Its only been running 30 mins and it had some tasks done by a 560 TI during build testing last night.

I started it off with 3 tasks for a while, but it is now down to 2 tasks at 98 % GPU Load
Spec is:
MSI GeForce GTX 750 Ti TWIN FROZR GAMING OC 2GB GDDR5 1085MHz Core, 1163MHz Boost, 5400MHz Memory, 640 CUDA Cores, DVI-D, HDMI, DSub, PCIe 3.0, 3 Year Warranty

Running at 35Deg C

All stock speeds, Lunatics x41 Cuda 42.

The interesting part will be to find out what this is:

computeCap 5.0

Given that Kepler Class is 3.0, and BigK (GK110) is 3.5

What I'll do is wade through the unreleased Cuda 6 documentation for hints, though I don't recall seeing mention of it in the early access stuff, I admit I wasn't really looking for that though. Probably should be some in the release candidate...

---

"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
If you need me to try anything, just let me know.

Reg

@Jason
If you need me to try anything, just let me know.

Reg

Well it's a bit of a mystery at the moment what features this compute capability has, and how to use them ;) I'll keep digging.

---

"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.

I know it's early days still but how does it compare against the 560Ti?

It seems to be 2x faster ! Sadly no tasks at moment.

It seems to be 2x faster ! Sadly no tasks at moment.

Very impressive, especially when you consider the power requirements.

GTX 750Ti 60W
GTX 560Ti 210W

Double the processing, for 1/3 of the power.


EDIT- looks like I've found my new video card. If the price is right, might as well get 2 of them to replace my present 560Ti (fan bearings have had it).

---

Grant
Darwin NT

It seems to be 2x faster ! Sadly no tasks at moment.

Very impressive, especially when you consider the power requirements.

GTX 750Ti 60W
GTX 560Ti 210W

Double the processing, for 1/3 of the power.


EDIT- looks like I've found my new video card. If the price is right, might as well get 2 of them to replace my present 560Ti (fan bearings have had it).

I would guess that factory OC card may be using more than the default 60w configuration rating. With it pumping out 2183 vs 1306 GFLOPS of the stock spec card at least.
No doubt it is still more efficient that the older cards.

---

SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[

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

In reading articles on this card a little while ago, there was no specific mention of either a 6 or 8 pin connector and there was mention that the card drew all its power from the PCI-E bus (60 watts). If the inferences that I read were not wrong and there is no other connector for power on each card (and all 3 are in the one machine) you may be running the risk of melting the 24 pin connector, or damaging the board in some other way.

---

In reading articles on this card a little while ago, there was no specific mention of either a 6 or 8 pin connector and there was mention that the card drew all its power from the PCI-E bus (60 watts).

Among the several GTX 750 and 750 ti cards I've looked over in the last few days, some do have am 6-pin additional power connector (widespread on expressly overclocked 750 Ti cards), and some do not (unknown or scarce on base spec 750 cards) .

References I've seen suggest that the design limit on card draw from the PCIe connector is 75 watts, and I think it likely that all the cards delivered without the 6-pin connector are expected to stay well under that limit under normal operating conditions. As I think the GPU chip supervises its own power consumption, possibly it may be difficult greatly to exceed that even under moderate user abuse.

I would hope the people who developed the PCIe specs and the designers of the motherboards both contemplated that someone might actually plug something into all of the offered slots.

These things are FAR lower power consuming than the higher end sort of gaming cards of the recent past.

---

I know it's early days still but how does it compare against the 560Ti?

It seems to be 2x faster !

If that's true then it's the biggest news since the move to the colo.

I was going to say I'm surprised the forums haven't caught fire but then the cynic in me tapped my shoulder and promptly reminded me that there is hardly ever any number crunching done in the Number Crunching forum:)

But are u sure? Because before you posted I had a look at your tasks and for some reason I came to the conclusion that the 750ti can do about 80% of what the 560ti can do. Specs would be closer to these numbers too but... specs and Seti never seem to go hand in hand.

Hopefully I'm wrong and my guess on how many tasks you were crunching on each card was off:)

Taken from the Boinc Manager event log just now .....


