found, I replaced them with the DDR6 variant -that also has a 90 Watt TBP sub-variant...
Also added the RTX 3050
Image
User avatar
scole of TSBT
Boinc Major General
Boinc Major General
Posts: 5980
Joined: Mon Feb 03, 2014 2:38 pm
Location: Goldsboro, (Eastern) North Carolina, USA

#10 Re: GPU Crunching on a budget: nVidia

Post by scole of TSBT »

Thanks Dirk. GFLOPs alone don't give an accurate comparison of older GPUs to newer GPUs. GFLOP to GFLOP, the newer GPUs are just faster. I guess due to faster memory and bandwidth speeds. I imagine a GTX 1650 Ti is probably faster than a GTX 780 on projects like PG and Collatz while pulling 75 watts vs 230.
Image
User avatar
Dirk Broer
Corsair
Corsair
Posts: 1962
Joined: Thu Feb 20, 2014 11:24 pm
Location: Leiden, South Holland, Netherlands
Contact:

#11 Re: GPU Crunching on a budget: nVidia

Post by Dirk Broer »

I have now included Cuda version, GPU speed, RAM speed, RAM bandwidth and SM count, but my guess is that the number of CUDA cores and the SM count (and the CUDA version) is more important than the speeds or the bandwidth. For those of use -like me- who still use a GTX 650: swith to a GTX 1650 or a GTX 1650 Ti (or the GTX 1660 Ti, which is even better, credit-wise).
Image
User avatar
Dirk Broer
Corsair
Corsair
Posts: 1962
Joined: Thu Feb 20, 2014 11:24 pm
Location: Leiden, South Holland, Netherlands
Contact:

#12 Re: GPU Crunching on a budget: nVidia

Post by Dirk Broer »

scole of TSBT wrote: Thu Oct 27, 2016 12:20 am Just a note though. If you plan to run more than one GTX 1050 / 1050 ti, you get more GFLOPs/watt and more GFLOPs/$ from a single GTX 1060 than two GTX 1050 /1050 ti. The GFLOPs/watt goes up as the GPU model goes up.
So, here I present the budget plus table:
Tesla, Fermi, Kepler, Maxwell, Pascal, Turing and Ampère chips compared for the 'budget plus card' segment:
ModelVideo ChipCUDA versionVideo Speed MHzRAM speed MHzRAM bandwidth GB/sCUDA coresSM/CU countGFLOP SPGFLOP DPTDP in Watt
GTX 260GT 2001.35761998111.919224476.959.62182
GTX 260 Core 216GT 2001.35761998111.921627536.567.07182
GTX 460GF 1042.16753600115.23367907.275.60160
GTX 560GF 1142.18104000128.033671089.090.72150
GTX 560 TiGF 1142.18234008128.338481263.0105.30170
GTX 660GK 1063.010326008144.296051981.082.56140
GTX 660 TiGK 1043.09806008144.2134472634.0109.80150
GTX 760GK 1043.010326008192.3115262378.099.07170
GTX 760 TiGK 1043.09806008192.3134472634.0109.80170
GTX 960GM 2065.211787012112.2102482413.075.39120
GTX 1060 3GBGP 1066.117088008192.2115293935.0123.00120
GTX 1060 6GBGP 1046.117098008192.21280104275.0136.70120
RTX 2060TU 1067.5168014008336.01920306451.0201.60160
RTX 2060 SuperTU 1067.5165014000448.02176347181.0224.40160
GTX 1660TU 1167.517858004192.11408225027.0157.10120
GTX 1660 SuperTU 1167.5178514000336.01408225027.0157.10125
GTX 1660 TiTU 1167.5177012000288.01536245437.0169.90120
RTX 3060GA 1068.5186014000336.038403014280.0223.20180
Image
Post Reply Previous topicNext topic

Return to “Graphics Processing Unit (GPU)”