Upcoming GTX 750 Ti tested

[sontin]

I think what Silicon is trying to say is that yes, the card that was designed from the ground up to be power efficient is power efficient, however considering that is the main goal, it makes the achievement somewhat less impressive? AKA. either company could make a card that efficient if they wanted to. You don't get FPS/$, you get FPS/W... while AMD is arguably more focused on FPS/$

So, a 50% improvement over the competition is "less impressive"?

And the last part: Lol.
So AMD is not trying to develop a GPU which they can sell in the notebook market because they want to sell cards which need twice the power?
Woha.

 

[Enigmoid]

Bonaire isn't "inefficient" as such I suppose - but obviously a part like this is a bit top-heavy and with a 1 GHz clock it's approaching the area where AMD cards seem to fall off the cliff a little in efficiency terms. That's why the 7750 and 7850 showed better efficiency characteristics than the 7770 and 7870, even though the former were salvage parts.

An 800-900 MHz Bonaire would be more efficient again, I guess. Also even slower of course, but it would still narrow the efficiency gap more to this 750 Ti.

Obviously AMD would have some work to do to catch Maxwell, but a year is a long time on this node as we've seen previously with Hawaii. That's the important thing, the extra year on the process.

It's a shame but I don't believe we'll see any more 28nm chips from AMD so it will remain an interesting "what if".

Bonaire has a bigger die than the 750Ti and a 850 mhz bonaire would get trashed by the 750TI.

That makes maxwell impressive.

 

[Crap Daddy]

So, a 50% improvement over the competition is "less impressive"?

And the last part: Lol.
So AMD is not trying to develop a GPU which they can sell in the notebook market because they want to sell cards which need twice the power?
Woha.

Given AMDs latest market share results I'd say he's right.

 

[SiliconWars]

Bonaire has a bigger die than the 750Ti and a 850 mhz bonaire would get trashed by the 750TI.

As mentioned previously, the die sizes are basically equal.

That makes maxwell impressive.

For Nvidia yes it's an impressive chip as they historically don't compete so well at the lower end.

However vs this non-existent 900MHz full die Bonaire it would likely be 15%-20% faster with the same power draw. In other words, nowhere near 50% or 60%.

This is Nvidia's new arch which explicitly targets perf/W
Bonaire is almost a year old
Bonaire was not targeted for perf/W, but perf/$ like Sushi said

Ask yourself what the chances of AMD not *at least* matching the 750 Ti would be with Pirate Islands targeting perf/W. You think they can't make 20% up? Just look at Hawaii, and that's not even next gen.

 

[Keysplayr]

As mentioned previously, the die sizes are basically equal.

For Nvidia yes it's an impressive chip as they historically don't compete so well at the lower end.

However vs this non-existent 900MHz full die Bonaire it would likely be 15%-20% faster with the same power draw. In other words, nowhere near 50% or 60%.

This is Nvidia's new arch which explicitly targets perf/W
Bonaire is almost a year old
Bonaire was not targeted for perf/W, but perf/$ like Sushi said

Ask yourself what the chances of AMD not *at least* matching the 750 Ti would be with Pirate Islands targeting perf/W. You think they can't make 20% up? Just look at Hawaii, and that's not even next gen.

What is your definition of "basically equal" ? 1mm2? 2mm2? difference?

148mm2 for GM107 and 160mm2 for Bonaire is not basically equal.
GM107 is 210million transistors shy of bonaire as well.

 

[SiliconWars]

As mentioned previously, Bonaire has the audio DSP's which have been estimated at 10mm2 or thereabouts.

 

[Keysplayr]

As mentioned previously, Bonaire has the audio DSP's which have been estimated at 10mm2 or thereabouts.

I suppose if we give you enough rope, you'll get Bonaire down to 20mm2 if you hack off all of the transistors that aren't used 100% of the time. maybe you can do the same for GM107?

 

[CrimsonWolf]

comparing this card with the GTX 480 is quite funny
http://www.anandtech.com/bench/product/1135?vs=1130

I actually ordered EVGA's GTX 750 Ti yesterday to replace a GTX 470 for a PC in my living room. I only game on it occasionally and is mostly used for streaming. Probably a bit of a side-grade performance wise, but it'll be nice to get rid of that furnace of a video card before summer comes.

 

[mohit9206]

The 750 and the 750Ti are now the go to cards for small form factor gaming htpc's and Oem pc's and for anyone that wants a good performance card in the $100-150 range that does not require a generator to run.These two cards are the most impressed i have been with any budget cards since the radeon 7750.I might upgrade my current 7750 to a 750 ,i will double the performance but power consumption will stay the same !! Unbeliveable..

 

[SiliconWars]

I suppose if we give you enough rope, you'll get Bonaire down to 20mm2 if you hack off all of the transistors that aren't used 100% of the time. maybe you can do the same for GM107?

Are you saying, that those audio DSP's are directly contributing to the perf/mm2 of Bonaire?

