Upcoming GTX 750 Ti tested

[f1sherman]

From TPU link above:

[...] partner-branded GeForce GTX 750[...]
Other than that, plenty of other sources confirm ref. design w/o 6-pin

Besides look at that picture again.
That dinky fan at 30% is cooling GPU @99% GPU load, and keeping 35 C.
Now use some commons physics sense to come up with TDP of that thing

 

[coffeejunkee]

You're right. There are models without pci-e connector too.

But it seems there are less cores than expected: http://videocardz.com/49581/asus-geforce-gtx-750-series-detailed-plus-real-specs-gm107

I thought this wouldn't even be Maxwell so I'm going to wait and see before I say any more.

 

[ams23]

With the latest leaked info, I am definitely convinced that 750 Ti and 750 are using a Maxwell derivative rather than a Kepler.M derivative. A Maxwell CUDA core will likely be capable of doing significantly more work than a Kepler CUDA core, and the perf. per watt and perf. per mm^2 appears to be superior to Kepler as well.

 

[Arkaign]

Yep, seems like a dry run of Maxwell design on an old but mature process. It's actually a good idea from a technical perspective, if not that exciting for the enthusiast class.

However, looking at what's possible with so few cores and so tiny a chip even on 28nm, a full blown boss mode Maxwell will be spectacular. Something tells me we'll see Radeon R390/390X at a similar time.

 

[Bubbleawsome]

Imagining the GM110.

 

[parvadomus]

So, in the end it packs 640 shaders @ 1085Mhz? It makes more sense now.

 

[f1sherman]

WTH 28nm(?), yet 2x perf/Watt vs Kepler

http://wccftech.com/nvidia-maxwell-...d-60watt-gpu-geforce-800-series-arrives-2014/

I don't think so... not on 28nm. Remember that NV roadmap?
It puts Maxwell at 200% of Kepler's DP perf/Watt, but that is supposed to be 20nm... right?

 

[boxleitnerb]

Why would they talk about unannounced products that are still 6+ months ahead? It's 28nm. But I guess they cherrypick like everybody else, so maybe 160-170%.

 

[f1sherman]

They wouldn't.
But above numbers puts "Maxwell 1st gen" at "Maxwell" in their roadmap.

Cherry picking? Could be, but their CUDA roadmaps are surprisingly accurate(save for Fermi time window slip)

 

[Arkadrel]

WTH 28nm(?), yet 2x perf/Watt vs Kepler

http://wccftech.com/nvidia-maxwell-...d-60watt-gpu-geforce-800-series-arrives-2014/



I don't think so... not on 28nm. Remember that NV roadmap?
It puts Maxwell at 200% of Kepler's DP perf/Watt, but that is supposed to be 20nm... right?

It sounds to good to be true.

Maybe its for something like computational calculations?
and not 2x perf/watt for gameing or such.

 

[tviceman]

It's 2x perf/watt compared to the GT 640. The marketing slides in question were comparing the gtx 750 ti to the gt 640 in gaming.

The 750 ti is about 65-75% more efficient on 28nm than the gtx 650. It will be ~2x more efficient than the gtx 650 when it gets the die shrink.

 

[sontin]

It sounds to good to be true.

Maybe its for something like computational calculations?
and not 2x perf/watt for gameing or such.

They are comparing the core/shader architecture. In games it will come down because of the bandwidth and other things.

However even a 50% improvement it's a leap like a shrink.

 

[USER8000]

It's 2x perf/watt compared to the GT 640. The marketing slides in question were comparing the gtx 750 ti to the gt 640 in gaming.

The 750 ti is about 65-75% more efficient on 28nm than the gtx 650. It will be ~2x more efficient than the gtx 650 when it gets the die shrink.

There are two versions of the GT640 though - one is GF116 based and the other GK107 based. Also,considering there were bus powered HD7850 2GB single slot cards released,I am not seeing what the big deal is:

http://www.kitguru.net/components/graphic-cards/zardon/afox-hd7850-single-slot-crossfire-review/

That card was released in 2012.

The thing is I have done some builds with the GTX660 and GTX650TI Boost cards,and you need to consider the GK106 has 50% more shaders and a 256 bit memory controller. Reduce that to a dedicated die with 640 shaders and a 128 bit memory controller,and I expect the card could easily be bus powered.

