Message boards :
Number crunching :
Nvidia Titan X
Message board moderation
Previous · 1 · 2 · 3 · Next
Author | Message |
---|---|
Akio Send message Joined: 18 May 11 Posts: 375 Credit: 32,129,242 RAC: 0 |
But how does it compare to the 780Ti which it is replacing? Bonus: Just did a major upgrade to my rig by installing 2 NVIDIA GeForce GTX 780 Ti GPU's at a not-too-shabby price, considering the single GPU card they replaced and how long the 780 Ti's have been available ;) Guess I got them just in the nick of time if they came out with a replacement GPU; I was never able to get a second GTX 460 mere months after I originally built the rig as it got discontinued. That's the way the technology wagon rolls though, hehe! That Titan card looks pretty impressive, but it'll be another 4 or so years before I am able to do any more upgrades to the rig. More than likely the next go round will require a new build to accommodate a better CPU and MB...And that's assuming the Old Lady approves such an ambitious spending endeavor. It took many months just to be allowed to spend hundreds on GPU's for this upgrade, hehe. Love her though! :) |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
Unfortunately since the 980Ti has yet to be released they haven't been able to compare it to the 780Ti. However if you want to compare the plain 980 vs 780Ti you will see it is faster. http://gpuboss.com/gpus/GeForce-GTX-980-vs-GeForce-GTX-780-Ti#differences But the 780TI has 840 more cuda cores. Power is the other biggy as 980 uses less. We will have to wait and see what the spec of the 980 Ti are when it comes out |
Akio Send message Joined: 18 May 11 Posts: 375 Credit: 32,129,242 RAC: 0 |
Yeah, I'm sure the 980Ti's will be considerably better than the 780Ti's that I bought and installed :) But hey, you have to admit that in regards to my upgrade that it was well worth it, hehe. There will be many future releases that will leave me drooling with "I wants!", and I'm sure the 980Ti will be one of those such releases, hehe. :) |
woohoo Send message Joined: 30 Oct 13 Posts: 972 Credit: 165,671,404 RAC: 5 |
I'm not seeing where the 980 would be faster than the titan x. |
mramakers Send message Joined: 20 Jul 04 Posts: 42 Credit: 3,694,335 RAC: 0 |
I'm not seeing where the 980 would be faster than the titan x. Me neither. GTX980 does 4612 GFLOPS Single precision Titan X does 6144 GFLOPS. My guess is that the 980Ti will not be that much slower compared to the titan. Therefore it will be faster than a 980. At least for crunching and that is what I'm interrested in. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
http://www.newegg.com/Product/Productcompare.aspx?Submit=ENE&N=-1&IsNodeId=1&Description=gtx%20980&bop=And&CompareItemList=-1%7C14-487-129%5E14-487-129-TS%2C14-487-079%5E14-487-079-TS&percm=14-487-079%3A%24%24%24%24%24%24%24 This is comparing core clock speed. The article mentioned above http://www.techspot.com/news/60183-gm200-based-geforce-gtx-980-ti-reportedly-coming.html is wrong according to http://www.tweaktown.com/news/44243/nvidia-preparing-geforce-gtx-980-ti-6gb-vram-faster-titan/index.html which goes along with what I was thinking. Why would you make the 980 Ti slower than the titan X or the 980. Both 980 and 980Ti will be faster than the titan X. The 980 Ti will gain 2 GB of Ram on the card approximately 1000 Cuda cores compared to the GTX 980 and be 10% faster than the Titan X But you have to wait on it |
mramakers Send message Joined: 20 Jul 04 Posts: 42 Credit: 3,694,335 RAC: 0 |
I'm confused. You're saying that NV makes a card that is slower than a GTX980, but for double the price? That's like shooting yourself in the foot. A GTX 980 would have to run at 1500MHz to be equal to a Titan X in single precision performance. 2048 x 2 x 1500 = 6144 GFLOPS 3072 x 2 x 1000 = 6144 GFLOPS |
woohoo Send message Joined: 30 Oct 13 Posts: 972 Credit: 165,671,404 RAC: 5 |
comparing core clocks doesn't work right when the underlying gpu core is different. like a 780 Ti is only about 1% faster on the core clock than a 780 but 25% faster overall because of the different gpu core. the titan black has a similar gpu core to the 780 Ti but more core clock so it's slightly faster than the 780 Ti at single precision. so yes the 980 has more core clock than the titan x but the gpu core is different. if the 980 Ti ends up using similar gpu core to the titan x then it won't need to be clocked that high to outrun a 980. |
mramakers Send message Joined: 20 Jul 04 Posts: 42 Credit: 3,694,335 RAC: 0 |
You meant slower in MHz when you said: "Interesting, but according to that article it would be slower than the Titan X which is slower than the current GTX 980s" That's where I started to get confused, because I was talking about GFLOPS I wasn't comparing clock speeds, you were. What I was refering to was the amount of cuda cores in relation with clock speed to calculate sp performance. That is the same for different gpu cores like 980 or titan x. |
woohoo Send message Joined: 30 Oct 13 Posts: 972 Credit: 165,671,404 RAC: 5 |
So one could guess 940 MHz for a 980 Ti |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
I'm quite keen, later today, to see if my 980 is liberated a bit under Win10/DirectX12. Current application architectures rely on a lot of mechanisms that mean we get generally somewhere around the 5% mark of those predicted peak GFLops, and a lot of that scaling limitation is related to driver architecture/latencies, as well as the applications which have to scale from the smallest G80 to the largest Maxwell. A good way to deal with that is with running multiple instances (as most of us do) One thing you'll probably find is even if lower frequency is specified as stock/base, the higher dissipation ( 250W TDP ) implies more cooling and power, which in turn implies higher headroom. I think being able to use all of a TiTan-X to its full potential is a way off, though I suspect even with current OS and applications, It'll reach a higher number of instances before diminishing returns, even though that will to some extent depend on the system feeding it. "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. |
