Message boards :
Number crunching :
GPU FLOPS: Theory vs Reality
Message board moderation
Previous · 1 . . . 13 · 14 · 15 · 16 · 17 · Next
Author | Message |
---|---|
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
I already seen Petri classified "For sale 4 - NVIDIA 1080 Founders Edition cards for sale lightly used in home gaming ..."(EDIT:) Reason for selling. I won the game!" |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
lel, my wallet won't allow that quite yet, but certainly will be something on the agenda down the road. I'm waiting to see if AMD / motherboard partners manage to resolve the IOMMU/virtualisation issues first. If they do, then some plans will need rejigging, and gumtree might start getting some of my old hardware. "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. |
petri33 Send message Joined: 6 Jun 02 Posts: 1668 Credit: 623,086,772 RAC: 156 |
I already seen Petri classified "For sale 4 - NVIDIA 1080 Founders Edition cards for sale lightly used in home gaming ..."(EDIT:) Reason for selling. I won the game!" LOL :) However, with the special app the scaling seems quite nice. You can comb through my results to find 1080Ti work. 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 |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13766 Credit: 208,696,464 RAC: 304 |
I like where the 1050Ti slots right underneath the 1070 for efficiency. I'm still impressed by the GTX 750Ti. 3 year old architecture and still in the top 10 for efficiency, and not all that far from the leaders either. Grant Darwin NT |
Shaggie76 Send message Joined: 9 Oct 09 Posts: 282 Credit: 271,858,118 RAC: 196 |
I've been mulling over the system I've been having problems with and considering options for changing up the hardware. It really bugs me that the RX480 is so far from theoretical throughput. Wikipedia says the RX480 claims to do 5834 GFLOPS in boost mode which is about the same as a 980 Ti's 5632 GFLOPS in theory but in practice my stats put the median credit/hour for the RX480 at only 730 compared to the 980 Ti's 1207 -- that's a 65% difference for when in theory the RX480 should be a tiny bit faster! The memory bandwidth for the RX480 is rated at 256 GB/sec while the 980 Ti is rated as 336 GB/s -- this is only a 33% difference and while it might explain some that margin it does leave me wondering why there's such a gulf between the cards. I can't help but wonder how much of that gap is software. So what about the RX470? The ratio of CPU to bandwidth is a bit better: 4940 GFLOPS to 211 GB/sec; let's compare it to, say, a GTX 1060 (6GB) which is about 4372 GFLOPS at 192 GB/sec -- it should be faster in both departments but the median credit/hour for the RX470 is 627 which is still considerably lower than the GTX 1060's 848 credit/hour. The other factor that I wondered about was the relative cost of the CPU-side -- the AMD implementation seems to be better yielding the code. Curiously on my i7-5960X this doesn't amount to much -- about 45 credit/hour/core so even if you add 45 or 90 to the RX480's score it's still painfully far behind. I looked at most of the RX500 series speculation and the only part expected to have more bandwidth looks the RX Vega -- most just have their cores clocked a bit faster which might not help much if there isn't bandwidth to feed them (and generally erodes the performance/watt). So back to my dilemma: the dual 980 Ti's just aren't stable for me even with only one task per card; I've had problems with dual 1070's so my options are get a single 1080 Ti, wait for Vega stuff that is likely to be slower, or cram in four RX470 or RX480's. Frankly I'm reluctant to try another SLI configuration so I'm leaning towards just a single 1080 Ti and hope it doesn't mind 2 or maybe even 3 tasks concurrently. |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13766 Credit: 208,696,464 RAC: 304 |
Frankly I'm reluctant to try another SLI configuration so I'm leaning towards just a single 1080 Ti and hope it doesn't mind 2 or maybe even 3 tasks concurrently. Did you try it without SLI? SLI is only for gaming, Grant Darwin NT |
Shaggie76 Send message Joined: 9 Oct 09 Posts: 282 Credit: 271,858,118 RAC: 196 |
Frankly I'm reluctant to try another SLI configuration so I'm leaning towards just a single 1080 Ti and hope it doesn't mind 2 or maybe even 3 tasks concurrently. Of course. I've tied both with and without a bridge -- I'm pretty sure that even if you have them configured for SLI OpenCL shows them separately anyway. |
Shaggie76 Send message Joined: 9 Oct 09 Posts: 282 Credit: 271,858,118 RAC: 196 |
One other factor that's been bugging me is during the weekly downtime my 980 Ti's often run out of tasks -- I can only imagine a faster card would run dry even sooner. Based on my last few scans the average run-time 980 Ti tasks is somewhere between 290 seconds and 315 seconds depending on how many VLARs there are -- let's call it 5 minutes a task, 100 tasks is just over 8 hours of work; suppose a 1080 Ti is only 50% faster (which I would say is conservative) it'll take one to burn down the 100 task limit in 5 and a half hours or so. I only crunch when electricity is cheap so ideally I crunch 108 hours a week; luckily the weekly 24-36 hour outage only tends to interrupt one 12-hour night-shift for me -- so right now my 980 Ti's are idle only 12-8.5 ~ 3.5 hours / 108 hours a week = so 3% of the time. Assuming a 1080 Ti would run dry in 5.5hrs that would mean mine would spend 6% of its week idle. Suppose it matches the paper-performance and is actually twice as fast as my 980 Tis; it would run dry in 4.2 hours, so over 7% idle per week! |
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
Petri is currently at a 130 minute cache with his setup. So yea it runs out quick with fast cards. EDIT: Your 980Ti's are at 9h50m cache right now ... if you have been running continuously the last 10 hours. How many Watts does your 980 Ti pull at full load? I've been trying to figure that out since I'm seriously considering one or two. The best I can figure it should be about 180W. |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13766 Credit: 208,696,464 RAC: 304 |
