GPU FLOPS: Theory vs Reality

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

To post messages, you must log in.

Previous · 1 . . . 16 · 17 · 18 · 19 · 20 · Next

AuthorMessage
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5126
Credit: 276,046,078
RAC: 462
Message 2023615 - Posted: 18 Dec 2019, 22:31:52 UTC - in response to Message 2023613.  

Someone who was very disappointed with a gx 1660 Super offered it for $110 buy-it-now on eBay..... it was Friday a payday. Crunch....

Tom


It wasn't a Gtx 1660 Super. Just a Gtx 1660 but I just confirmed it does boot. After I take a look at Shaggie's list I will cancel my request for a refund (probably).

Tom


Just looked at the performance data for a regular gtx 1660 and it's not even close in either effeciency or total performance to a gtx 1660 Ti so probably not close to the Super either.

Tom
A proud member of the OFA (Old Farts Association).
ID: 2023615 · Report as offensive
Ian&Steve C.
Avatar

Send message
Joined: 28 Sep 99
Posts: 4267
Credit: 1,282,604,591
RAC: 6,640
United States
Message 2023618 - Posted: 18 Dec 2019, 22:47:50 UTC - in response to Message 2023615.  

1. 110 is a great price for a 1660, non-super.

2. Make sure the performance numbers you’re looking at are in compute scenarios and not gaming loads. The only difference between and 1660 and a 1660 super is the memory being GDDR6 instead of GDDR5.
Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

ID: 2023618 · Report as offensive
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5126
Credit: 276,046,078
RAC: 462
Message 2023623 - Posted: 18 Dec 2019, 23:29:02 UTC - in response to Message 2023618.  

1. 110 is a great price for a 1660, non-super.

2. Make sure the performance numbers you’re looking at are in compute scenarios and not gaming loads. The only difference between and 1660 and a 1660 super is the memory being GDDR6 instead of GDDR5.


Doesn't seem to be listed that high on Shaggie's list. That was the only place I was looking. Since it came up and I have it, I guess I will keep it for now. The numbers I saw on Shaggie's list seemed to show it was less efficient and less powerful than the Gtx 1070's I have. If I am reading that list right it is likely that a Gtx 1660 Super out performs the Gtx 1070 in both production and in efficiency.

If that is true.... I want MORE :)

Tom
A proud member of the OFA (Old Farts Association).
ID: 2023623 · Report as offensive
Ian&Steve C.
Avatar

Send message
Joined: 28 Sep 99
Posts: 4267
Credit: 1,282,604,591
RAC: 6,640
United States
Message 2023633 - Posted: 19 Dec 2019, 0:13:01 UTC - in response to Message 2023623.  

There’s no way that it’s less efficient than the GTX 10-series cards.

Run it. Then compare the runtimes against Juan’s 1070s or anyone else running the special app.
Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

ID: 2023633 · Report as offensive
Profile tazzduke
Volunteer tester

Send message
Joined: 15 Sep 07
Posts: 190
Credit: 28,269,068
RAC: 5
Australia
Message 2023690 - Posted: 19 Dec 2019, 9:09:00 UTC

Greetings Tom M

How is your GTX 1660 travelling, tried to have a look at your statistics, but it seems you have it the PC with multi GPU setup.

Hoping to see some feedback/results sometime soon, may help me source some new GPU's that could easily replace a GTX 1060 and GTX 970.

Happy Crunching.

Regards
Mark
ID: 2023690 · Report as offensive
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5126
Credit: 276,046,078
RAC: 462
Message 2023699 - Posted: 19 Dec 2019, 12:08:40 UTC - in response to Message 2023690.  

Greetings Tom M

How is your GTX 1660 travelling, tried to have a look at your statistics, but it seems you have it the PC with multi GPU setup.

Hoping to see some feedback/results sometime soon, may help me source some new GPU's that could easily replace a GTX 1060 and GTX 970.

Happy Crunching.

Regards
Mark


Mark,
The gpu task software I run will give you a discreet listing of all the gpu's I have and which one processed that task. Since I have only just installed the Gtx 1660 (not-Super) I don't have much evidence one way or the other.

I have run my gtx 1660 Super as a single gpu for a couple of hours last Saturday? And it continues to perform at least as well as a gtx 1070 while using less power (I think).

The Super hasn't made it onto the last published report by Shaggie but the Gtx 1660 Ti was the leader in efficiency. If you see a GPU brand you like at or near $230 that is currently the best price I have seen.

My experience is with Linux and Tbar's All-In-One. Shaggy's graph pulls production data and skips non-prodcution tasks like the AIO.

Tom
A proud member of the OFA (Old Farts Association).
ID: 2023699 · Report as offensive
Profile StFreddy
Avatar

Send message
Joined: 4 Feb 01
Posts: 35
Credit: 14,080,356
RAC: 26
Hungary
Message 2023732 - Posted: 19 Dec 2019, 17:55:09 UTC - in response to Message 2023699.  

