GPU Wars 2014: Postponed to 2015?

Message boards : Number crunching : GPU Wars 2014: Postponed to 2015?
Message board moderation

To post messages, you must log in.

1 · 2 · 3 · 4 · Next

AuthorMessage
Profile shizaru
Volunteer tester
Avatar

Send message
Joined: 14 Jun 04
Posts: 1130
Credit: 1,967,904
RAC: 0
Greece
Message 1478872 - Posted: 18 Feb 2014, 22:57:48 UTC

Haven't had time to digest these but at first glance it looks like we're back to the good ol' days of Fermi (sans toaster jokes!) :) I think Kepler proved hard to get excited about as far as Seti is concerned. Maxwell may end up far more promising but as we all know by now, specs alone can be deceptive when it comes to Seti...

Here are your links for today:

AnanadTech - Compute (page 21 of 24)
The NVIDIA GeForce GTX 750 Ti and GTX 750 Review: Maxwell Makes Its Move

TPU - Performance per Watt (page 26 of 30)
NVIDIA GeForce GTX 750 Ti 2 GB
ID: 1478872 · Report as offensive
Thomas
Volunteer tester

Send message
Joined: 9 Dec 11
Posts: 1499
Credit: 1,345,576
RAC: 0
France
Message 1479500 - Posted: 20 Feb 2014, 7:02:02 UTC - in response to Message 1478872.  

Haven't had time to digest these but at first glance it looks like we're back to the good ol' days of Fermi (sans toaster jokes!) :) I think Kepler proved hard to get excited about as far as Seti is concerned. Maxwell may end up far more promising but as we all know by now, specs alone can be deceptive when it comes to Seti...

Here are your links for today:

AnanadTech - Compute (page 21 of 24)
The NVIDIA GeForce GTX 750 Ti and GTX 750 Review: Maxwell Makes Its Move

TPU - Performance per Watt (page 26 of 30)
NVIDIA GeForce GTX 750 Ti 2 GB

Thanks for the heads-up Alex ! :)
ID: 1479500 · 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 1479541 - Posted: 20 Feb 2014, 8:53:07 UTC
Last modified: 20 Feb 2014, 8:53:53 UTC

Too bad that nVidia originally engineered Maxwell to have x86 cores as onboard counterparts, were set back so long, and were forced to seek alternatives. The RISC based ARM core is impressive, but far less understood by developers. It seems Intel and AMD bastardry backfires yet again. When will they realise a world run by lawyers would be a lonely one ?
"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: 1479541 · Report as offensive
draco
Volunteer tester

Send message
Joined: 6 Dec 05
Posts: 119
Credit: 3,327,457
RAC: 0
Latvia
Message 1479547 - Posted: 20 Feb 2014, 9:02:31 UTC - in response to Message 1479541.  

OMG!
i that way, looks like, we soon have x86 architecture in smartphones too.
world lead to more and more dumb....
ID: 1479547 · 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 1479555 - Posted: 20 Feb 2014, 9:50:13 UTC - in response to Message 1479547.  

OMG!
i that way, looks like, we soon have x86 architecture in smartphones too.
world lead to more and more dumb....


+1
"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: 1479555 · Report as offensive
Profile tullio
Volunteer tester

Send message
Joined: 9 Apr 04
Posts: 8797
Credit: 2,930,782
RAC: 1
Italy
Message 1479568 - Posted: 20 Feb 2014, 10:43:23 UTC

Which is the difference from GTX 750 ti and GTX 750 OC? I have found the latter on an Amazon catalogue and it would cost me 126 euro in Italy.
tullio
ID: 1479568 · Report as offensive
archae86

Send message
Joined: 31 Aug 99
Posts: 909
Credit: 1,582,816
RAC: 0
United States
Message 1479651 - Posted: 20 Feb 2014, 17:09:14 UTC - in response to Message 1479568.  

Which is the difference from GTX 750 ti and GTX 750 OC? I have found the latter on an Amazon catalogue and it would cost me 126 euro in Italy.
tullio

In this case the ti designation cards use the same physical chip, but with more functional units enabled for use, and a slightly higher TDP.

The OC designation generally means a version clocked more above the manufacturer's nominal clock rate than a card from the same vendor without "OC". But regardless of OC or not you can only determine the clock rate on a specific card by consulting specs, reviews, or such.

