GeForce GTX 750 Ti Review: Maxwell Adds Performance Using Less Power

[RussK1]

Other than finding efficiency it all leaves me meh!... Sure it looks good against its GK106 cousin but, that not saying a lot as the GK106 wasn’t sincerely efficient or great performance. While this Maxwell has it better in efficiency it just doesn’t raise the bar in performance of a $150 mainstream offering. It feels like the Nvidia had more of an edict to keep it under 75W, than making a solid mainstream gaming offering.

Maybe that was the point?

I believe low wattage was indeed the point. I find it hard to consider a card drawing only 60W but with HD7790 performance "disappointing," particularly as long as the price is right. This is a solid mainstream card.

Both right... This is a perfect card for non-hardcore gamers. Although I think $150 is still too much. If it was priced closer to $100 it would shake things up quite a bit.

A $100 would mean it would be the best entry/mid value card on any book.

Definitely! Although nVidia has no incentive to do so. AMD has nothing to compete with performance per watt in it's class.

 

[Casecutter]

Other than finding efficiency it all leaves me meh!... Sure it looks good against its GK106 cousin but, that not saying a lot as the GK106 wasn’t sincerely efficient or great performance. While this Maxwell has it better in efficiency it just doesn’t raise the bar in performance of a $150 mainstream offering. It feels like the Nvidia had more of an edict to keep it under 75W, than making a solid mainstream gaming offering.

Maybe that was the point?

I believe low wattage was indeed the point. I find it hard to consider a card drawing only 60W but with HD7790 performance "disappointing," particularly as long as the price is right. This is a solid mainstream card.

Both right... This is a perfect card for non-hardcore gamers. Although I think $150 is still too much. If it was priced closer to $100 it would shake things up quite a bit.

A $100 would mean it would be the best entry/mid value card on any book.

Definitely! Although nVidia has no incentive to do so. AMD has nothing to compete with performance per watt in it's class.

Sorry, to get you all so riled up. My point is, a card that for most truthfully means an AIB OC which are now more like $160-170. That’s pricey for really only able to provide medium settings/no AA in most titles at 1080p, slightly bettering a R7 260X. This is a really nice entry card for a novice that buy’s a stock box i5, while a little short on grunt to be mainstream it a plug-n-play. Today with 1920x1080 the norm for most displays having to ante-up that much cash should more often say at least some AA more often. If your building your own low priced gaming machine, any good 400+w PSU is already on your list, then this come in a little short for such a build, the GTX 660 is a better buy.

 

[smeezekitty]

Other than finding efficiency it all leaves me meh!... Sure it looks good against its GK106 cousin but, that not saying a lot as the GK106 wasn’t sincerely efficient or great performance. While this Maxwell has it better in efficiency it just doesn’t raise the bar in performance of a $150 mainstream offering. It feels like the Nvidia had more of an edict to keep it under 75W, than making a solid mainstream gaming offering.

I think this is a pretty big deal because before NVidia was notably lacking in decent low power cards before

 

[Novuake]

Definitely! Although nVidia has no incentive to do so. AMD has nothing to compete with performance per watt in it's class.

This power envelope was dominated by AMD for some time.

There are always the back and forth in this segment.

 

[logainofhades]

Nvidia and AMD both have had the crown for performance per watt at different times. Nvidia seems to be moving in that direction with Maxwell. Nvidia did this one right. The OEM crowd has a decent performing GPU, that will not require a PSU upgrade, and don't have to worry about card length issues either. As soon as they can make some low profile versions, they will dominate the OEM bracket. AMD needs something to answer if they care at all about OEM PC owner marketshare.

 

[lowguppy]

AMD's 7750 was by far the top low profile/no aux power video card for something like 18 months (an eternity in GPU markets) and nVidia wasn't even close. This card is a 40-50% performance improvement over the 7750, completely trumping it. All that without a node shrink is very impressive. I can't wait to see what they can do once their process does shrink this fall.

If anyone hasn't noticed, there's a pretty big push towards small form factor and quiet computing. Low wattage GPUs are crucial to making that segment true gaming PCs and not just HTPCs.

