Upcoming GTX 750 Ti tested

[f1sherman]

Curious that it s still DX11.0 only like its predecessors....

More like... annoying

 

[Qwertilot]

Not entirely sure about thinking about performance/watt in percentage terms.

Certainly if I was building something I'd start with a given performance target - say 1080/30/sane settings - and then see how low I could rationally get the power draw to do that.
(Or cost of course, depending on what I felt like optimising for. I suspect it'll be power for me next time.).

The 750's seem to do that quite comfortably, although what native next gen console games will need could be a question. I'd only consider more performance relevant if getting something tangiable like 60FPS, much higher settings or (especially) higher resolution for it.

One other minority sport is I guess those tiny pico psu's which do ~160W total. You could easily get a 750, low power haswell model etc into that budget, which could make for some very small machines indeed.

 

[USER8000]

Not entirely sure about thinking about performance/watt in percentage terms.

Certainly if I was building something I'd start with a given performance target - say 1080/30/sane settings - and then see how low I could rationally get the power draw to do that.
(Or cost of course, depending on what I felt like optimising for. I suspect it'll be power for me next time.).

The 750's seem to do that quite comfortably, although what native next gen console games will need could be a question. I'd only consider more performance relevant if getting something tangiable like 60FPS, much higher settings or (especially) higher resolution for it.

One other minority sport is I guess those tiny pico psu's which do ~160W total. You could easily get a 750, low power haswell model etc into that budget, which could make for some very small machines indeed.

Toms Hardware saw this though:
http://media.bestofmicro.com/2/W/422600/original/01-GTX-750-Ti-Complete-Gaming-Loop-170-seconds.png

Measurement Procedure:Non-Contact Direct Current Measurement at the PCIe Slot
Non-Contact Direct Current Measurement at the External PCIe Power Supply
Direct Voltage Measurement 3.3 V / 12 V

According to Toms Hardware those are card measurements not power consumption at the wall.

There are very quick peaks upto 141W,and the problem is picoPSUs don't have the strongest 12V lines and also the higher wattage ones require a decent power block.

I would be more comfortable using a decent TFX form factor PSU.

 

[el etro]

To be fair, the the R7 265 is running quite a bit cooler than the 750 Ti in the review because Ryan's using the numbers from the Sapphire Dual-X cooler R7 265 http://www.anandtech.com/show/7754/the-amd-radeon-r7-265-r7-260-review-feat-sapphire-asus/2

Reference HD 7850 uses the same chip and closer clocks, runs at 66 versus 61 degrees at furmark(don't represent gaming condition). We are not talking here about GPUs that operate at 95 degrees in reference cards operating at halve or 25% less temperature.

Thoses comparisons are moot , actualy the most powerfull card
will likely get the CPU to be maxed out and hence it will inflate
the system power draw because of the CPU, not only because
of the GPU...

That's why perf/w matters more. Another problem is to measure CPU power/threads needed to make each card max its usage. But take this in account: CPU power consumption is related too to the operating clocks. If the processor is clocking at 3.5Ghz instead of 2.6Ghz it will interfere on the power consumption.
And this is why we need new rules to CPU testing on VGA benchmarks.

Toms Hardware saw this though:
http://media.bestofmicro.com/2/W/422600/original/01-GTX-750-Ti-Complete-Gaming-Loop-170-seconds.png

Some custom 290 averages 214 watts with 375 watts peak on the test. This peak is not so influent on PSU chosing, peak method that TPU calculates is much better.

 

[Atreidin]

So does anybody have an idea of how this might perform as an add-in card for physx?

 

[Madmick]

To be frank, there are smarter options than that.

http://www.newegg.com/Product/Produc...82E16817139049

$30 and he'd never have to worry about powering most single chip graphics cards ever again.

Which would be great if it wasn't an old HP office build on a unique form factor that isn't ATX compliant, and that you're just continuing to billow the budget here for a guy who really didn't want to spend more than $60 on a card. He was frustrated when I had to explain to him that because of Cryptomining, that GDDR5 version of the 6670 that my little brother got like 1 to 1 1/2 years ago for $65 (after rebate) that has served him marvelously for WoW now starts around $80-$85 for the crappy DDR3 version.

What an awesome deal that is btw - note that it comes with a 6+2 pin PCI-E - that should tell you how confident Corsair is about their ability to handle most graphics cards. I've sold a ton of these in the past and never had one back yet.

Yes, this is my standard entry PSU when I make recommendations to new builders.

 

[coffeejunkee]

I don`t understand why you people use so much energy in trying to make certain people change their minds:

snip.

