780ti not faster than a GTX 580

Message boards : Number crunching : 780ti not faster than a GTX 580
Message board moderation

To post messages, you must log in.

AuthorMessage
Miklos M.

Send message
Joined: 5 May 99
Posts: 955
Credit: 136,115,648
RAC: 73
Hungary
Message 1497950 - Posted: 31 Mar 2014, 20:20:19 UTC

I am wondering why my two 780ti cards are not faster than a GTX 580. Doing Distributed Rainbow units.
ID: 1497950 · 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: 22160
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1497954 - Posted: 31 Mar 2014, 20:32:34 UTC

I suggest you go and ask this question of the folks over Distributed Rainbow land's forums where you should be able to get an answer.
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1497954 · Report as offensive
Miklos M.

Send message
Joined: 5 May 99
Posts: 955
Credit: 136,115,648
RAC: 73
Hungary
Message 1497955 - Posted: 31 Mar 2014, 20:35:13 UTC - in response to Message 1497954.  

I understand that this is the wrong site for the question, but they are not there too often.
ID: 1497955 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1497959 - Posted: 31 Mar 2014, 20:45:54 UTC

It depends on what they are doing to the GPU. If the kind of processing they are doing is saturating another resource then a faster GPU won't help that kind of processing.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1497959 · Report as offensive
Miklos M.

Send message
Joined: 5 May 99
Posts: 955
Credit: 136,115,648
RAC: 73
Hungary
Message 1497965 - Posted: 31 Mar 2014, 20:53:36 UTC - in response to Message 1497959.  

That makes sense, although I see other people with the same computer GPU's running a wu in about 25 minutes or even less and mine took the first two to crunch in 85 minutes each. I keep asking here, because this I found to be the only site where one can get answers.
ID: 1497965 · Report as offensive
juan BFP Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 16 Mar 07
Posts: 9786
Credit: 572,710,851
RAC: 3,799
Panama
Message 1497984 - Posted: 31 Mar 2014, 22:12:12 UTC
Last modified: 31 Mar 2014, 22:16:22 UTC

Don´t know if that is you answer so if i´m wrong please forgive me.
Some time ago i made the same question, why the 780 not apears to be faster than a 770 (actualy an OC 680) or even the 580 as it should be.
Our "master guru" Jason give me the technical explanation, i try to find his post but i was unable, will try to explain in one word: "latency".
For some reason the 780 series (he including have one 780SC) loose a lot of time with the latencies of the driver and the GPU itself, don´t ask me why i simply have no knowledgue to explain. Maybe if he read this post he could give us all the explanation again.
AFAIK he is working on the problem and we expected that to be solved with the x42 builds, but that is not for now.
So it´s possible the problem you related on Distrigen could be related to that too. So it´s hard for somebody here help you with that. Maybe in Distrigen someone could have the answer, i don´t do Distrigen so i can´t help too.
My sugestion, when you crunch SETI crunch SETI only when do Dristigen do Distrigen only and see if something changes, yes i agree is a long shoot but when i try to run E@H and MB crunching at the same time on the my 780FTW the times increases exponentialy with no other explanation, so the problem is similar.
BTW I don´t have the Ti model only FTW´s but their processor are the same so in theory both must have the same latency limitation.
ID: 1497984 · Report as offensive
juan BFP Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 16 Mar 07
Posts: 9786
Credit: 572,710,851
RAC: 3,799
Panama
Message 1498014 - Posted: 1 Apr 2014, 0:00:11 UTC
Last modified: 1 Apr 2014, 0:00:58 UTC

I find a part of the Jason´s explanation...

This is exactly the phenomena that leads to most of the x42 changes. Pretty soon after receiving my own 780 SC I was exploring to find which parts of code these fast GPUs spend most of their time in, and why just one task doesn't push 100%.

It turns out that so much time is spent waiting for the CPU to respond, that it takes as much as 30-60% of the tasks. That is:
as the GPU core gets faster:
- Memory on the system stays roughly the same speed,
- So is the VRAM, +/-,
- PCIe stays roughly the same speed,
- Drivers get bigger/more complex.

So the first 3 don't explain it, but there are clues in the last. Measuring this backstage at CA we found that for fast GPUs we talk to the host CPU too often in too small messages, each of which incurs fixed waiting time.