 

[Onus]

If/when a single-slot, low-profile version of this card becomes available, I'll be looking to buy it for my mini-ITX box. Until then, the HD7750 it has remains the most powerful card that will fit.
Considering the plethora of games available on Steam that aren't all that demanding, that lowly HD7750 is sufficient for great settings even at 1920x1080. The most demanding game I play is Guild Wars 2, and it looks quite nice on that card; the GTX750Ti would just make it a lot smoother, and maybe let me increase view distance.

 

[logainofhades]

I still would like 3 of the 750ti to replace my HD 5850 in my file server, the 5850 in my FX 8320 @ 4.0ghz rig, and my FX 8320 rig that is currently sitting with a GTS 450 than I am testing out for a friend. Normally that rig has an HD 4870.

 

[RedJaron]

I really think AMD botched the sub-$100 market with the R7 250. The 250 gives you inferior performance compared to the 7750, but you don't get a price drop. Yes, the 750 Ti is now the performance king, but you're also paying $160 for it. I think there's still a market for people that want a decent mainstream gaming GPU for $100 or less that doesn't require a power cable. I mean, they could release a 255 that's a rebranded 7750, but that makes things confusing with current 250 pricing. And who would want a 250 when you can get a 255 for $5-$10 more?

 

[Tumn1s]

Whoever setup the mining benchmarks for this article doesn't have a clue what they're doing... I get 450kh/s out of my R9 270x on stock clocks. Learn how to configure CGminer, thanks.

 

[RussK1]

Whoever setup the mining benchmarks for this article doesn't have a clue what they're doing... I get 450kh/s out of my R9 270x on stock clocks. Learn how to configure CGminer, thanks.

I've read a lot of people "claiming 450kH/s" from a card with 1,280 shaders but yet to see proof. Until then, it's bs. Like a lot of people on the Litecoin Wiki page. No validations and for some reason want to lie about it.

It's a 300-350kH/s card with matching WU's or close to it (.9). Getting 450kH/s and -/+300 WU's means the hashrate is junk.

I can see people getting 400+kH/s with the 7870XT with the Tahiti LE chip but not Pitcairn. Needless to say, every card is different from the components used to it's firmware. A simple Google will easily show that most with Pitcairn GPU's with 1,280 shaders average hashrate is right around 300kH/s.

 

[Novuake]

If/when a single-slot, low-profile version of this card becomes available, I'll be looking to buy it for my mini-ITX box. Until then, the HD7750 it has remains the most powerful card that will fit.
Considering the plethora of games available on Steam that aren't all that demanding, that lowly HD7750 is sufficient for great settings even at 1920x1080. The most demanding game I play is Guild Wars 2, and it looks quite nice on that card; the GTX750Ti would just make it a lot smoother, and maybe let me increase view distance.

I am assuming you are not on Aurora Glade?

 

[helltrager]

So a question I've been wanting to ask for ages. Why are they deliberately hobbling the cards when it comes to memory transfer? My GTX260 had 448mb's and now my GTX670 Ti OC has 128mb's. Isn't it true that the games industry has hit the time/profit ceiling about 5-6 years ago, but the card companies want you to think you need better cards to run what you can run on Ultra with a 260? Not to mention the whole SLi scam. Two cards at half the PCI-e multiplier? And they have to communicate which slows performance. Something I've tested. Why can't I get an affordable card like a GTX670 Ti with 448mb's memory trnasfer. It's not like it can't be done. I only run one screen though, I understand that Sli and Crossfire, along with top price cards will run three screens for that surround look, but who the hell really needs it? I still play brand new games on Ultra with my shitty GTX670 TI OC. And the frame rate it's perfectly fine. I guess it's what you don't know, rather than what you know. Like the $400 HDMI cables that came out when Blu-ray players did, haha.

 

[Someone Somewhere]

They never made a 670Ti...

There's been substantial improvements. Look at the fact that a 6990 is about equal with a R9 280, for example.

Also, there's been some testing on the PCIe thing. There's no bottleneck from the interface till you get down to about PCIe2.0x4, by memory.

 

[helltrager]

They never made a 670Ti...

There's been substantial improvements. Look at the fact that a 6990 is about equal with a R9 280, for example.

Also, there's been some testing on the PCIe thing. There's no bottleneck from the interface till you get down to about PCIe2.0x4, by memory.