It's not a question of changing minds. It's about trying to find out how efficient Maxwell really is. The TPU chart has an unreal vibe to it, at least to me. So I decided to see what AT says. Yes, I realize the test system is not representative for the average user, but all cards use it. Power and fps from just one game also doesn't give a representative picture. Saying you should only look at gpu power usage and not system power usage is nonsense though, you can't use a gpu without a system. Also, what if a gpu archictecture requires the cpu to work harder? Measuring just the card itself would completely miss that.

I did those calculations to see how R7 260X compared to 750 Ti according to AT numbers. 70% is at least a bit better than the 61% TPU shows. I also looked at 650 Ti to see the increase against the Kepler but it's a bit underwhelming. I added the R7 265 as a sort of afterthought but was really surprised to see it actually have better perf/W than the 750 Ti. Had me doublechecking. Again, I realize the test system is not representative. And it's just one game. But these are the numbers we have.

My main concern is actually the big difference between benchmarks across sites. On TPU the 750 Ti gets more fps than 7870 and GTX 660 in Crysis 3. On AT it's slower than R7 265 which itself is slower than GTX 660.

Other weird results on TPU:

7790/R7 260X
Batman Arkham Asylum: slower than 7770

750 Ti
Bioshock Infinite: faster than 7850/GTX650 Ti Boost
Crysis 3: faster than 7870/GTX660
Far Cry 3: faster than 7850
Splinter Cell Blacklist: faster than 7870

Also strange things can be seen if you compare TPU perf/W charts: 750 Ti chart shows GTX 690 more efficient than 7750 compared to 750 Ti. But if you look at GTX 690 review (http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_690/29.html) you see 7750 being a lot more efficient than GTX 690.

 

[SiliconWars]

My main concern is actually the big difference between benchmarks across sites. On TPU the 750 Ti gets more fps than 7870 and GTX 660 in Crysis 3. On AT it's slower than R7 265 which itself is slower than GTX 660.

Other weird results on TPU:

7790/R7 260X
Batman Arkham Asylum: slower than 7770

750 Ti
Bioshock Infinite: faster than 7850/GTX650 Ti Boost
Crysis 3: faster than 7870/GTX660
Far Cry 3: faster than 7850
Splinter Cell Blacklist: faster than 7870

Also strange things can be seen if you compare TPU perf/W charts: 750 Ti chart shows GTX 690 more efficient than 7750 compared to 750 Ti. But if you look at GTX 690 review (http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_690/29.html) you see 7750 being a lot more efficient than GTX 690.

This is basically why you have to take TPU's numbers with a huge grain of salt. I knew there were plenty of other things that were whacked but couldn't be bothered looking for them all.

W1zzard really needs to get his act together on his benchmarking runs.

 

[parvadomus]

Can the 750s SLI? or has that functionality removed too?
Now can we summarize how the efficiency gains have been achieved:
- L2 cache increased enabling more shaders to operate on a narrow bus, with slow memory.
- No SLI logic.
- Only one GPC (GPC communication logic totally removed).
- FP64 even more anemic than before.

Am I missing something?

 

[USER8000]

Can the 750s SLI? or has that functionality removed too?
Now can we summarize how the efficiency gains have been achieved:
- L2 cache increased enabling more shaders to operate on a narrow bus, with slow memory.
- No SLI logic.
- Only one GPC (GPC communication logic totally removed).
- FP64 even more anemic than before.

Am I missing something?

One GPC also means no interconnect.

The GTX750TI cannot SLI.

 

[OCGuy]

Can the 750s SLI? or has that functionality removed too?
Now can we summarize how the efficiency gains have been achieved:
- L2 cache increased enabling more shaders to operate on a narrow bus, with slow memory.
- No SLI logic.
- Only one GPC (GPC communication logic totally removed).
- FP64 even more anemic than before.

Am I missing something?

That is what also allows the 1400mhz clocks too, right?

I don't even know the point of trying to go out on a limb against an industry wide understanding of what these cards are/can be.

 

[parvadomus]

That is what also allows the 1400mhz clocks too, right?

I don't even know the point of trying to go out on a limb against an industry wide understanding of what these cards are/can be.

We are trying to know how much of an improvement Maxwell really is.
It has 1350mhz memory.

 

[el etro]

W1zzard really needs to get his act together on his benchmarking runs.

He needs to do little reviews, reduce the game count by half.

 

[sontin]

We are trying to know how much of an improvement Maxwell really is.
It has 1350mhz memory.

It delivers twice the performance of the GK107 while using the same power on the same 28nm process.

I dont understand what else you need "to know".

 

[parvadomus]

It delivers twice the performance of the GK107 while using the same power on the same 28nm process.

I dont understand what else you need "to know".

Im talking about architectures not only about a cut down GPU die.

 

[JoeRambo]

Im talking about architectures not only about a cut down GPU die.

So your claim is that previuos 28nm stuff from Nvidia had SLI logic and FP64 units that were burning truckload of watts even when not used?

