Advanced search

Message boards : Graphics cards (GPUs) : GTX 1050 Ti?

Author Message
MossyRock
Send message
Joined: 26 Nov 13
Posts: 17
Credit: 50,096,588
RAC: 0
Level
Thr
Scientific publications
watwat
Message 47400 - Posted: 11 Jun 2017 | 22:39:30 UTC
Last modified: 11 Jun 2017 | 22:40:43 UTC

Since my new BOINC machine will be in a 2U server case, I'm limited to low-profile graphics boards. I considered 2U cases that have provisions for horizontally-mounted, full-sized boards which would require a riser card, but there were too many uncertainties with that approach.

This is what I'm looking at - an MSI GeForce GTX 1050Ti:

https://www.newegg.com/Product/Product.aspx?Item=N82E16814137081

I know this card isn't state of the art. And, I know it is about half the power as the GTX 1080.

1) Do you see any issues with this board with respect to using it for GPUGrid?
2) Is it already considered "obsolete" and I just shouldn't bother?
3) Do you know of any low-profile graphics boards that would have better performance?

Profile Retvari Zoltan
Avatar
Send message
Joined: 20 Jan 09
Posts: 1814
Credit: 9,960,094,194
RAC: 6,342,775
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 47401 - Posted: 12 Jun 2017 | 7:02:08 UTC - in response to Message 47400.

Since my new BOINC machine will be in a 2U server case, I'm limited to low-profile graphics boards. I considered 2U cases that have provisions for horizontally-mounted, full-sized boards which would require a riser card, but there were too many uncertainties with that approach.

This is what I'm looking at - an MSI GeForce GTX 1050Ti:

https://www.newegg.com/Product/Product.aspx?Item=N82E16814137081

I know this card isn't state of the art.
The GTX 1050Ti is a Pascal (GP107) GPU, so it's state of the art (until the release of Volta GPUs, but that will make the GTX 1080 and the GTX 1080Ti also "obsolete").

And, I know it is about half the power as the GTX 1080.
That is the GTX 1060. The GTX 1050Ti has less computing power than the third of a GTX 1080 (768 vs 2560 CUDA cores, and lower clocks).

1) Do you see any issues with this board with respect to using it for GPUGrid?
This is a low middle range card, so it's slow.

2) Is it already considered "obsolete" and I just shouldn't bother?
It's not obsolete, but it's not intended for such demanding applications like GPUGrid. This specific card will emit all of its heat inside the case, so you have to make sure that the case will have proper airflow. Blade servers tend to have cooler fans with very high (up to 15.000(!)) RPM which emits unbearable noise. A 2U server probably has bigger and lower (~5000) RPM fans, but it still not recommended to live nearby such equipment.

3) Do you know of any low-profile graphics boards that would have better performance?
Passive cooled computing cards exists (designed for rackmount cases with low clocks and low voltages to achieve low heat dissipation), but they are much too expensive for BOINC use.

MossyRock
Send message
Joined: 26 Nov 13
Posts: 17
Credit: 50,096,588
RAC: 0
Level
Thr
Scientific publications
watwat
Message 47410 - Posted: 12 Jun 2017 | 12:57:00 UTC
Last modified: 12 Jun 2017 | 12:57:49 UTC

Thanks for taking the time to answer my questions. This is excellent information.

I didn't realize that this card was considered to be so low in the performance lineup. I'd be starting right out of the gate with a serious under-performer.

I've been away from GPUGrid for several years now and wasn't aware how demanding the WUs have become.

I'm going to have to re-think my entire build. Although I didn't want a 4U chassis due to rack space constraints, it looks like now I have no choice if I want decent GPU power, which means a full-height card.

JoergF
Avatar
Send message
Joined: 20 Apr 15
Posts: 189
Credit: 224,425,186
RAC: 373,030
Level
Leu
Scientific publications
watwat
Message 47413 - Posted: 12 Jun 2017 | 15:15:05 UTC
Last modified: 12 Jun 2017 | 15:21:41 UTC

3) Do you know of any low-profile graphics boards that would have better performance?
Passive cooled computing cards exists (designed for rackmount cases with low clocks and low voltages to achieve low heat dissipation), but they are much too expensive for BOINC use.



Well, if you want to spend 400$ ... there is one "active cooled" card on market with a smashing crunching performance but single slot height: that is the KFA² GeForce GTX 1070 Katana Single Slot. So it IS possible to have a fast (active cooled) card of that size indeed.

Having said this, Zoltan is correct in regard to the airflow. You need to pay close attention to this.
____________
Die Liebe allein versteht das Geheimnis, andere zu beschenken und dabei selbst reich zu werden. [Clemens von Brentano]
Only love understands the secret of giving and getting richer at the same time [Clemens of Brentano]

MossyRock
Send message
Joined: 26 Nov 13
Posts: 17
Credit: 50,096,588
RAC: 0
Level
Thr
Scientific publications
watwat
Message 47416 - Posted: 13 Jun 2017 | 1:00:38 UTC
Last modified: 13 Jun 2017 | 1:01:24 UTC

JoergF,

That's a pretty slick card!

For me the issue isn't the number of slots the card occupies, but the height of the card.

It appears to me to be standard-height. I need low-profile, or reduced-height.

Thanks.

KSUMatt
Avatar
Send message
Joined: 11 Jan 13
Posts: 214
Credit: 831,004,493
RAC: 0
Level
Glu
Scientific publications
watwatwatwatwatwatwatwatwat
Message 47417 - Posted: 13 Jun 2017 | 4:59:46 UTC - in response to Message 47410.

I've been away from GPUGrid for several years now and wasn't aware how demanding the WUs have become.


The GPUGrid team have been good at keeping their models and tasks up-to-date for the most recent GPUs available. Every time a new generation of Nvidia GPUs come out we see an increase in computational demand from this project. This has both its pros and cons.

Erich56
Send message
Joined: 1 Jan 15
Posts: 346
Credit: 1,457,365,177
RAC: 2,705,716
Level
Met
Scientific publications
watwatwat
Message 47475 - Posted: 18 Jun 2017 | 19:28:09 UTC - in response to Message 47417.

Every time a new generation of Nvidia GPUs come out we see an increase in computational demand from this project. This has both its pros and cons.

In my eyes, the cons overweigh.
Let's just face what this means: a volunteer who want's to be up to date would have to buy new cards about every year.
If I take myself as example:
last year, in addition to my GTX750ti and my GTX970 I bought two GTX980ti.
With latter, at that time, the longest tasks ran about 6 hours. Today, they run about 12+ hours.
Last year, the longest task with the GTX970 took about 11 hours, now it takes about 22 hours.
And, most sadly, the GTX750Ti - with the unintended stops inbetween (due to a bug in the crunching software, as described in some other threads) may take up to 3 days for a very demanding "longruns".

What does this mean?
What concerns the GTX750Ti - I should forget about it.
What concerns the GTX970 - I'll be able to use it for GPUGRID maybe one more year, then also forget about it.
What concerns the GTX980Ti - hopefully I can use them for a little more than one additional year - maybe not.

To take account of all the different cards still in use by many crunchers, I suggested in another thread that a third type of task be implemented, like "extra long runs" or similar. So that the crunchers could at least properly allocate the tasks available to their different cards.

Post to thread

Message boards : Graphics cards (GPUs) : GTX 1050 Ti?