If this link works, it will take you to the fourth page of a review on the GTX 750 and 750 ti cards on Anandtech--a page which details the "ti" version differences. You may find other parts of the review useful as well.

Anandtech GTX 750 review page 4

The quick answer to the ti differences question is:

resource      std   ti
cores         640  512
texture units  40   32
TDP            55   60

ID: 1479651 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1479661 - Posted: 20 Feb 2014, 17:22:14 UTC - in response to Message 1479651.  

The quick answer to the ti differences question is:

resource      std   ti
cores         640  512
texture units  40   32
TDP            55   60

Did you really mean it that way round - with the standard version having more cores and texture units than the Ti version?
ID: 1479661 · Report as offensive
Profile tullio
Volunteer tester

Send message
Joined: 9 Apr 04
Posts: 8797
Credit: 2,930,782
RAC: 1
Italy
Message 1479666 - Posted: 20 Feb 2014, 17:33:46 UTC - in response to Message 1479651.  
Last modified: 20 Feb 2014, 17:34:00 UTC

Thanks archae86, but it seems to me that your data are different from those of the link. It is the Ti which has more cores compared to the standard and also more texture units. The TDP is correct. Since I am not a graphic card expert, can you confirm this? I have a 400 watt power supply on my SUN WS 2008 vintage and no 6 pin connector. Can I install a standard version? Thanks.
Tullio
ID: 1479666 · Report as offensive
Batter Up
Avatar

Send message
Joined: 5 May 99
Posts: 1946
Credit: 24,860,347
RAC: 0
United States
Message 1479668 - Posted: 20 Feb 2014, 17:34:54 UTC

NVIDIA GPU Specification Comparison
  		GTX 750 Ti 	GTX 750 	
CUDA Cores 		640 	512 	
Texture Units 		40 	32 	
ROPs 	24 		16 	
Core Clock 		1020MHz 	1020MHz 	
Boost Clock 	 	1085MHz 	1085MHz 	
Memory Clock 	 	5.4GHz GDDR5 	5GHz GDDR5 	
Memory Bus Width 	 	128-bit 	 	
VRAM 		2GB 	1GB 	
FP64 	 	1/32 	1/32 	
TDP 		60W 	55W 	
Transistor Count 	 	1.87B 	1.87B 	
Manufacturing Process 	 	TSMC 28nm 	TSMC 28nm 	
Architecture 	 	Maxwell 	Maxwell 	
GPU 	 	GM107 	GM107 	GK107
Launch Date 	 	02/18/14 	02/18/14 	
Launch Price 	 	$149 	$119 	

ID: 1479668 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1479671 - Posted: 20 Feb 2014, 17:40:08 UTC - in response to Message 1479668.  

NVIDIA GPU Specification Comparison
  			GTX 750 Ti 	GTX 750 	
CUDA Cores 		640 		512 	
Texture Units 		40 		32 	
ROPs 			24 		16 	
Core Clock 		1020MHz 	1020MHz 	
Boost Clock 	 	1085MHz 	1085MHz 	
Memory Clock 	 	5.4GHz GDDR5 	5GHz GDDR5 	
Memory Bus Width 	 	128-bit 	 	
VRAM 			2GB 		1GB 	
FP64 	 		1/32 		1/32 	
TDP 			60W 		55W 	
Transistor Count  	1.87B 		1.87B 	
Manufacturing Process	TSMC 28nm 	TSMC 28nm 	
Architecture 	 	Maxwell 	Maxwell 	
GPU 	 		GM107 		GM107
Launch Date 	 	02/18/14 	02/18/14 	
Launch Price 	 	$149 		$119 	
ID: 1479671 · Report as offensive
archae86

Send message
Joined: 31 Aug 99
Posts: 909
Credit: 1,582,816
RAC: 0
United States
Message 1479675 - Posted: 20 Feb 2014, 17:59:15 UTC

As several have pointed out, I got the ti vs non-ti numbers wrong in my first post in this thread. I was not even consistent, as I got the TDP right-way round, and the cores and texture units backwards. As the time to edit has expired, I'll just let the several corrective postings by others stand.

Sorry about that.