How many Watts does your 980 Ti pull at full load? I've been trying to figure that out since I'm seriously considering one or two. The best I can figure it should be about 180W. If it were fully loaded, 250W. But present applications appear to be able to only get up to around 65-70% of maximum power load. Petri's specials i'd expect to be higher still, but still probably not yet knocking on 100%. EDIT- Looking at the Credit/Wh the GTX 980Tis aren't that good a performer. What you save in purchase price will be offset by running costs. Grant Darwin NT |
Shaggie76 Send message Joined: 9 Oct 09 Posts: 282 Credit: 271,858,118 RAC: 196 |
How many Watts does your 980 Ti pull at full load? I've been trying to figure that out since I'm seriously considering one or two. The best I can figure it should be about 180W. It can keep them at 250W each if I double up tasks -- since I've been trying single-tasked to try to find stability. With one task the still hit 250 occasionally but depending on the task they might drop to 180 or so -- it's wildly variable. |
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
That is good to know, I just about bought a pair of 980Ti Hybrids last night. My 980 peeks at 140, your Ti at 250, Petri sees ~150 vs 225 for the 1080Ti My search was in vain for a 980 Ti running CUDA apps, but I am guessing that they should get close to 90% of a 1080. The power use just doesn't seem worth it for the Ti's. I'm a little disappointed with my 1070, it's 10-15W less than my 980, but does 10% less work. Kind of makes the 980 look good. |
LooneyTunes Send message Joined: 12 Dec 08 Posts: 51 Credit: 16,807,684 RAC: 0 |
I am seeing about a 1minute difference in GPU processing time between the 1050Ti and the GTX960 with the 1050 being the slighter faster of the two cards. When it comes down to it, dollar for dollar the 1050Ti is a great choice especially should some tuning be done. Especially when you look at the difference in power requirements of the two cards. The reason I went for the Ryzen system was power consumption and cost per core/thread. Couple that with a cheap 1050Ti and give it a tweak speed wise and all in all it becomes quite a potent little package for crunching. "I know I am insignificant...Just look how many stars their are...!" |
petri33 Send message Joined: 6 Jun 02 Posts: 1668 Credit: 623,086,772 RAC: 156 |
That is good to know, I just about bought a pair of 980Ti Hybrids last night. If I did not see wrong your 1070 does a guppi in 200 secs and your 980 does the same ar guppi in 240 secs, I'll try to find the tasks here. So an edit is on the way in a moment. EDIT #1: https://setiathome.berkeley.edu/workunit.php?wuid=2527379505 That is your 1070 and it beats nicely a GTX Titan Pascal too. EDIT #2: https://setiathome.berkeley.edu/workunit.php?wuid=2527559249 That is your 980 and it loses to your 1070 by 15% and beats nicely a 980Ti on another platform. EDIT #3: Your 1080 does a 0.01 guppi in 161 seconds. The 1070 takes 200 seconds and your 980 takes 233 seconds. They All Do Nice! Petri 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 |
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
Yea, my 1070 did pick up the pace a bit with rearranging and app change. My original post was with the 1070(15cu)/980(16cu) in one box with x41p_zi+ with -unroll 15. As you seen I changed my mind to 1080 Hybrids, Yes they perform well. So now I have x41p_zi3t1f -autotune: 1080/1080/980 (all Hybrids) 1070/750Ti (air cooled) 750Ti I'm thinking of swapping the 1070 and 980 since the 2 air cooled cards are a tad noisy together. But I'm waiting for my 100W PSU to arrive, my 750W is running at 650W, and the 1070/750 is only 550W. Also time for some clocking :) Thanks for looking at the times, it saves me the work :D |
Carlos Send message Joined: 9 Jun 99 Posts: 29964 Credit: 57,275,487 RAC: 157 |
This thread seem to come close to answering a question I have. I am currently running 3-750Ti's and feel the need for more speed. Toying with the idea of getting a 1080ti ($700) but I am seeing that the 1050ti at about $140 a card might be a better option. Assume that I will put $700 in to upgrading. What would give me the best credit per hour? I can only fit 3 cards into my main cruncher. Suggestions? |
Wiggo Send message Joined: 24 Jan 00 Posts: 35109 Credit: 261,360,520 RAC: 489 |
This thread seem to come close to answering a question I have. I am currently running 3-750Ti's and feel the need for more speed. Toying with the idea of getting a 1080ti ($700) but I am seeing that the 1050ti at about $140 a card might be a better option. Assume that I will put $700 in to upgrading. What would give me the best credit per hour? I can only fit 3 cards into my main cruncher. Suggestions? I'd suggest 3x 3GB 1060's, but I'm probably a bit biased there (just be sure to get proper 2 slot design jobs as quite a few around now are 2.1-2.3 designs are won't fit together in 2 slot spacings). ;-) Cheers. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
I was going to tell him to go with the 6GB 1060 but I don't own either so take that for what it's worth, lol... |
Wiggo Send message Joined: 24 Jan 00 Posts: 35109 Credit: 261,360,520 RAC: 489 |
I was going to tell him to go with the 6GB 1060 but I don't own either so take that for what it's worth, lol... I've spoken with several people here who also run 1060's about the memory size difference and unless you're gaming with them you'll see no difference with them here (it maybe different with other projects though), in fact most of those with the 6GB versions now wish that they'd with the cheaper option. ;-) Cheers. |
Carlos Send message Joined: 9 Jun 99 Posts: 29964 Credit: 57,275,487 RAC: 157 |
Three of these? Looks like you are getting about 22k per card. So three would give me 66k? |
©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.