 

[Crap Daddy]

Also,considering there were bus powered HD7850 2GB single slot cards released,I am not seeing what the big deal is:

http://www.kitguru.net/components/gr...ssfire-review/

What are you talking about? Those cards use 110W and have one six pin connector.

 

[USER8000]

What are you talking about? Those cards use 110W and have one six pin connector.

LOL,my bad!!

I must have confused it with this:

http://www.techpowerup.com/185003/triplex-shows-off-slot-powered-radeon-hd-7850.html

Its a bus powered HD7850 2GB with the same cooler.

It looks exactly the same!!

 

[parvadomus]

It's 2x perf/watt compared to the GT 640. The marketing slides in question were comparing the gtx 750 ti to the gt 640 in gaming.

The 750 ti is about 65-75% more efficient on 28nm than the gtx 650. It will be ~2x more efficient than the gtx 650 when it gets the die shrink.

A pile of PR BS. The GT640 (DDR3) is obviously bandwidth constrained @4xAA.

 

[altairauditore]

With respect to the 3D perf scores scored by both the new Maxwell cards, It seems that the leaked scores make sense:

The GTX 750 and 750Ti perform pretty similar to the GTX 650Ti :

GPU 3DMark 11 Score
Nvidia GTX 750 5250
Nvidia GTX 750 Ti 5963
Nvidia GTX 650 Ti 4819
Nvidia GTX 660 2GB 6727

Now check the following image

It seems that Nvidia wants to replace the GTX 650ti with the 2 new 750 series cards.. which makes sense becuase of the huge performance gap between the GTX 650 and the GTX 650ti.

The GTX 750 and 750ti will fit into the same market segment.. in terms of performance.. but the whats awesome is that with Maxwell, they have awesome power consumption..

So the GTX 650 Wattage ~ GTX 750ti wattage.. but the GTX 750ti will be so much faster..

This is going to be awesome for lower powered desktops :awe:

 

[sontin]

Look at the PC on the desk. D:

 

[altairauditore]

More data here: http://wccftech.com/nvidia-maxwell-geforce-gtx-750-ti-gtx-750-official-specifications-confirmed-60watt-gpu-geforce-800-series-arrives-2014/

 

[altairauditore]

Look at the PC on the desk. D:

i think I can spot a GPU with a big axial cooling fan in there

 

[altairauditore]

Also, GTX 650Ti = 100W
GTX 750Ti = 65W

Wow.. plus there's also a performance improvement.. so that's where Nvidia gets the 2x perf/Watt figure

 

[Lonyo]

But the 650Ti is a cut down 660, not a designed to be low power card, so you would expect a card designed to be smaller to be more efficient.
Historically with third tier chips, they often ran higher voltages than the "top" cards using that GPU in order to ensure max yield, so it's entirely possible vs a 650Ti to get 2x the per/watt, because the 650Ti is a harvested GPU, not a low power GPU, and will probably have some leaky bits, and may not be perfectly balanced due to being heavily harvested.

And people have already made comments on the GT640 which can either be an old process, hampered by DDR3, or it could be a standard 28nm GDDR5 GPU. So any perf/watt comparison is pretty pointless unless it's against a standard 650.

 

[jpiniero]

If the 750 Ti is really 60W, that's pretty stellar. I have my doubts.

 

[Arkadrel]

just wanted to say you can cherry pick benchmarks to show same thing with a 7750 (no 6pin connector):

Nvidia Geforce 640 DDR3 = 50w TDP
AMD 7750 (no 6pin connector) = 55w TDP

Now look at their fps in this benchmark, its over 2x (perf/watt).

 

[altairauditore]

I din't really mean to say that it IS 2x.. but it could come pretty close.. Even If it doesnt, GTX 650ti class performance @ 60W is pretty good at the same process node.

Sorry for my 2x statement.

Yes, It's true that the lower bin parts often run using more power because they aren't as "perfect" as the higher bins.. and yes there's also the issue of the turned off SMX's still hogging power although they aren't doing anything.

If you take GTX 650 (GK107?) and the new Maxwell core, it seems to be a fairer comparison .. and Maxwell wins in this case..