While some of the card vendors suggest extravagantly high rated power supplies even for GTX 750 cards, if you are buying a non-overclocked card and refrain from overclocking and overvolting it yourself, I think quite modest power supplies--provided exclusively through the PCIe connector, will very likely prove entirely adequate. In particular, I'd be very surprised if Tullio's 400 watt supply were to prove inadequate. But I'd not buy a card with the 6-pin extra power connector if I lacked means to provide that connection.

A reference in one or more of the reviews has me thinking that 75 watt card total is the nominal PCIe connector-supplied power limit. As the TDP is just for the GPU chip, the rest of the card could put you a little closer to this than you might think. But I think most of the companies offering base 750 cards are not providing 6-pin extra power connectors, and not even all of the ti overclocked cards provide them.

On the graphics card scale of modern times, these things are actually pretty low power--considerably higher than the sub $50 fanless cards I used to buy, but well below what were considered mid-range cards in the last generation, and very far below any card with gamer bragging rights.
ID: 1479675 · Report as offensive
Profile tullio
Volunteer tester

Send message
Joined: 9 Apr 04
Posts: 8797
Credit: 2,930,782
RAC: 1
Italy
Message 1479699 - Posted: 20 Feb 2014, 18:47:56 UTC - in response to Message 1479675.  

Thanks everybody for their help. Sic stantibus rebus, I shall not buy the Gigabyte GTX 750 OC since it is probably overclocked. I shall wait for a not overclocked version.
Tullio
ID: 1479699 · Report as offensive
Profile shizaru
Volunteer tester
Avatar

Send message
Joined: 14 Jun 04
Posts: 1130
Credit: 1,967,904
RAC: 0
Greece
Message 1479762 - Posted: 20 Feb 2014, 20:26:52 UTC

A few random tidbits:

@Tullio - OC stands for (is the abbreviation of) "overclocked", so yes, you were looking at an overclocked card... At the bottom of this link is a chart and - in bold - has the differences between the 750 and 750 OC. A few columns over is the 750Ti too.
http://www.techpowerup.com/reviews/ASUS/GTX_750_OC/

@archae86 - always nice to see you drop by :)

Important:
Forgot to mention that since this is a new architecture, there is a chance it may not work out-of-the-box with Seti.
ID: 1479762 · Report as offensive
Gone

Send message
Joined: 31 May 99
Posts: 150
Credit: 125,779,206
RAC: 0
United Kingdom
Message 1480187 - Posted: 21 Feb 2014, 14:10:13 UTC

Hi, I just took delivery of 3 x 750 Ti and will fire them up in a couple of hours.
I will post back here if they work or not.

Any suggestions as to which version I should run Cuda 42 or Cuda 50 ?

Reg
ID: 1480187 · Report as offensive
Profile arkayn
Volunteer tester
Avatar

Send message
Joined: 14 May 99
Posts: 4438
Credit: 55,006,323
RAC: 0
United States
Message 1480205 - Posted: 21 Feb 2014, 14:58:19 UTC - in response to Message 1480187.  

Hi, I just took delivery of 3 x 750 Ti and will fire them up in a couple of hours.
I will post back here if they work or not.

Any suggestions as to which version I should run Cuda 42 or Cuda 50 ?

Reg


CUDA 50 would be your best bet.

ID: 1480205 · Report as offensive
Profile shizaru
Volunteer tester
Avatar

Send message
Joined: 14 Jun 04
Posts: 1130
Credit: 1,967,904
RAC: 0
Greece
Message 1480215 - Posted: 21 Feb 2014, 15:20:45 UTC - in response to Message 1480205.  
Last modified: 21 Feb 2014, 15:29:58 UTC

Hi, I just took delivery of 3 x 750 Ti and will fire them up in a couple of hours.
I will post back here if they work or not.

Any suggestions as to which version I should run Cuda 42 or Cuda 50 ?

Reg


CUDA 50 would be your best bet.


