The Supercomputer in my Basement

[Gorgon]

Being a bit of a nerd I added up the Single-Precision performance of all my GPUs courtesy of the data in TechPowerUp's GPU Database and realized that I have 102.3 TFlops of compute performance.

 

This capacity exceeds the 91.75 TFlops of compute performance of the IBM BlueGene/L Supercomputer which fifteen years ago in November 2004 was ranked as the world's fastest computer.

image courtesy of IBM

 

The Top500 list recently merged with the Green 500 list which measures the energy efficiency of Supercomputers. The top ranked position on the Green 500 list is the DGX SaturnV Volta system at 15.113 GFlops/W.

 

My current power consumption is 2.9kW so my efficiency is 35.289 GFlops/W

[Dissitesuxba11s]

Do you have a list of all the GPUs that you have currently folding?

[Windows7ge]

I always like hearing about when enthusiast gear outperforms 10+ year old enterprise gear. It mean's technology is always moving forward and "The Good Stuff" is trickling down our way.  

[Metallus97]

DAYM BRO!

Thats is actually a really nice efficiency!

even my servers can’t beat your GPUs  

[Gorgon]

8 minutes ago, Dissitesuxba11s said:

Do you have a list of all the GPUs that you have currently folding?

They're in the Build Logs but here's the condensed version:

• 2 GTX 1060 6GB
• 1 GTX 1660 Ti
• 1 GTX 1070 Ti
• 1 RTX 2060
• 1 RTX 2060 Super
• 1 RTX 2070 A
• 5 RTX 2070 Super
• 1 RTX 2080 Super

[Guest]

3 minutes ago, Gorgon said:

They're in the Build Logs but here's the condensed version:

• 2 GTX 1060 6GB
• 1 GTX 1660 Ti
• 1 GTX 1070 Ti
• 1 RTX 2060
• 1 RTX 2060 Super
• 1 RTX 2070 A
• 5 RTX 2070 Super
• 1 RTX 2080 Super

What about those two hydro 2070 Supers?

[Gorgon]

15 minutes ago, Den-Fi said:

What about those two hydro 2070 Supers?

I have 4 RTX 2070 Super Hybrids and one RTX 2070 Super, so 5 total

[Guest]

Just now, Gorgon said:

I have 4 RTX 2070 Super Hybrids and one RTX 2070 Super, so 5 total

Ohhhhh right. I read that 5 as a 1 because my attention span is too short.

[Gorgon]

@GOTSpectrum @palespartanBased on my mix of GPUs a rough calculation suggests our team is providing 2.96 PetaFlops of processing power to Folding@Home or about the same as Sweden's Tetralith Supercomputer currently ranked 74th in the world which has 64,512 CPU Cores

[leadeater]

4 hours ago, Gorgon said:

My current power consumption is 2.9kW so my efficiency is 35.289 GFlops/W

Theirs includes cooling 

[Gorgon]

Just now, leadeater said:

Theirs includes cooling 

Mine includes power at the wall for all the systems excluding 2 10" fans in the basement window and a 24" box fan suspended from the joists. It's free heat (well, expensive heat) this time of year in the northern Hemisphere. 11C this morning walking to work

[palespartan]

21 hours ago, Gorgon said:

@GOTSpectrum @palespartanBased on my mix of GPUs a rough calculation suggests our team is providing 2.96 PetaFlops of processing power to Folding@Home or about the same as Sweden's Tetralith Supercomputer currently ranked 74th in the world which has 64,512 CPU Cores

This is awesome. We as linus tech tips team are in the top 100 super computers in the world. This is the original point of distributed computing to be able to give the power of super computers to work on projects that might otherwise not be able to afford them.

[porina]

Just stumbled on the thread. There is an additional factor to consider in that supercomputers are rated for double-precision flops, whereas consumer GPUs only have high single-precision flops and crippled DP flops. You can approximate one DP FLOPS is equivalent to two SP FLOPS, but not the other way around usually. DP optimised hardware could be split into two SP, but two SP can't usually be combined into one DP.

 

Still, a factor of 2 probably wont change the results much.

[GOTSpectrum]

@RILEYISMYNAME This is still very much the case, QuickBit material?

[Gorgon]

Update for April 2020 ...

 

The current calculation shows 134 TFLOPS Single Precision:

CPU/GPU	GFLOPS (est)	Count	Threads	Clock (MHz)	IPC	GFLOPS (Calc)
2600		1	10	3400	16	544
2700		1	15	3350	16	804
2700		1	10	3250	16	520
2700x		1	14	3900	16	874
3900x		1	22	4055	32	2,855
9900k		1	14	4400	32	1,971
G5500		1	4	3800	32	486
1231v3		1	6	3400	32	653
1060	4375	3	1280	1950	2	14,976
1660ti	5437	1	1536	1950	2	5,990
1070ti	8186	1	2432	1950	2	9,485
2060	6451	1	1920	1950	2	7,488
2060S	7180	1	2176	1950	2	8,486
2070A	7465	1	2304	1950	2	8,986
2070S	9060	5	2560	1950	2	49,920
5700XT	9754	2	2560	1950	2	19,968
					Total	134,006
					CPU	8,707
					GPU	125,299

References:

https://en.wikipedia.org/wiki/FLOPS

https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units

https://en.wikipedia.org/wiki/List_of_AMD_graphics_processing_units