kittyman Send message Joined: 9 Jul 00 Posts: 51468 Credit: 1,018,363,574 RAC: 1,004 |
The newest thingys the kitties are sporting are 680s. It would take a helluva winning lottery ticket to upgrade the dozen or so power hungry 580s crunching here.... Meowsigh....maybe some day. A fella on GPUGRID tried to run the numbers to convince me that the payback on converting them was less than a year with power savings. My numbers did not sort as well as his did...LOL. "Freedom is just Chaos, with better lighting." Alan Dean Foster |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Yeah, based on its low power usage & great performance at cool temps, I don't think my 680 will be retiring for a long time to come, and will likely end up in a small formfactor network storage / media device this year. The 480 monster was easier to retire early based on power, heat and noise, and is sitting somewhere under a pile of junk. It did some great work fermi-ising X-branch, so putting it to pasture made a lot of sense. The 780 and 980 are being juggled around exploring Win7 versus Win10 tech preview this weekend. I'd have to say the 980 was the perfect choice for my development needs right now, machine hardware and power/heat/noise budget, and needing to get on top of the massive changes. "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. |
kittyman Send message Joined: 9 Jul 00 Posts: 51468 Credit: 1,018,363,574 RAC: 1,004 |
Yeah, based on its low power usage & great performance at cool temps, I don't think my 680 will be retiring for a long time to come, and will likely end up in a small formfactor network storage / media device this year. The 480 monster was easier to retire early based on power, heat and noise, and is sitting somewhere under a pile of junk. It did some great work fermi-ising X-branch, so putting it to pasture made a lot of sense. Yeah, the 680 is not a beast, but it is a calm, cool, and collected cruncher. My top rig has 3 680s and a 570 that I should retire and slot another 680 in it's place. "Freedom is just Chaos, with better lighting." Alan Dean Foster |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13746 Credit: 208,696,464 RAC: 304 |
The 780 and 980 are being juggled around exploring Win7 versus Win10 tech preview this weekend. Will be interesting to see how the 980 does on Win10. Given how greatly shorty crunch times increase on Maxwell cards compared to previous hardware I'm hopeful that DX12 will improve shorty crunch times significantly on Maxwell hardware. Grant Darwin NT |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13746 Credit: 208,696,464 RAC: 304 |
Yeah, the 680 is not a beast, but it is a calm, cool, and collected cruncher. For those that are interested, from NVidia's site, GTX 570, 219W GTX 680, 200W GTX 750Ti, 60W GTX 960, 120W GTX 980, 165W GTX 780Ti, 250W Titan X, 250W Grant Darwin NT |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
The 780 and 980 are being juggled around exploring Win7 versus Win10 tech preview this weekend. Yes, my pre-testing/verification guess is that being DirectX12//WDDM2 capable, that DirectX11+Win7-8-8.1 could be some kindof emulation going on. That would fit with the virtualisation/synchronisation/latency costs that would bloat in VHAR shorties, but we shall see. Could just be time to scrap the whole codebase and start from scratch :-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. |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
FWIW at stock speeds I think the 680 tech spec of 200W is rather higher than actual, though that bodes well for cooling and power implementation overkill, and OC headroom. "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. |
woohoo Send message Joined: 30 Oct 13 Posts: 972 Credit: 165,671,404 RAC: 5 |
My poor math says it would cost around $400 a month to power a dozen 580s. A gtx 960 is faster so you might only need eight of them to produce the same output for a total street price of over $1700. It might cost $125 a month to run those eight newer cards so the break even point might be about seven months. |
kittyman Send message Joined: 9 Jul 00 Posts: 51468 Credit: 1,018,363,574 RAC: 1,004 |
Yeah, the 680 is not a beast, but it is a calm, cool, and collected cruncher. So... I am not into maths right now. 750tis can be had for about $140.00 each. Drop a dozen 580s and swap in the 750tis. $1,680.00 to acquire. Maybe 140 watts per card less than a 580. X12 = 1,680 watts x 24 hours = 40,320 watts a day less. Here's where my bad math sometimes gets me confused, but.... 40.32 for 1 day x .14/kwh = $5.6448 per day power cost savings. $1,680 divided by 5.6448 = 297.61904......... Maybe, if the math is correct, they COULD pay for themselves in less than a year. Not sure how crunchability of a 750ti compares to a 580 though...might take a hit there. It does take a hit because my numbers are based on a .14/kwh prime day rate. It gets more complicated because my cost is .14 during business days 10am to 10pm. On holidays, nights, and weekends, my cost is only about .05/kwh. So the payback gets extended a great deal with that in mind. Hmmmmm......... Meownowthinking. "Freedom is just Chaos, with better lighting." Alan Dean Foster |
©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.