Tom, if you want to have an idea how a gtx1660 (non ti and non super) performs, check my host: ID: 8838390
This PC uses the special sauce. The CPU is mostly crunching WCG, occasionally SETI. The GPU only crunches SETI 7/24, RAC is around 50k. The GPU alone consumes 90 Watts while crunching SETI. It is pretty efficient.
ID: 2023732 · Report as offensive
Profile tazzduke
Volunteer tester

Send message
Joined: 15 Sep 07
Posts: 190
Credit: 28,269,068
RAC: 5
Australia
Message 2023769 - Posted: 19 Dec 2019, 21:02:09 UTC

Greetings Tom M and StFreddy

Thankyou both for your input, it is definitely food for thought.

Happy crunching.

Regards
ID: 2023769 · Report as offensive
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5126
Credit: 276,046,078
RAC: 462
Message 2023783 - Posted: 19 Dec 2019, 23:32:50 UTC - in response to Message 2023732.  

Tom, if you want to have an idea how a gtx1660 (non ti and non super) performs, check my host: ID: 8838390
This PC uses the special sauce. The CPU is mostly crunching WCG, occasionally SETI. The GPU only crunches SETI 7/24, RAC is around 50k. The GPU alone consumes 90 Watts while crunching SETI. It is pretty efficient.


Thank you. I looked and the bulk of your first page gpu tasks are running at or under 90 odd seconds with a few outliers. That leads me to suspect that even the Gtx 1660 is running as fast or faster than my Gtx 1070's most of the time.

Maybe for New Years I can afford another gtx 1660 (some version) to replace my last gtx 1060 3Gb on that box. Then I can start swapping out gtx 1070's......

Yes, its the "upgrade grind" once again ladies and gentleman.

Tom
A proud member of the OFA (Old Farts Association).
ID: 2023783 · Report as offensive
wujj123456

Send message
Joined: 5 Sep 04
Posts: 40
Credit: 20,877,975
RAC: 219
China
Message 2024964 - Posted: 25 Dec 2019, 19:14:32 UTC
Last modified: 25 Dec 2019, 19:20:05 UTC

I have two 1660 super and one 1080 for my computer: https://setiathome.berkeley.edu/results.php?hostid=8846357. Unfortunately I am running the mutex version so the timing number can't be directly used from website. From the timing I had before switching to mutex version is that my 1080 is just ~10% faster.
If one is patient enough, there are still pending validation results of same true angle range like these ones from early Dec:
https://setiathome.berkeley.edu/result.php?resultid=8313712992
https://setiathome.berkeley.edu/result.php?resultid=8313712563
https://setiathome.berkeley.edu/result.php?resultid=8312950821

However, another interesting observation from "nvidia-smi dmon" is that my 1080 isn't consuming any more power than the 1660 supers. The three cards (top to bottom) are 1080 FE (x16), 1660 super dual-fan (x4), 1660 super single-fan (x4). They consume around 100-110W, 110-120W, 100-110W on average with similar sm utilization percentage. Though I am pretty sure factory OC is going to make the 1660 super less power efficient, I don't know by how much here and they did stay within TDP. I reserved two cores for the 3 cards and most of time one core worth of compute is idling and thus I don't believe I've starved any GPUs off resources. If we go by TDP, 1080 would seem to be less efficient, but at least on my host, the actual power consumption puts 1080 ahead in power efficiency.

I wonder what power readings people get on their high-end cards with the special sauce. Are they consuming as high as their TDP rating suggested? I am curious if it's something odd with my setup or it's consistent with what others see as well. There is no easy way to collect actual power consumption at scale with the data we have, so the charts likely won't arbitrarily change. However, I might buy higher-end cards in the future if they are actually more power efficient. These 1660 supers are bought for boinc doubling as small heaters. I don't really have other uses for them, so might as well go with the most power efficient card.
ID: 2024964 · Report as offensive
Ian&Steve C.
Avatar

Send message
Joined: 28 Sep 99
Posts: 4267
Credit: 1,282,604,591
RAC: 6,640
United States
Message 2024966 - Posted: 25 Dec 2019, 19:18:54 UTC - in response to Message 2024964.  
Last modified: 25 Dec 2019, 19:22:56 UTC

if you are running -nobs and not overcomiting the CPU, you can use the CPU time as the run time, it will be close enough.

to get maximum performance you should set the -nobs command line argument to bring the GPU utilization up to 95+%. I noticed when running without nobs, the work bounces around on different cores and that makes the GPU utilization jump around and not work as hard as it could. when you run -nobs, GPU utilization pegs and it will use as much power as you let it. all of my cards run rear max TDP limit with the exception of really low end cards like 750ti or 1650 which can't even pull up to their limit when trying their hardest lol.
Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

ID: 2024966 · Report as offensive
wujj123456

Send message
Joined: 5 Sep 04
Posts: 40
Credit: 20,877,975
RAC: 219
China
Message 2024969 - Posted: 25 Dec 2019, 19:24:16 UTC - in response to Message 2024966.  

if you are running -nobs and not overcomiting the CPU, you can use the CPU time as the run time, it will be close enough.

