Geforce 750 Ti - Maxwell!?

[Mondozei]

Sweclockers, a highly reputed Swedish site which never publishes anything unless they are highly confident about something, has just posted a news story about Nvidia's new Maxwell series.

According to their sources, the first card with a new architecture will be in the 700-series, not the 800-series. And it will be the 750 Ti. It will replace the 650 Ti Boost.

The 750 Ti will have the Denver CPU, so it will be the first time we'll see how the 64-bit ARM chip performs(mostly GPGPU-related tasks), since it the same Denver CPU will come to Tegra K1 only until this fall.

In addition, Sweclockers more or less say that it will be on 28 nm. And it will be released in the middle of februari - not end of march which many thought would be the starting point for Maxwell's new desktop computers(the mobile GPU segment of maxwell, which includes rebrands, is coming out more or less now with the new MSI laptops).

So it seems that Nvidia has pushed forward the schedule after the leaks from Clover about their 800M-series for laptops.

The question remains about the middle range 800-series cards, will they come a few weeks later or a full month or two later?

In any case, if this is accurate, and Sweclockers is a very careful site(although as always, no site is perfect), then we'll be able to analyze Maxwell for desktop a lot earlier than we thought, just a 4 weeks from now instead of 10 weeks.

 

[Ancalagon44]

Why on earth would a desktop GPU include a Denver CPU?

 

[Mondozei]

Why on earth would a desktop GPU include a Denver CPU?

Far from a majority of people use GPGPU, but the need is there, especially for science.
Not everything has to be done in supercomputers and you can pool cheaper GPU's together if you have a tight budget for smaller datasets.

I always smile when I see people who can't understand that people have varying needs and not everyone is the same.

 

[46andtool]

if this 750 ti maxwell card is built on revised architecture than you would think that they would introduce it using 800 series nomenclature. I cant imagine it would be much slower than a 760 gtx!

 

[Spjut]

It would be awesome if future games could offload work to the card's CPU

 

[sushiwarrior]

It performs pretty impressively, but for some reason the 750 and 750Ti have a 6 pin connector, I thought they would use less power than that.

 

[ams23]

Tom's Hardware hinted something along these lines 10 days ago:

http://www.tomshardware.com/reviews/tegra-k1-kepler-project-denver,3718-3.html

Moving forward, every GPU architecture Nvidia develops will be mobile-first. That was a decision management made during Kepler’s design and then applied to Maxwell from the start. It doesn’t mean Maxwell will show up in Tegra first (in fact, the first Maxwell-powered discrete GPUs are expected in the next few weeks). But the architecture was approached with its mobile configuration and power characteristics in mind, scaling up from there. Sounds like a gamble for a company so reliant on the success of its big GPUs, right? Nvidia says that the principles applied to getting mobile right will be what help it maximize the efficiency of its discrete products moving forward—and we’ll have the hardware to put those claims to the test once GeForce GTX 750 materializes.

 

[Skurge]

Well, I saw that these were coming soon and they were on the spec sheet, but I didn't bother looking at the core. I just figured it was just a cut down 760 so I didn't take a second look.

AIBs tell me it will be slower than a 650Ti Boost, about as fast as a 260X.

 

[sushiwarrior]

Well, I saw that these were coming soon and they were on the spec sheet, but I didn't bother looking at the core. I just figured it was just a cut down 760 so I didn't take a second look.

AIBs tell me it will be slower than a 650Ti Boost, about as fast as a 260X.

Somewhere between a 260x and a 270 is spot on.

They apparently use the Denver CPU for memory compression (DCC?), so even though it's a 128-bit it does surprisingly well at high resolutions, especially versus other Nvidia cards.

 

[Silverforce11]

Targeting mobiles first is the right approach because NV has so many mobile design wins for several generations already, they pretty much dominate that market. This is what really hurt AMD, not desktop Fermi or Kepler, since they've maintained a good marketshare on desktop, but tanked so hard on mobile.