It's that waiting time, totally system, chipset and drivers dependant, that causes the 780SC here to run at only 50% with one task.

Reducing and hiding these 'latencies' is one of the big parts of x42 phase 1
By changing to talk to the host less frequently, and in bigger lumps, a lot of that wastage does get alleviated, traffic through the drivers reduced, and utilisation goes up.

ID: 1498014 · Report as offensive
Miklos M.

Send message
Joined: 5 May 99
Posts: 955
Credit: 136,115,648
RAC: 73
Hungary
Message 1498025 - Posted: 1 Apr 2014, 0:28:22 UTC - in response to Message 1497984.  

I have the Distr running on GPU and Rosetta on CPU what do you think?
ID: 1498025 · Report as offensive
juan BFP Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 16 Mar 07
Posts: 9786
Credit: 572,710,851
RAC: 3,799
Panama
Message 1498027 - Posted: 1 Apr 2014, 0:32:59 UTC - in response to Message 1498025.  

Sorry i can´t help you with distr or rosetta, anything I could say about them is a total unfundamented guess. As Rob say you need to ask in ther forums since each program works diferent and what works on one could simply not work in the other.
ID: 1498027 · 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 1498028 - Posted: 1 Apr 2014, 0:33:38 UTC
Last modified: 1 Apr 2014, 0:34:35 UTC

Yeah that'll do, thanks Juan for finding that, as I'm a bit tied up at the moment.

From the OP's query, Even though I know nothing of this particular app, I can make some comments. If you see other similar hosts/cards achieving better times with similar work & application, the best is to first check every possible difference with available information, like:
- Different App version ?
- Different OS ?
- Overclock / watercooling on host you're comparing against ?
- Different CPU ? (and associated Overclocks etc too)

There are so many reasons a given application may perform on one system with the same GPU, and on another not. Working out where the key differences are is a start, and extends through these levels and back:

1) GPU-application code (was it made scalable to the new GPU architecture?)
2) GPU library runtimes (different version of Cuda/OpenCL?)
3) Driver, User mode Level. ( IS the device being shared ? is the driver beta, outdated, or other wise flakey)
4) OS Driver Level (as with #3, but add in if OS is changing, or misconfigured, such as power saving settings... e.g. Windows is changing also, to share nicer between applications. )
5) Other Hardware drivers. ( Are any devices in the system using outdated BIOS, Firmware [especially SSD] , Drivers... especially PCI Express with generic Microsoft drivers dated 2006 ? )

So there's quite a lot to narrow down before a specific problem/fault could be pointed at, but hopefully gives some ideas.
"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: 1498028 · Report as offensive
Miklos M.

Send message
Joined: 5 May 99
Posts: 955
Credit: 136,115,648
RAC: 73
Hungary
Message 1498029 - Posted: 1 Apr 2014, 0:41:33 UTC - in response to Message 1498027.  

I have shut down Rosetta and see what happens.
ID: 1498029 · Report as offensive
Miklos M.

Send message
Joined: 5 May 99
Posts: 955
Credit: 136,115,648
RAC: 73
Hungary
Message 1498038 - Posted: 1 Apr 2014, 1:39:09 UTC

seems to make no difference with Rosetta stopped.
ID: 1498038 · Report as offensive
Miklos M.

Send message
Joined: 5 May 99
Posts: 955
Credit: 136,115,648
RAC: 73
Hungary
Message 1499242 - Posted: 3 Apr 2014, 18:15:38 UTC

I think I my have found out something by trying a few other projects. For PrimeGrid and Collatz my GTX 780ti cards are on par with other fast performers with the same cards. But for DistrRain no one seems to use GTX 780 ti cards. Like a wake up call for me. Perhaps they tried it, but did not like how it performed vs. the GTX 580(not much better). But my new cards sure like to run Boinc and they run Collatz and PrimeGrid also a lot faster than the GTX 580.
Mora of the story: do more research before complaining. I am now running one wu of GPUGrid to test it there too, comparing it to other results with the same cards.
Also, thank you all for your kind patience and help. I know I can always count on the Seti boards for help!
ID: 1499242 · Report as offensive

Message boards : Number crunching : 780ti not faster than a GTX 580


 
©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.