Or in terms of perf/mm2 are they basically just "dead area"?

Regardless, I'll concentrate on what actually counts.

The 11 month gap between the chips
The fact that this is Nvidia's next gen
The fact that Maxwell is designed for efficiency
The fact that Bonaire was neither binned nor aiming for efficiency

You think all of that is less relevant than an 8% area advantage that is basically only there because of TrueAudio? I don't. Then again, I'll be one of the few here who isn't totally shocked if AMD does another Hawaii at the low end.

 

[ams23]

@ SiliconWars: the reality is that, per Tom's Hardware analysis, GTX 750 Ti has 38-76% (!) higher perf. per watt than it's nearest priced competitors (one which was launched only three months ago, one which has only been paper launched), all on the same 28nm fab. process. Like it or not, GM107 is unrivaled at this time in perf. per watt and perf. per mm^2, period. There is no existing GPU from either major IHV that can touch it in these two metrics. It is what it is, get over it already.

 

[Mondozei]

Bonaire has a bigger die than the 750Ti and a 850 mhz bonaire would get trashed by the 750TI.

That makes maxwell impressive.

No it makes it logical for a new architecture to be better than an older one.
Anything else would be a scandal.

 

[insertcarehere]

More than anything else, this card would be damn good for Nvidia in mobile form, considering that this has ~=GTX770m performance in a much smaller die and consumes less power even in desktop variant right now.

 

[Enigmoid]

More than anything else, this card would be damn good for Nvidia in mobile form, considering that this has ~=GTX770m performance in a much smaller die and consumes less power even in desktop variant right now.

Which is good. At least Nvidia still cares about mobile. AMD doesn't give a crap. Still waiting for Bonaire in mobile (good chip and strong competitor to the 765m and 770m). Almost 1 year and nothing yet.

No it makes it logical for a new architecture to be better than an older one.
Anything else would be a scandal.

Its a new architecture without a node shrink.

Are you saying, that those audio DSP's are directly contributing to the perf/mm2 of Bonaire?

Or in terms of perf/mm2 are they basically just "dead area"?

Regardless, I'll concentrate on what actually counts.

The 11 month gap between the chips
The fact that this is Nvidia's next gen
The fact that Maxwell is designed for efficiency
The fact that Bonaire was neither binned nor aiming for efficiency

You think all of that is less relevant than an 8% area advantage that is basically only there because of TrueAudio? I don't. Then again, I'll be one of the few here who isn't totally shocked if AMD does another Hawaii at the low end.

While the audio DSP does not contribute to the architectural efficiency it does increase the area and thus cost of the chip regardless. At present it has limited/no use but costs AMD to implement.

 

[Binky]

Galaxy announced a 750 ti low-profile. Would this be the new performance king for low-profile machines? I can't find one for sale.

 

[f1sherman]

Which is good. At least Nvidia still cares about mobile. AMD doesn't give a crap. Still waiting for Bonaire in mobile (good chip and strong competitor to the 765m and 770m). Almost 1 year and nothing yet.

Bonaire had been excellent chip when it arrived.
AMD had a winner back then.
But it turned out to be far too underutilized, and the fact there is no mobile version is a crime.
How the hell did that happen? Wrong timing... huh

Still very decent chip but hopelessly out-gunned now that GM107 is out. Whatta waste...

 

[SiliconWars]

While the audio DSP does not contribute to the architectural efficiency it does increase the area and thus cost of the chip regardless. At present it has limited/no use but costs AMD to implement.

Yes but any additional cost Bonaire has to manufacture due to die size has been gained back and then some due to the extra 11 months of manufacturing. Remember too that there was only one Bonaire SKU at the start, so all the good Bonaire chips were maximum price - this isn't the same with GM107 which has the Ti and non Ti version.

This is part of what sushi meant when talking about AMD going for perf/$ instead of perf/W. You can make more "working" chips with more variability, and that's exactly what AMD has done with Bonaire. Nvidia has split GM107 by having a clear good part and one "not so good", right from the start. For all we know the Ti version might only be 20-30% of all the working chips coming off the wafer. Even if it's 80%, you can still see how having 20% of the lesser version is a drawback compared to AMD having one "Bonaire XT" SKU commanding full price for 3 quarters of a year.

Even now Bonaire probably has more revenue per wafer than GM107 (at $120 prices vs GM107's $150) - even with GM107's smaller size. A year ago Bonaire was one SKU with 896 shaders at 1GHz. Now it has 2 SKU's - one with 896 shaders at 1100MHz and one with 768 shaders at 1GHz. In other words, they are definitely getting more useable chips out of the same wafers.

Now if you consider the R7 265 at $150 - Pitcairn has been around for 3 years and this is basically the worst part that they throw out now. Yield must be as close to maximum as you're gonna get - and it's a salvage part of a much more expensive chip.

The point is, time to market counts way, way more than die size here. If AMD chooses to fight Maxwell with old, cut down chips that aren't *that* much bigger then they are clearly able to do so.

