Upcoming GTX 750 Ti tested

[SiliconWars]

I guess if AMD doesnt "exceed this" you can always claim they didnt even try. Maybe they just didnt feel like it or something.

AMD *will* matched or exceed this. Thing is, it's probably going to be on 20nm, not 28nm.

I am no expert but i am pretty sure it takes longer than a couple months to come up with a completely new architecture, much less bring it to market.

What do you think AMD's graphics engineers do? 6 months developing cards then 6 months holiday?

There is no other way they could beat the gm107 efficiency.

"There is no way AMD can beat Titan".

It would take a completely new amazing architecture that would be an engineering marvel. So yeah, i bet AMD could do this if they wanted. But it would take a lot longer than a couple months. The gm107 is in a league of its own.

Have you heard of Pirate Islands by any chance? What do you think AMD has been doing since Southern/Volcanic Islands?

Are you aware that with the true last generations (Kepler vs Southern Islands), AMD was 3 months ahead? Why wouldn't Pirate Islands be at least close to Maxwell's release time?

I am not saying AMD cannot make a better chip or they just arent capable, whether they can or not is another matter all by itself. But what i am saying is your statement is really really impossible. AMD cannot just pop out a more efficient design than maxwell in a couple months.......even if they wanted to. This is totally and completely unrealistic. Fallacy. I really dont even know why you would be saying something like this?

Because we have absolutely no idea how close to release Pirate Islands is. We have no idea if AMD has something else planned in the interim period.

What I'm saying is that *if* AMD has gone back to improve their low end in the way they improved with Hawaii, or *if* a Pirate Islands pipe-cleaner is due in a couple of months (remember the 4770?) *then* they will surely be capable of matching or exceeding the 750 Ti.

 

[SiliconWars]

I think getting near the performance of the previous arch at nearly half the power cost on the same process is a pretty big marvel. When this gets a shrink it could be close to 4 times the performance\watt of Kepler.

In the unlikely event of 20nm lasting 4 years it might get close to 3x, but nowhere near 4. People are going to be (unpleasantly) surprised when the new cards on an immature 20nm are compared to the last cards on a very mature 28nm. I really don't think the perf/Watt gap is going to be that big.

 

[Genx87]

In the unlikely event of 20nm lasting 4 years it might get close to 3x, but nowhere near 4. People are going to be (unpleasantly) surprised when the new cards on an immature 20nm are compared to the last cards on a very mature 28nm. I really don't think the perf/Watt gap is going to be that big.

Even if it got 3x that is a pretty significant increase in performance\watt. That said I expect them to get closer to their target of 4x than 3x.

 

[OCGuy]

In the unlikely event of 20nm lasting 4 years it might get close to 3x, but nowhere near 4. People are going to be (unpleasantly) surprised when the new cards on an immature 20nm are compared to the last cards on a very mature 28nm. I really don't think the perf/Watt gap is going to be that big.

Well that wouldn't be good for the 20nm power-hog Hawaii, that is for sure.

16nm FinFET is right behind 20nm, and TSMC lists expected efficiency and density gains.

Maxwell on 20/16 is going to be something to watch, that is for sure.

 

[Will Robinson]

The efficiency numbers for the 750Ti per mm2 do indeed look great.I hope AMD will get that good one day.
Gamers however buy a video card for performance and at a similar price to R7 265 it gets beaten fairly convincingly.
Remember your expensive 780 and Titan cards a few short months ago?
They had to slash the prices because of AMD's perf/dollar.
Where was perf/watt then hmm?

 

[USER8000]

Well that wouldn't be good for the 20nm power-hog Hawaii, that is for sure.

16nm FinFET is right behind 20nm, and TSMC lists expected efficiency and density gains.

Maxwell on 20/16 is going to be something to watch, that is for sure.

What roadmap do you see which says AMD is directly porting over AMD Hawaii to 20NM,or even using the same design for their next generation??

If you have some inside sources,can you please share the slides with us.

 

[OCGuy]

What roadmap do you see which says AMD is directly porting over AMD Hawaii to 20NM,or even using the same design for their next generation??

You think Hawaii was meant for 28 nm originally?

 

[SiliconWars]

You think Hawaii was meant for 28 nm originally?

Yes. I'm still not completely convinced that Pirate Islands will be 20nm yet...

 

[Zanovar]

Yes. I'm still not completely convinced that Pirate Islands will be 20nm yet...

:S

 

[Roland00Address]

I wonder why they're using Kepler in Tegra instead of switching to Maxwell?

Because Kepler was done 6 to 12 months ago and you can't just switch the gpu part of a soc when you already have a finalized mask of the soc. We now have working processors with Tegra K, and it will be about 6 months before it actually is in devices you can buy. Designing a soc takes 2 to 3 years from design to final product, 2 years if you throw money at it, 3 years for most companies. These 6 months will be for products that don't need a lot of verification (aka tablets) and will be even longer for products that need lots of verification such as phones.