Thanks for clearing that up, you're right about my card, it's 660 Ti OC, I never payed much notice considering I neither noticed any performance increases. I only bought it because my father inlaw wanted to buy me something for Christmas. So I just lumped it in with an Asus Phoebus. You can notice the difference with that though.
What's an R9 280? I could look it up but I'm about to go out, is it a GeForce or an ATI? Or is there a new kid on the block.
About the memory transfer, I still don't see why they can't just leave them at higher bandwidths. It's not like it costs them anything. The whole card is probably only 5$ to have made in China.
And the games ceiling comment still stands. I haven't come across a game I can't play on ultra at 1920 x 1080.
Anyway, all the best, since Win 8 came out I've lost all interest in PC's anyway. I've switched to consoles and Linux.

 

[Someone Somewhere]

R9 280 is a rebranded 7950.

I'm not sure about your units - they completely don't make sense.

Lots of people play on higher resolutions, or want framerates about 60Hz.

I'll have to have a look, but I think it's a LOT more than that. Even just the metal in the card is probably worth more than $5 at the moment. Semiconductor fabs are multi-billion dollar facilities, and these GPUs are at the bleeding edge of manufacturing techniques.

The 260 had a 448-bit interface, because it used DDR3 and was a fairly high-end cards. It had a bandwidth of 111.9GB/s. The GTX760 is 192.2GB/s, and the naming scheme has changed somewhat - the 260 was the second fastest nVidia card of the generation, but the 760 is #4 (not counting the Titan/Titan Black), and there may be more.

 

[smeezekitty]

About the memory transfer, I still don't see why they can't just leave them at higher bandwidths. It's not like it costs them anything. The whole card is probably only 5$ to have made in China.

NVidia loves deliberately gimping their product in various ways. No idea why

And the games ceiling comment still stands. I haven't come across a game I can't play on ultra at 1920 x 1080.
Anyway, all the best, since Win 8 came out I've lost all interest in PC's anyway. I've switched to consoles and Linux.

No one says you have to use 8

 

[helltrager]

R9 280 is a rebranded 7950.

I'm not sure about your units - they completely don't make sense.

Lots of people play on higher resolutions, or want framerates about 60Hz.

I'll have to have a look, but I think it's a LOT more than that. Even just the metal in the card is probably worth more than $5 at the moment. Semiconductor fabs are multi-billion dollar facilities, and these GPUs are at the bleeding edge of manufacturing techniques.

The 260 had a 448-bit interface, because it used DDR3 and was a fairly high-end cards. It had a bandwidth of 111.9GB/s. The GTX760 is 192.2GB/s, and the naming scheme has changed somewhat - the 260 was the second fastest nVidia card of the generation, but the 760 is #4 (not counting the Titan/Titan Black), and there may be more.

You don't need to check my units, hehe, I was being facetious. Have a great day buddy.

 

[knowom]

Might have to get one of these in the future to go with my undervolted & underclocked 0.752 V E7200 still using IGP with it finally a low wattage GPU that really performs.

As i said few times above, do not rush. This is new tech that is being sold crazy expensive seriously. Wait 2 months and see new lineups with better prices appear.

Might have to get one of these in the future to go with my undervolted & underclocked 0.752 V E7200 still using IGP with it finally a low wattage GPU that really performs.

That would be a waste. A STOCK E7200 will bottleneck it in some cases. Underclocked that far would be totally pointless

As in Nvidia architectures prior, ROP partitions and L2 cache slices are aligned. And like the GeForce GTX 650 Ti’s GK106 processor, GM107 sports two partitions with eight units each, giving you up to 16 32-bit integer pixels per clock. Where the two GPUs really diverge is their L2 cache capacity. In GK106, you were looking at 128 KB per slice, adding up to 256 KB in an implementation with two ROP partitions. GM107 appears to wield 1 MB per slice, yielding 2 MB of memory used for servicing load, store, and texture requests. According to Nvidia, this translates to a significant load shifted away from the external memory system, along with notable power savings.

Going easy on memory bandwidth is smart, since GM107 exposes a pair of 64-bit memory controllers to which 1 or 2 GB of 1350 MHz GDDR5 DRAM is attached. Peak throughput is, interestingly, exactly what we got from GeForce GTX 650 Ti: 86.4 GB/s. The memory is feeding fewer CUDA cores, but they’re managed more efficiently. So, the big L2 is supposed to play an instrumental role in preventing a bottleneck.