GeForce GTX 560 Ti 1312 GFLOPS peak) (That's a Ti 448 too)

GeForce GTX 750 Ti 2183 GFLOPS peak)



24/02/2014 11:49:14 | | CUDA: NVIDIA GPU 0: GeForce GTX 560 Ti (driver version 334.89, CUDA version 6.0, compute capability 2.0, 1280MB, 1163MB available, 1312 GFLOPS peak)
24/02/2014 11:49:14 | | CUDA: NVIDIA GPU 1: GeForce GTX 750 Ti (driver version 334.89, CUDA version 6.0, compute capability 5.0, 2048MB, 1948MB available, 2183 GFLOPS peak)
24/02/2014 11:49:14 | | OpenCL: NVIDIA GPU 0: GeForce GTX 560 Ti (driver version 334.89, device version OpenCL 1.1 CUDA, 1280MB, 1163MB available, 1312 GFLOPS peak)
24/02/2014 11:49:14 | | OpenCL: NVIDIA GPU 1: GeForce GTX 750 Ti (driver version 334.89, device version OpenCL 1.1 CUDA, 2048MB, 1948MB available, 2183 GFLOPS peak)



I am using PCIe x1 adaptors to run the 750TIs which have their own molex connector to power the PCI e x16 slot.

---

Did you try to run the benchmark tools? It could be interesting to see the real performance board on SETI.

You could DL them at: http://lunatics.kwsn.net/index.php?module=Downloads;catd=5

If you do, try first with the GPu direct connected to the PCI-E 3.0 and after that with it conected to the extender. That´s will give you, first the real performance of the GPU and the second how much you loose by using the adaptor (if you loose anything of course).

And please share with us the findings.

---

With my relatively new (and self-proclaimed) role as Cassandra of these boards I'm going to go ahead and guess that Boinc is just "seeing" the flops wrong (it's not like it runs any benchmarks to get that number, just blurts out whatever it's told AFAIK).

The Gflops for the 448 are spot on but the Gflops for the 750ti should be around 1300 (according to Wiki anyway, which can be a bit dodgy of course).

I do hope I'm wrong though.

With my relatively new (and self-proclaimed) role as Cassandra of these boards I'm going to go ahead and guess that Boinc is just "seeing" the flops wrong (it's not like it runs any benchmarks to get that number, just blurts out whatever it's told AFAIK).

The Gflops for the 448 are spot on but the Gflops for the 750ti should be around 1300 (according to Wiki anyway, which can be a bit dodgy of course).

I do hope I'm wrong though.

I was thinking the same. The speed (peak flops) of an NVidia card depends - critically - on the number of shaders or cuda cores (pick your own terminology).

For some absurd reason, NVidia makes it stupidly difficult to detect that number programmatically. You can ask the API for the number of 'Streaming Multiprocessors' on the card, and it'll tell you. But ask it how many shaders per SM? Nada.

Instead, you have to know (in advance) that it's some function of the Compute Capability of the card. The original NV cards were 8 shaders per SM: Fermis jumped it to 32, some Keplers went up to 48 - but now it seems to have become variable. Accordingly to the (possibly the same) Wiki article, these little Maxwells seem to be back down to 16 S per SM - in which case, BOINC might over-estimate them by 2x or 3x.

Unless our resident NV guru can tell us that NVidia have introduced a new API call at last?

Edit - COPROC_NVIDIA::set_peak_flops() in /lib/coproc.cpp seems to be stuck in a timewarp. Don't believe anything you read there.

That´s exactly why i ask him to run the benchmarks so we could see their performance in the "real SETI world AP/MB".

---

I will build a host with 2 x 750s in , one on PCIe x16 and one via x1 and then run benchmarks.
I am seeing varying run times at the moment running 2 tasks per card on both 560 Ti and 750 Ti
For me the real beauty of these cards is that they need no extra power cables.

Reg

I will build a host with 2 x 750s in , one on PCIe x16 and one via x1 and then run benchmarks.
I am seeing varying run times at the moment running 2 tasks per card on both 560 Ti and 750 Ti
For me the real beauty of these cards is that they need no extra power cables.

Reg

sorry, dont understand.
cards TDP is about 60w.
PCIe power possibilities is around 75W.
if you use one card, then all is ok.
if you want use two cards - there is no possibility to use it without extra power cables?

---

Look one of his previsous post, he is ussing this adaptor who drives the power for the second GPU from the molex connector not the PCI-e. The question is what is the impact on crunching the use of the slow PCI x1 slot will do (speed/time to crunch a WU).

---

Look one of his previsous post, he is ussing this adaptor who drives the power for the second GPU from the molex connector not the PCI-e.

molex != extra power cable? :-O

---