Really, the 750 Ti is a nice card but the price is completely out of whack...cuz Nvidia. It should be $100, putting huge pressure on whatever AMD has down there now. Problem is they would then have to drop the price of the 640.

If the 265 does find itself at $150 at retail, you can be sure that the 750 Ti will drop by at least $20.

 

[SiliconWars]

Galaxy announced a 750 ti low-profile. Would this be the new performance king for low-profile machines? I can't find one for sale.

This low-profile 7850 would still be faster, if you can find one - http://www.kitguru.net/components/graphic-cards/zardon/afox-hd7850-single-slot-crossfire-review/

Clearly the 750 Ti will be far ahead in perf/Watt and is the obvious choice unless it's a lot more expensive.

 

[TeknoBug]

http://www.youtube.com/watch?v=2MH50WGxFvY

 

[Arkadrel]

http://www.youtube.com/watch?v=2MH50WGxFvY

Not bad, watched a bit and dont think I saw the FPS drop below 43 at any point, and usually it was over 50.

Not at all bad for a 750ti @1350mhz (running 1920x1080 and mostly Ultra settings)


Its pretty crazy what can be done within the TDP this card has.
Im really looking forwards to 20nm, and again with 16nm+ FinFETs.

Im actually wondering if we ll reach the point of deminishing returns like with CPUs.

Will there be a day soon where the hardware is so much ahead of the software (games),
that even low powered GPUs will be enough for everyone?

 

[Rezist]

Prices for the 750ti in canada range from 170 to 190$. I hope it drops but it probably won't.

 

[Enigmoid]

Yes but any additional cost Bonaire has to manufacture due to die size has been gained back and then some due to the extra 11 months of manufacturing. Remember too that there was only one Bonaire SKU at the start, so all the good Bonaire chips were maximum price - this isn't the same with GM107 which has the Ti and non Ti version.

This is part of what sushi meant when talking about AMD going for perf/$ instead of perf/W. You can make more "working" chips with more variability, and that's exactly what AMD has done with Bonaire. Nvidia has split GM107 by having a clear good part and one "not so good", right from the start. For all we know the Ti version might only be 20-30% of all the working chips coming off the wafer. Even if it's 80%, you can still see how having 20% of the lesser version is a drawback compared to AMD having one "Bonaire XT" SKU commanding full price for 3 quarters of a year.

Even now Bonaire probably has more revenue per wafer than GM107 (at $120 prices vs GM107's $150) - even with GM107's smaller size. A year ago Bonaire was one SKU with 896 shaders at 1GHz. Now it has 2 SKU's - one with 896 shaders at 1100MHz and one with 768 shaders at 1GHz. In other words, they are definitely getting more useable chips out of the same wafers.

Now if you consider the R7 265 at $150 - Pitcairn has been around for 3 years and this is basically the worst part that they throw out now. Yield must be as close to maximum as you're gonna get - and it's a salvage part of a much more expensive chip.

The point is, time to market counts way, way more than die size here. If AMD chooses to fight Maxwell with old, cut down chips that aren't *that* much bigger then they are clearly able to do so.

Really, the 750 Ti is a nice card but the price is completely out of whack...cuz Nvidia. It should be $100, putting huge pressure on whatever AMD has down there now. Problem is they would then have to drop the price of the 640.

If the 265 does find itself at $150 at retail, you can be sure that the 750 Ti will drop by at least $20.

I agree completely that the price is steep. The 265 may be hit a little by mining prices.

I would expect comparatively lower binned chips from Nvidia (on the desktop vs bonaire) because the bulk of GM 107 and especially the good chips are probably shipping to OEMs for mobile which AMD did not do with Bonaire.

Nvidia can also get higher margin mobile prices which AMD does not with Bonaire.

If anything I'd say that AMD did not go with a cut down bonaire because it would compete too much with the 7770 and for a newer chip prices are generally higher, dropping off with time. Nvidia uses GK106 for the 650 TI and 650 TI boost. The primary reason for Nvidia to launch the 750/750TI is that they can stop disabling so many working GK106 chips with much larger die sizes.

 

[SiliconWars]

I dunno wtf AMD is doing in mobile (and by most accounts they don't either) so you got me there.

 

[xpea]

I dunno wtf AMD is doing in mobile (and by most accounts they don't either) so you got me there.

well AMD tries hard but, and it's a big BUT, Enduro is a joke compared to Optimus, so very few OEMs pick up AMD...

 

[Binky]

This low-profile 7850 would still be faster, if you can find one - http://www.kitguru.net/components/graphic-cards/zardon/afox-hd7850-single-slot-crossfire-review/

Clearly the 750 Ti will be far ahead in perf/Watt and is the obvious choice unless it's a lot more expensive.

That link is a single slot (but not low profile) 7850. I've seen rumors of a low profile 7850 but never actually seen one for sale in the US.