Really sad to see ppl hell bent on talking down what is impressive product from Nvidia. Nvidia has been focused on perf/power since Fermi fiasco days and it is really showing.

 

[f1sherman]

So your claim is that previuos 28nm stuff from Nvidia had SLI logic and FP64 units that were burning truckload of watts even when not used?

Really sad to see ppl hell bent on talking down what is impressive product from Nvidia. Nvidia has been focused on perf/power since Fermi fiasco days and it is really showing.

Indeed. Both Intel and Nvidia have realized that in TDP classified market perf/Watt = perf,
and if you have TDP to spare you can easily convert it to perf.

BTW... GTX 650/Ti have no SLI too.

Sadly or luckily - this prevents hideous configurations like HD 7750/7770 CrossFire and even Hybrid CrossFireX :biggrin:

 

[TeknoBug]

Here's what I gathered from a local store:

(picking the cheapest in Canadian $):
XFX HD7790 (1GB) $149
Zotac GTX750Ti (2GB) $159
MSI R7 260X (2GB) $159
MSI HD7850 (2GB) $214
MSI GTX660 (2GB) $219

Comparing to the 750Ti, The GTX660 pulls ahead in the Anandtech benchmarks, pretty much trades places with the 7850, Better than both the 7790 and T7 260X. I'd say the 750Ti is actually in a good place price-wise and still gives decent performance for a budget card compared to other low cost ones.

There is also an EVGA 750Ti with G-sync function for $174.

 

[f1sherman]

You'd still need G-Sync monitor ^^

http://www.geforce.com/hardware/technology/g-sync/supported-gpus

 

[coffeejunkee]

It delivers twice the performance of the GK107 while using the same power on the same 28nm process.

I dont understand what else you need "to know".

Ok, lets calculate. Numbers from AT Bench, Crysis 3.

power
750 Ti: 184W
640: 152W

fps
750 Ti: 60,6
640: 22

perf/w
750 Ti: 0.33
640: 0.14

Yes, 750 Ti has 228% efficiency of GT 640, so it actually has more than double perf/w.

But how about GTX 650? That's also GK107.

power
750 Ti: 184W
650: 176W

fps
750 Ti: 60,6
650: 38.8

perf/w
750 Ti: 0.33
650: 0.22

750 Ti has 149% efficiency of GTX 650, not quite the 200% we are looking for.

What this mostly shows is that GT 640 is just pretty inefficient.

How about the card it's closest to in perf, GTX 650 Ti Boost?

power
750 Ti: 184W
650 Ti Boost: 236W

fps
750 Ti: 60,6
650 Ti Boost: 72.9

perf/w
750 Ti: 0.33
650 Ti Boost: 0.31

750 Ti has 107% efficiency of GTX 650 Ti Boost.

Yes, the numbers are skewed because of the high power usage of the test system. But it's not going to turn the 149% against GTX 650 into 200%.

 

[parvadomus]

So your claim is that previuos 28nm stuff from Nvidia had SLI logic and FP64 units that were burning truckload of watts even when not used?

Really sad to see ppl hell bent on talking down what is impressive product from Nvidia. Nvidia has been focused on perf/power since Fermi fiasco days and it is really showing.

I didnt say gm107 was not impressive. A lot of people should stop with the stupid defensive attitude. You are NOT Nvidia.

Having said that I still hold that there are a lot of tiny tricks that contributes to the efficiency of this particular die.
The L2 cache might be the biggest contributor, and maybe even tweaks over the mature 28nm process. Maybe the same tweaks that helped bonaire remain efficient despite having too much shaders for a 128 bit bus.

 

[stahlhart]

So your claim is that previuos 28nm stuff from Nvidia had SLI logic and FP64 units that were burning truckload of watts even when not used?

Really sad to see ppl hell bent on talking down what is impressive product from Nvidia. Nvidia has been focused on perf/power since Fermi fiasco days and it is really showing.

I didnt say gm107 was not impressive. A lot of people should stop with the stupid defensive attitude. You are NOT Nvidia.

Having said that I still hold that there are a lot of tiny tricks that contributes to the efficiency of this particular die.
The L2 cache might be the biggest contributor, and maybe even tweaks over the mature 28nm process. Maybe the same tweaks that helped bonaire remain efficient despite having too much shaders for a 128 bit bus.

Get your tempers under control.
-- stahlhart

 

[Jodell88]

Is it safe to assume that Nvidia is going to be competitive with AMD when it comes to OpenCL performance?

 

[Will Robinson]

Assume nothing :sneaky:

 

[Cookie Monster]

Is it safe to assume that Nvidia is going to be competitive with AMD when it comes to OpenCL performance?

I dont think so. They want to push CUDA so i doubt this. Its quite sad because they probably could match AMD's performance or atleast have somewhat respectable performance except they don't choose to. The drivers are just not even touched for openCL.