That is just all fine and dandy but that doesn't change the fact a 1 GHz CPU doesn't have chance with keeping up with even a low end card

Might have to get one of these in the future to go with my undervolted & underclocked 0.752 V E7200 still using IGP with it finally a low wattage GPU that really performs.

That would be a waste. A STOCK E7200 will bottleneck it in some cases. Underclocked that far would be totally pointless

As in Nvidia architectures prior, ROP partitions and L2 cache slices are aligned. And like the GeForce GTX 650 Ti’s GK106 processor, GM107 sports two partitions with eight units each, giving you up to 16 32-bit integer pixels per clock. Where the two GPUs really diverge is their L2 cache capacity. In GK106, you were looking at 128 KB per slice, adding up to 256 KB in an implementation with two ROP partitions. GM107 appears to wield 1 MB per slice, yielding 2 MB of memory used for servicing load, store, and texture requests. According to Nvidia, this translates to a significant load shifted away from the external memory system, along with notable power savings.

Going easy on memory bandwidth is smart, since GM107 exposes a pair of 64-bit memory controllers to which 1 or 2 GB of 1350 MHz GDDR5 DRAM is attached. Peak throughput is, interestingly, exactly what we got from GeForce GTX 650 Ti: 86.4 GB/s. The memory is feeding fewer CUDA cores, but they’re managed more efficiently. So, the big L2 is supposed to play an instrumental role in preventing a bottleneck.

That is just all fine and dandy but that doesn't change the fact a 1 GHz CPU doesn't have chance with keeping up with even a low end card

wait what!!!! 1ghz, dude core 2 DUO E7200 is not 1GHZ. Go look at specs before saying those random stuff.

Note where it says underclocked and undervolted, and the CPU-Z validator saying 995MHz.

http://www.hardwaresecrets.com/fullimage.php?image=10725
According to benchmarks this CPU does not bottleneck a HD 7770 i don't see why it would bottleneck this card. But i'd rather let that up to CPU experts. But i am pretty confident in what i was saying, it will most certainly not bottleneck the 750Ti. Well in his case he is doing the undervolting and underclocking, however its his own setup, talking about a normal setup there should be no problem. But btw why you undervolt and underclock ur cpu?

It's a 100% fanless 0DB passive cooled DAW that's just for midi sequencing and recording it doesn't need a powerful CPU just a cool quiet and energy efficient one. For every day use I'll probably have it clocked more reasonably between 1.2GHz to 1.6GHz.

The CPU-Z was just testing it's threshold limits and for the banner. I actually wired up my PC cases reset button to the bios jumpers because I got sick of resetting it trying to figure out the sweet spot for post or hang up with the underclock same overclocking hassles apply.

 

[knowom]

Might have to get one of these in the future to go with my undervolted & underclocked 0.752 V E7200 still using IGP with it finally a low wattage GPU that really performs.

That would be a waste. A STOCK E7200 will bottleneck it in some cases. Underclocked that far would be totally pointless

As in Nvidia architectures prior, ROP partitions and L2 cache slices are aligned. And like the GeForce GTX 650 Ti’s GK106 processor, GM107 sports two partitions with eight units each, giving you up to 16 32-bit integer pixels per clock. Where the two GPUs really diverge is their L2 cache capacity. In GK106, you were looking at 128 KB per slice, adding up to 256 KB in an implementation with two ROP partitions. GM107 appears to wield 1 MB per slice, yielding 2 MB of memory used for servicing load, store, and texture requests. According to Nvidia, this translates to a significant load shifted away from the external memory system, along with notable power savings.

Going easy on memory bandwidth is smart, since GM107 exposes a pair of 64-bit memory controllers to which 1 or 2 GB of 1350 MHz GDDR5 DRAM is attached. Peak throughput is, interestingly, exactly what we got from GeForce GTX 650 Ti: 86.4 GB/s. The memory is feeding fewer CUDA cores, but they’re managed more efficiently. So, the big L2 is supposed to play an instrumental role in preventing a bottleneck.

