GPU FLOPS: Theory vs Reality


Advanced search

Message boards : Number crunching : GPU FLOPS: Theory vs Reality
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 7 · 8 · 9 · 10 · 11 · 12 · 13 . . . 17 · Next

AuthorMessage
AMDave
Volunteer tester

Send message
Joined: 9 Mar 01
Posts: 234
Credit: 11,671,730
RAC: 0
United States
Message 1811414 - Posted: 22 Aug 2016, 16:46:53 UTC - in response to Message 1811353.  

And still wondering if/when something like a 1050Ti will be built, and how it will compare to the 750?

ID: 1811414 · Report as offensive
Profile petri33
Volunteer tester

Send message
Joined: 6 Jun 02
Posts: 1668
Credit: 623,086,772
RAC: 156
Finland
Message 1811422 - Posted: 22 Aug 2016, 16:58:01 UTC - in response to Message 1811405.  

I should add that the GPU data is per card -- I can't tell which card did which task (part of why I have to exclude multi-gpu setups from my scans).

Your quad 1080 rig is more like 12871 cr/h in GPU throughput -- pretty amazing.


Thanks, I think it is one of a kind.

The cr/Wh is not too bad either. The 750 does well in that too.
To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones
ID: 1811422 · Report as offensive
Profile jason_gee
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 24 Nov 06
Posts: 7489
Credit: 91,093,184
RAC: 0
Australia
Message 1811428 - Posted: 22 Aug 2016, 17:01:56 UTC - in response to Message 1811364.  
Last modified: 22 Aug 2016, 17:07:23 UTC

Hi Shaggie,

Would it be possible to run a targeted scan of my (petri33) host and Mr Kevvys Linux hosts to get insight of Arecibo performance with special CUDA app. -=Vyper=- is running special too on his Linux.

All specials run one at a time.

My power Draw is between 120-170W, mostly 136W with Arecibo work load.


That is of interest since I wonder if productivity comes wit a cost.


I think that would be very useful. When I was running with the special code for the first Windows experiments, care of the 980 the temperature raised in the room noticeably (though single instance, compared to normally 2 instances underloaded with baseline apps).

A ways off: We'll probably need to make use of available data internally in new apps for reporting some summary data for development. NVML, which I'm looking into, can supply spot samples of Wattage (can take average and max), fans, number of compute instances running on the GPU (take max), and clocks/p-state. Operations are 'counted', and peak_flops known/calculable by device.
[nearly forgot about the misleading utilisation figure :) ]

That leads naturally to optional external tools for user control, profile driven, by raw throughput, compute/energy efficiency, temperature, or energy(Joules, W*t, or Wh). The last two being handy for those of us suffering from power costs or heatwaves.

New world :D
"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.
ID: 1811428 · Report as offensive
AMDave
Volunteer tester

Send message
Joined: 9 Mar 01
Posts: 234
Credit: 11,671,730
RAC: 0
United States
Message 1811568 - Posted: 22 Aug 2016, 23:01:49 UTC
Last modified: 22 Aug 2016, 23:10:37 UTC

For those of you who have a hankering for numbers (plus charts and graphs):

and

and for our German friends:


*EDIT*

ID: 1811568 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1811575 - Posted: 22 Aug 2016, 23:11:41 UTC - in response to Message 1811568.  
Last modified: 22 Aug 2016, 23:11:47 UTC

Now that would be something to see, lol
ID: 1811575 · Report as offensive
woohoo
Volunteer tester

Send message
Joined: 30 Oct 13
Posts: 972
Credit: 165,671,404
RAC: 5
United States
Message 1811585 - Posted: 22 Aug 2016, 23:24:30 UTC

that would be some serious power distribution at the motherboard level

but it would reduce the amount of excess power cables
ID: 1811585 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13736
Credit: 208,696,464
RAC: 304
Australia
Message 1811736 - Posted: 23 Aug 2016, 5:58:11 UTC - in response to Message 1811585.  

but it would reduce the amount of excess power cables

At least ones to the video cards.
It will just mean more cables connected to the motherboard.
Grant
Darwin NT
ID: 1811736 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1811747 - Posted: 23 Aug 2016, 6:23:16 UTC - in response to Message 1811736.  
Last modified: 23 Aug 2016, 6:23:29 UTC

but it would reduce the amount of excess power cables

At least ones to the video cards.
It will just mean more cables connected to the motherboard.



Yeah i thought about that too..

My ASUS X99e ws has 1-24 pin 2-8 pin and 1-6 pin. I thought that was a lot.

I can't begin to imagine how many pins or what combination you would need to power a board with 3-4 GPUs on it.
ID: 1811747 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13736
Credit: 208,696,464
RAC: 304
Australia
Message 1811751 - Posted: 23 Aug 2016, 6:29:25 UTC - in response to Message 1811747.  

I can't begin to imagine how many pins or what combination you would need to power a board with 3-4 GPUs on it.

Just a few...

If you look closely, there is a single 20-pin power connector like any other modern motherboard. There are also four 8-pin and two 6-pin power connectors. AMD marked the PCIe power connectors as follows:

P0 ABCD PWR
P1 ABCD PWR
P0 EFGH PWR
P1 EFGH PWR

These connectors are in addition to the processor 4+4 pin connector marked P0 and P1 CORE PWR.