We will see Maxwell in the successor to Tegra K, we will be getting a new tegra soc about every year from now on.

 

[OCGuy]

Yes. I'm still not completely convinced that Pirate Islands will be 20nm yet...

20nm was originally projected in 2013. AMD would have stuck with 28nm?

They couldn't design/produce a different chip quickly enough to just burst out another 28nm....

 

[SiliconWars]

20nm was originally projected in 2013. AMD would have stuck with 28nm?

They couldn't design/produce a different chip quickly enough to just burst out another 28nm....

Well, Maxwell was supposed to be 20nm also but Nvidia changed their minds at one point. AMD did this in the past with Cayman and will have known exactly when 20nm was going to be ready proper.

I doubt Hawaii would have existed at all had 20nm been on 2 and a bit years after 28nm instead of 3 and a bit, so basically Hawaii was something that AMD intended to bring the fight to Nvidia with at 28nm from the start. AMD knew a long time ago that 28nm was going to be a long-lasting node. It might not be done yet...

While I do believe Pirate Islands will be 20nm, I wouldn't say I was 100% sure of it. Neither am I sure that all the rest of the Maxwell chips will be. Maybe they'll do 1/2 28nm now and 1/2 and 20nm later. It's just too hard to say for sure these days.

 

[know of fence]

Very exciting new cards, but I'm surprised at the omission of display port. Doesn't Gsync require a display port connection? - Not a rhetorical question.
I'm completely confused regarding new interfaces and 4K or 144Hz. There will be a lot of incentive to upgrade your monitor in the near future. But will this also require a new graphics card?
4K HDMI 2.0@60Hz [-]
4K over DisplayPort [-]
Gsync [-]
4K DualLink DVI [?]

 

[tviceman]

At that price 750Ti is best suited to HPC or perhaps crypto mining,no serious gamer is going to value perf/watt over performance/$.

No "serious" is going to buy a $150 card anyways.

People with DELL, ACER, or ASUS off-the-shelf systems with half-way decent CPU's but lackluster PSU's will happily pay a $10 premium for this card if they want to play and have games run decently. Likewise, anyone with an SFF HTPC that values low power consumption AND/OR quiet operation is also going to want this card and, again, will gladly pay a $10 premium for it.

Use your head to figure these things out, not your biased opinions.

 

[Erenhardt]

No "serious" is going to buy a $150 card anyways.

People with DELL, ACER, or ASUS off-the-shelf systems with half-way decent CPU's but lackluster PSU's will happily pay a $10 premium for this card if they want to play and have games run decently. Likewise, anyone with an SFF HTPC that values low power consumption AND/OR quiet operation is also going to want this card and, again, will gladly pay a $10 premium for it.

Use your head to figure these things out, not your biased opinions.

I wonder if these people actually upgrade parts, or entire systems...
I think it is good for mobile. The need something efficient to compete with DualGraphics

 

[TeknoBug]

No "serious" is going to buy a $150 card anyways.

People with DELL, ACER, or ASUS off-the-shelf systems with half-way decent CPU's but lackluster PSU's will happily pay a $10 premium for this card if they want to play and have games run decently. Likewise, anyone with an SFF HTPC that values low power consumption AND/OR quiet operation is also going to want this card and, again, will gladly pay a $10 premium for it.

Use your head to figure these things out, not your biased opinions.

And it looks like it's adding an onramp to accessing PC gaming for the budget consumer, I imagine this would help Nvidia make sales to people that actually want to buy a video card for gaming and not mining like the hijacked pricing of AMD cards lately. I'd happily pay $150-170 for a 750Ti, having looked at benchmarks and videos on youtube, it seems to be capable of playing just about every game out today including BF4, Crysis 3, Titanfall and Metro Last Light.

Thankfully the R7 260X is still reasonably priced and is somewhat similar to the 750Ti in performance, so there's still a choice especially for those who want to take advantage of Mantle in Thief and future EA titles.

 

[USER8000]

And it looks like it's adding an onramp to accessing PC gaming for the budget consumer, I imagine this would help Nvidia make sales to people that actually want to buy a video card for gaming and not mining like the hijacked pricing of AMD cards lately. I'd happily pay $150-170 for a 750Ti, having looked at benchmarks and videos on youtube, it seems to be capable of playing just about every game out today including BF4, Crysis 3, Titanfall and Metro Last Light.

Thankfully the R7 260X is still reasonably priced and is somewhat similar to the 750Ti in performance, so there's still a choice especially for those who want to take advantage of Mantle in Thief and future EA titles.

It is not true in all markets though. In the UK,aftermarket GTX660 and R9 270 cards cost the same as the aftermarket GTX750TI cards,and there are deals to be had on and off on R9 270 cards and HD7850 cards which pushes them closer to the same price as normal GTX750TI cards. Its mainly the R9 280X which has been price inflated over here to some degree.