Unfortunately I am not using -nobs. I don't see much improvement when I tried it on 0.98b. I decided to save the cores to run other projects/stuff.
ID: 2024969 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2024971 - Posted: 25 Dec 2019, 19:34:22 UTC - in response to Message 2024969.  

If you don't want to use -nobs, you could also use an affinity lock instead. That keeps the cpu thread feeding the gpu task locked onto that card and task and keeps the load from jumping around.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2024971 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2024973 - Posted: 25 Dec 2019, 19:44:35 UTC

You know it would have been really interesting if Shaggie had run one of his server scrapes during the weekend when a lot of hosts would have been reporting as stock and using the stock OpenCL application when in reality they would have been running the special sauce app disguised as the stock app. Might have shown a very different view of the top cards.

Actually glad that didn't happen what with all the issues the servers were having though.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2024973 · Report as offensive
Ian&Steve C.
Avatar

Send message
Joined: 28 Sep 99
Posts: 4267
Credit: 1,282,604,591
RAC: 6,640
United States
Message 2024974 - Posted: 25 Dec 2019, 19:46:00 UTC - in response to Message 2024973.  

Yeah I was thinking that all the people running the special app under the guise of the stock app would really skew the scan with such huge outliers lol.
Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

ID: 2024974 · Report as offensive
Profile tazzduke
Volunteer tester

Send message
Joined: 15 Sep 07
Posts: 190
Credit: 28,269,068
RAC: 5
Australia
Message 2025050 - Posted: 26 Dec 2019, 9:19:52 UTC

Well Santa dropped of an MSI GTX 1660 Super for Christmas, so have that now in the box replacing a GTX 970 ITX model.

Time to let it have a run and see how we go.

Cheers
ID: 2025050 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13855
Credit: 208,696,464
RAC: 304
Australia
Message 2025051 - Posted: 26 Dec 2019, 9:33:27 UTC - in response to Message 2025050.  
Last modified: 26 Dec 2019, 9:35:08 UTC

Well Santa dropped of an MSI GTX 1660 Super for Christmas, so have that now in the box replacing a GTX 970 ITX model.

Time to let it have a run and see how we go.
Looks like it's just ahead of a GTX 1070, close to par with a 1660Ti.
Grant
Darwin NT
ID: 2025051 · Report as offensive
Profile zoom3+1=4
Volunteer tester
Avatar

Send message
Joined: 30 Nov 03
Posts: 66355
Credit: 55,293,173
RAC: 49
United States
Message 2025471 - Posted: 29 Dec 2019, 16:46:53 UTC
Last modified: 29 Dec 2019, 16:47:47 UTC

1660Ti = 1536 cores.
1660 S = 1408
In case anyone was wondering. Asus has not made a blower or turbo in a 1660, the strix got to a 1660Ti and below that you'd have to Tuf it out. :(
Savoir-Faire is everywhere!
The T1 Trust, T1 Class 4-4-4-4 #5550, America's First HST

ID: 2025471 · Report as offensive
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5126
Credit: 276,046,078
RAC: 462
Message 2026191 - Posted: 4 Jan 2020, 0:17:05 UTC

Well a few minutes ago I dropped the new Gtx 1660 Supers into my "box" and fired it up.

The only question I have left is to swap a gtx 1070 for a gtx 1660 or not?
I accidentally acquired a gtx 1660 when the listing said "Gtx 1660 Super" for an unheard of low price. No luck on the Super but it is a perfectly good Gtx 1660 that has been (previously) in the "box".

Opinions welcomed.

Tom
A proud member of the OFA (Old Farts Association).
ID: 2026191 · Report as offensive
Profile tazzduke
Volunteer tester

Send message
Joined: 15 Sep 07
Posts: 190
Credit: 28,269,068
RAC: 5
Australia
Message 2026201 - Posted: 4 Jan 2020, 1:04:09 UTC - in response to Message 2026191.  

Well a few minutes ago I dropped the new Gtx 1660 Supers into my "box" and fired it up.

The only question I have left is to swap a gtx 1070 for a gtx 1660 or not?
I accidentally acquired a gtx 1660 when the listing said "Gtx 1660 Super" for an unheard of low price. No luck on the Super but it is a perfectly good Gtx 1660 that has been (previously) in the "box".

Opinions welcomed.

Tom


Give it a run, uses less energy than the 1070 and should be about the same or close to in RAC.

Is the multiprocs number important as well????

GTX 1660 TI = multiProcs 24
GTX 1660 Super = multiProcs 22
GTX 1660 = multiProcs 22
GTX 1070 = multiProcs 15

No harm in trying and finding out though?

Regards
Mark

PS. I am impressed with my GTX 1660 Super, intend to get another one, so I dont have to use the GTX 970.
ID: 2026201 · Report as offensive
Previous · 1 . . . 16 · 17 · 18 · 19 · 20 · 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.