A look at PCIe 4 at Tom's Hardware.
Grant
Darwin NT
ID: 1811751 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22202
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1811756 - Posted: 23 Aug 2016, 6:42:12 UTC

It will be interesting to see how they manage to get that amount of power (current) around the motherboard. It will need some serious power tracks! Maybe by using a dedicated "power plain" in the stack of plains?
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1811756 · Report as offensive
W-K 666 Project Donor
Volunteer tester

Send message
Joined: 18 May 99
Posts: 19062
Credit: 40,757,560
RAC: 67
United Kingdom
Message 1811764 - Posted: 23 Aug 2016, 7:17:58 UTC - in response to Message 1811756.  

It will be interesting to see how they manage to get that amount of power (current) around the motherboard. It will need some serious power tracks! Maybe by using a dedicated "power plain" in the stack of plains?

And if not very carefully designed can lead to pcb warping. Been there, done that, got the tee-shirt.
ID: 1811764 · Report as offensive
Al Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1682
Credit: 477,343,364
RAC: 482
United States
Message 1812178 - Posted: 24 Aug 2016, 13:22:16 UTC

From the comment discussion in the PCI Express 4.0 Brings 16 GT/s And At Least 300 Watts At The Slot link, they believe that the guy who wrote the article has got it wrong, due to it needing the power connectors still, but this time to the board instead of the card, the costs, the incompatibilities, for really no gain. The conclusion some had come to was that it was the the PCI-E Spec itself would officially allow that much power at the slot, if not more, because as someone had stated, the current spec tops out at 1 8 pin and one 6 pin, 2 8 pins are not in the spec yet. The first thing I thought when I read it was Cool! No more PSU cords to the cards, then I thought, Holy Cr*p, you get 4 cards pulling 300w+ thru the motherboard, that is a recipe for disaster.

ID: 1812178 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22202
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1812181 - Posted: 24 Aug 2016, 13:29:15 UTC - in response to Message 1811764.  

ONLY pcb warping? - you weren't trying hard enough....

I love the smell of burning FR4 in the morning :-0
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1812181 · Report as offensive
AMDave
Volunteer tester

Send message
Joined: 9 Mar 01
Posts: 234
Credit: 11,671,730
RAC: 0
United States
Message 1812190 - Posted: 24 Aug 2016, 13:53:31 UTC - in response to Message 1812178.  

... you get 4 cards pulling 300w+ thru the motherboard, that is a recipe for disaster.

Much ado about nothing.  The technology to mitigate heat buildup has existed for decades.  Mobos will simply need to evolve into a more advanced structure.  See here and here.
ID: 1812190 · Report as offensive
Al Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1682
Credit: 477,343,364
RAC: 482
United States
Message 1812346 - Posted: 25 Aug 2016, 2:10:52 UTC - in response to Message 1812190.  

lol yeah, sure, tell that to MB mfg's who are mostly trying to compete on cost, breaking away from a tried and true setup that has worked for years, decades actually, to go to space shuttle or other high tech, mostly unproven materials for a commodity product? With no actual tangible benefit to the end user other (that I can tell) than not having it have cables from the PSU to the add in cards? Good luck with that one. ;-)

ID: 1812346 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13736
Credit: 208,696,464
RAC: 304
Australia
Message 1812386 - Posted: 25 Aug 2016, 4:48:44 UTC - in response to Message 1812178.  
Last modified: 25 Aug 2016, 4:48:59 UTC

From the comment discussion in the PCI Express 4.0 Brings 16 GT/s And At Least 300 Watts At The Slot link, they believe that the guy who wrote the article has got it wrong

Yep, Tom's hardware have revised their article.
Basically, PCIe 4.0 max power capabilities: TBD
Update, 8/24/16, 2:06pm PT:PCI-SIG reached out to tell us that the power increase for PCI Express 4.0 will come from secondary connectors and not from the slot directly. They confirmed that we were initially told incorrect information.
Grant
Darwin NT
ID: 1812386 · Report as offensive
woohoo
Volunteer tester

Send message
Joined: 30 Oct 13
Posts: 972
Credit: 165,671,404
RAC: 5
United States
Message 1812387 - Posted: 25 Aug 2016, 4:50:38 UTC

Damn, no 1600w motherboards
ID: 1812387 · Report as offensive
Al Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1682
Credit: 477,343,364
RAC: 482
United States
Message 1812700 - Posted: 26 Aug 2016, 5:56:46 UTC

Such a shame... lol

ID: 1812700 · Report as offensive
AMDave
Volunteer tester

Send message
Joined: 9 Mar 01
Posts: 234
Credit: 11,671,730
RAC: 0
United States
Message 1812851 - Posted: 26 Aug 2016, 23:17:22 UTC - in response to Message 1811568.  
Last modified: 26 Aug 2016, 23:30:38 UTC

Someone read the tea leaves wrong:

ID: 1812851 · Report as offensive
Profile Shaggie76
Avatar

Send message
Joined: 9 Oct 09
Posts: 282
Credit: 271,858,118
RAC: 196
Canada
Message 1812955 - Posted: 27 Aug 2016, 13:01:50 UTC

Over coffee this morning I decided to format the performance data collected from my list ran in a different way. Assuming the Cobblestone metric is 200 per day per GFLOP and given the theoretical GFLOP throughput for all the GPUs (taken from Wikipedia) I get a picture like this:



I can only hope that this means that the modern GPUs are merely under-utilized.
ID: 1812955 · Report as offensive
Previous · 1 . . . 7 · 8 · 9 · 10 · 11 · 12 · 13 . . . 17 · Next

Message boards : Number crunching : GPU FLOPS: Theory vs Reality


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.