That is just all fine and dandy but that doesn't change the fact a 1 GHz CPU doesn't have chance with keeping up with even a low end card

wait what!!!! 1ghz, dude core 2 DUO E7200 is not 1GHZ. Go look at specs before saying those random stuff.

Note where it says underclocked and undervolted, and the CPU-Z validator saying 995MHz.

http://www.hardwaresecrets.com/fullimage.php?image=10725
According to benchmarks this CPU does not bottleneck a HD 7770 i don't see why it would bottleneck this card. But i'd rather let that up to CPU experts. But i am pretty confident in what i was saying, it will most certainly not bottleneck the 750Ti.

If it were running stock clocks, it would be marginal to OK. However, if you look at the CPU-Z image, you'll see it's at 995MHz. That's not stock clocks.

Even in underclocked it would still only bottleneck it circumstantially to some extent really anyway and I'm clearly aiming more for low heat and low wattage performance efficiency.

Honestly I'll probably wait on a die shrink and or fanless solution, but this generation of GPU's is one to look out for and consider it actually appears compelling enough to consider the upgrade from a 8800GT or 260GTX even though the performance isn't really much different the efficiency and heat output is night and day different.

 

[knowom]

They never made a 670Ti...

There's been substantial improvements. Look at the fact that a 6990 is about equal with a R9 280, for example.

Also, there's been some testing on the PCIe thing. There's no bottleneck from the interface till you get down to about PCIe2.0x4, by memory.

Thanks for clearing that up, you're right about my card, it's 660 Ti OC, I never payed much notice considering I neither noticed any performance increases. I only bought it because my father inlaw wanted to buy me something for Christmas. So I just lumped it in with an Asus Phoebus. You can notice the difference with that though.
What's an R9 280? I could look it up but I'm about to go out, is it a GeForce or an ATI? Or is there a new kid on the block.
About the memory transfer, I still don't see why they can't just leave them at higher bandwidths. It's not like it costs them anything. The whole card is probably only 5$ to have made in China.
And the games ceiling comment still stands. I haven't come across a game I can't play on ultra at 1920 x 1080.
Anyway, all the best, since Win 8 came out I've lost all interest in PC's anyway. I've switched to consoles and Linux.

1080P is on the verge of being replaced by 4K probably in 3 to 4 years they'll start to become much more standard. That of course will certainly change GPU requirements a fair amount. I do agree with you for the record though about memory bit rates being crippled deliberately by both Nvidia and AMD.

I'm not sure if maybe power requirements plays a role in that or not. but back around when the 8800GT series launched I was expecting high end GPU's to transition in the direction of 1024 memory bit rates, but instead kind of trended back in the other direction entirely gradually getting smaller each generation.

 

[knowom]

About the memory transfer, I still don't see why they can't just leave them at higher bandwidths. It's not like it costs them anything. The whole card is probably only 5$ to have made in China.

NVidia loves deliberately gimping their product in various ways. No idea why

And the games ceiling comment still stands. I haven't come across a game I can't play on ultra at 1920 x 1080.
Anyway, all the best, since Win 8 came out I've lost all interest in PC's anyway. I've switched to consoles and Linux.

No one says you have to use 8

No, but Tom's Hardware keeps not so gently prodding XP users to engage the kill switch at Microsoft's mercy not that it's doing much good they still haven't developed a proper replacement for it.

The best they've come up with is windows 7 which is far from perfect and really at best what I'd deem a middle of the road compromise if anything.

 

[Someone Somewhere]

I'm curious as to what you prefer about XP over 7. I don't see an issue, unless you're dealing with seriously legacy programs.

 

[Novuake]

No, but Tom's Hardware keeps not so gently prodding XP users to engage the kill switch at Microsoft's mercy not that it's doing much good they still haven't developed a proper replacement for it.

The best they've come up with is windows 7 which is far from perfect and really at best what I'd deem a middle of the road compromise if anything.

I SERIOUSLY disagree. What is wrong with Win7 compared to XP?

 

[smeezekitty]

I keep an XP machine around for very old software. But 7 is quite good. Even Vista is tolerable with decent hardware

Really no reason to keep a main machine on XP at this point. It is over 12 years old.
About like using Win 95 in 2007!