 

[f1sherman]

http://techreport.com/review/26050/nvidia-geforce-gtx-750-ti-maxwell-graphics-processor

http://www.3dcenter.org/artikel/launch-analyse-nvidia-geforce-gtx-750-750-ti

 

[OCGuy]

While I do believe Pirate Islands will be 20nm, I wouldn't say I was 100% sure of it. Neither am I sure that all the rest of the Maxwell chips will be. Maybe they'll do 1/2 28nm now and 1/2 and 20nm later. It's just too hard to say for sure these days.

Maxwell will absolutely become be 20 or 16. GM2XX then, GM1XX now.

The only comments floating around that are not at least partially excited about what the recent Maxwell launch means for 20nm seem to be wishful thinking. The pattern of AMD being on the forefront of a new node, as well as efficiency, looks to be coming to an end.

 

[USER8000]

Maxwell will absolutely become be 20 or 16. GM2XX then, GM1XX now.

The only comments floating around that are not at least partially excited about what the recent Maxwell launch means for 20nm seem to be wishful thinking. The pattern of AMD being on the forefront of a new node, as well as efficiency, looks to be coming to an end.

Where is this AMD roadmap you seem to be looking at though?? None of use know what the next AMD uarch is going to look like,but you seem to know already it won't be efficient and it will be delayed and not be on a new node.

Maybe you need to contact one of the major news sites,it would be a big coup for them - they might even pay you for the exclusive!!

Edit to post!!

Can you tell us when the Maxwell based GK104 and GK106 replacements are out?? I assume you are fairly certain as you have roadmaps for them. I want to replace my current GTX600 series card in my main desktop,so it would be useful to know when the replacements are due and how much they cost.

 

[coffeejunkee]

I calculated performance per Watt based on AT data for 750 Ti, 650 Ti, 260X and 265. AT only shows power consumption for Crysis 3 (and Furmark but I don't think it's representative). I used fps from Crysis 3 at medium settings.

Power

750 Ti: 184W
650 Ti: 191W
260X: 222W
265: 230W

FPS

750 Ti: 49,9
650 Ti: 45,1
260X: 43,2
265: 63,3

Perf/W

750 Ti: 0,27 (= 100%)
650 Ti: 0,24 (= 89%)
260X: 0,19 (= 70%)
265: 0,28 (= 104%)

 

[mindbomb]

you have to use graphics card power consumption rather than system power consumption, as system will dilute any difference between cards.

 

[SiliconWars]

you have to use graphics card power consumption rather than system power consumption, as system will dilute any difference between cards.

Yes but you can't use a graphics card without a CPU and the rest of a system.

It's fine to marvel at Maxwell's efficiency taken in a vacuum, or in mobile when we see it - but on the desktop the gains are largely swallowed up by the rest of the system.

Why would you pay $150 for this chip if the R7 265 is the same price? The 750 Ti loses in perf/W system wide - which is all that counts - and it loses in perf/$...and it loses in performance.

 

[Enigmoid]

I calculated performance per Watt based on AT data for 750 Ti, 650 Ti, 260X and 265. AT only shows power consumption for Crysis 3 (and Furmark but I don't think it's representative). I used fps from Crysis 3 at medium settings.

Power

750 Ti: 184W
650 Ti: 191W
260X: 222W
265: 230W

FPS

750 Ti: 49,9
650 Ti: 45,1
260X: 43,2
265: 63,3

Perf/W

750 Ti: 0,27 (= 100%)
650 Ti: 0,24 (= 89%)
260X: 0,19 (= 70%)
265: 0,28 (= 104%)

System power is a really poor measure. Especially when its an overclocked 4960X.

Crysis 3 is also a poor game for comparison (don't know why review sites use it so much) as it places a hard load on the CPU.

Edit:

@SiliconWars Its an overclocked IV-E. Exactly the wrong system to measure power consumption in.

 

[sontin]

Yes but you can't use a graphics card without a CPU and the rest of a system.

What an argument.

It's fine to marvel at Maxwell's efficiency taken in a vacuum, or in mobile when we see it - but on the desktop the gains are largely swallowed up by the rest of the system.

No, it's not. If you have no headroom for 50-100W more then the GTX750TI is the best option.

Why would you pay $150 for this chip if the R7 265 is the same price? The 750 Ti loses in perf/W system wide - which is all that counts - and it loses in perf/$...and it loses in performance.

Seriously? Going from 150W to 200W is still a 33% increase. The 265 only offers ~16% more performance.

@SiliconWars Its an overclocked IV-E. Exactly the wrong system to measure power consumption in.

Bingo. The plattform alone is using a huge amount of power. And nobody will combine a GTX750TI with a overclocked 6-Core CPU...