I'm not pretending to be even remotely qualified to take a shot at answering but I do want to say I wouldn't be surprised if it turns out to be Cuda 42. As I understood the Anand article, Maxwell architecture is closer to Fermi than Kepler. If that is truly the case and I'm not oversimplifying my reasoning (though I probably am) that could even mean that 42 works while 50 doesn't? We'll know soon enough:)

Congratz on your new cards RED and if you are up for it, you may need to consult with someone like Richard or Joe (or anyone who thoroughly understands Server-side validation) in case it might be a good idea to run stock for a few days or until you complete 10 tasks from each app? I honestly have no idea if this would help the project in any way, just bringing it up so a guru can shoot it down or expand on it:)

Edit: Apologies for getting your name wrong Reg :)
ID: 1480215 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1480253 - Posted: 21 Feb 2014, 16:31:45 UTC - in response to Message 1480215.  

Hi, I just took delivery of 3 x 750 Ti and will fire them up in a couple of hours.
I will post back here if they work or not.

Any suggestions as to which version I should run Cuda 42 or Cuda 50 ?

Reg

CUDA 50 would be your best bet.

I'm not pretending to be even remotely qualified to take a shot at answering but I do want to say I wouldn't be surprised if it turns out to be Cuda 42. As I understood the Anand article, Maxwell architecture is closer to Fermi than Kepler. If that is truly the case and I'm not oversimplifying my reasoning (though I probably am) that could even mean that 42 works while 50 doesn't? We'll know soon enough:)

Congratz on your new cards RED and if you are up for it, you may need to consult with someone like Richard or Joe (or anyone who thoroughly understands Server-side validation) in case it might be a good idea to run stock for a few days or until you complete 10 tasks from each app? I honestly have no idea if this would help the project in any way, just bringing it up so a guru can shoot it down or expand on it:)

Edit: Apologies for getting your name wrong Reg :)

Unless you intend to put then all in the same machine, why not try cuda42 on one, cuda50 on another? Heck, even cuda32 on a third for giggles. I don't know whether Jason has got or is getting a Maxwell yet - he might even have a cuda55 or cuda60 he needs testing.

I don't think there's any need to run stock first - even if you're upgrading an existing CUDA host, your current cards are beefy enough for the speeds to match well enough, and they'll settle down soon enough. But - and this may be redundant advice given the current state of the splitters - do please turn down your cache/work request before you start, in case tasks get trashed: this is very much untested technology.
ID: 1480253 · Report as offensive
Gone

Send message
Joined: 31 May 99
Posts: 150
Credit: 125,779,206
RAC: 0
United Kingdom
Message 1480256 - Posted: 21 Feb 2014, 16:36:42 UTC

750 Ti installedmon this host http://setiathome.berkeley.edu/show_host_detail.php?hostid=7218229

Behaving like any the card so far. Latest Nvidia driver 334.89 and currently doing MB on Cuda 42.

21/02/2014 16:22:34 | | CUDA: NVIDIA GPU 0: GeForce GTX 750 Ti (driver version 334.89, CUDA version 6.0, compute capability 5.0, 2048MB, 1938MB available, 2183 GFLOPS peak)
21/02/2014 16:22:34 | | OpenCL: NVIDIA GPU 0: GeForce GTX 750 Ti (driver version 334.89, device version OpenCL 1.1 CUDA, 2048MB, 1938MB available, 2183 GFLOPS peak)


I will try it on Cuda 50 after a while and then try all 3 cards.


Reg
ID: 1480256 · Report as offensive
archae86

Send message
Joined: 31 Aug 99
Posts: 909
Credit: 1,582,816
RAC: 0
United States
Message 1480263 - Posted: 21 Feb 2014, 16:46:21 UTC - in response to Message 1480187.  

Big Reg wrote:
Hi, I just took delivery of 3 x 750 Ti and will fire them up in a couple of hours.
Which particular manufacturer/model did you obtain? And what clock speed are they set to by default and have you yet tampered with any of the default settings?

Lastly, I got excited because one of your hosts which currently shows as having a 750 ti got a validated WU, but on reading the details the particular task seems to have been done on a 560 ti.

I myself have a 750 base model on order (EVGA 01G-P4-2751-KR, 1020 MHz), but will bring it up first on Einstein, and probably won't get first results until a week or two from now. If 750 results are still scarce here by the time I get it running, I'll do some trials on SETI, but suspect there will be others long before me.
ID: 1480263 · Report as offensive
1 · 2 · 3 · 4 · Next

Message boards : Number crunching : GPU Wars 2014: Postponed to 2015?


 
©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.