 

[f1sherman]

:hmm:

remember this thread?

Somewhere between a 260x and a 270 is spot on.

They apparently use the Denver CPU for memory compression (DCC?), so even though it's a 128-bit it does surprisingly well at high resolutions, especially versus other Nvidia cards.

Denver on low Maxwell parts? CAN'T BE. Is Mantle that good :biggrin:
Please tell moar

 

[sushiwarrior]

:hmm:

remember this thread?

Denver on low Maxwell parts? CAN'T BE. Is Mantle that good :biggrin:
Please tell moar

Not a whole lot else to say, I'm not 100% clear on how Denver is used but the big thing I know is that memory compression.

I couldn't give exact numbers for mantle, but low end CPU's should see 30-45%, higher end stuff will see more like 15-25%? Total ballparks here, but the story is low end CPU's gain the most.

 

[f1sherman]

Nice. That's more than whole Mantle thread + Denver bonus

 

[24601]

If it were just memory compression I would be disappointed.

I was hoping for CPU driver load offloading.

Would explain how Nvidia would deal with the GDDR throughput stagnation though.

 

[Imouto]

Would explain how Nvidia would deal with the GDDR throughput stagnation though.

I bet they're thinking more about how compressing the memory will allow their cards to mount way lower amounts of VRAM. Memory compression allows massive physical RAM savings.

But this is a huge win for Nvidia as you say, the memory throughput will be insane.

 

[Keysplayr]

It performs pretty impressively, but for some reason the 750 and 750Ti have a 6 pin connector, I thought they would use less power than that.

I thought it would use more. Feeding the GPU and the CPU. Funny how different we can see things.

 

[Enigmoid]

I can potentially see massive CUDA acceleration with the Denver CPU.

 

[Termie]

TechPowerUp provides more clarity in this post re: the role of the Denver CPU:

http://www.techpowerup.com/196956/nvidia-readies-geforce-gtx-750-ti-based-on-maxwell.html

 

[Ventanni]

How long before someone loads Linux/Android/whatever on the GPU and runs it that way just for kicks? Didn't they do that with the Aegia cards at some point?

 

[Mand]

So, can someone explain what Unified Virtual Memory is really doing?

Does it allow my GPU to use system RAM for video tasks, thereby decreasing the importance of VRAM on the card?

Does it allow my CPU to use VRAM for other tasks, performing some benefit I'm not clear on?

 

[Pottuvoi]

Why on earth would a desktop GPU include a Denver CPU?

High serial performance computing unit with low latency which can launch jobs for the rest of the GPU and/or calculating the serial parts of algorithms and leave the embarrassingly paraller tasks for the standard ALUs.
There's a plenty of interesting things one could do with that...

 

[JDG1980]

I wonder if Maxwell will have the same limitations on integer GPGPU that holds back Kepler in cryptocoin mining. If Nvidia fixes that, then they might see a big boost in sales, at least in the short run.

 

[krumme]

Somewhere between a 260x and a 270 is spot on.

They apparently use the Denver CPU for memory compression (DCC?), so even though it's a 128-bit it does surprisingly well at high resolutions, especially versus other Nvidia cards.

Thanx for info.

Adressing that segment for NV makes perfect sense.

The 770 is very efficient, but they need something cheaper. If 128b bus can make that kind of perf its really an improvement. Reminds me of the old ati 4670 - but it was a good deal cheaper and without 6 pin. I think we will get there also for the 260x and 750, and then nv holds a significant perf advantage at the lowest end.

And the mobile parts is very much needed now when kaveri and broadwell steps in eradicating the former low end for mobile. Still it goes to show how the mobile dgpu market is fading very slowly but steadily for each year. The lowest end dgpu for mobile 2014 is a pretty costly card compared to the pathetic dgpu stuff that used to get into machines just 2-3 years ago. But its great progress for consumers.

 

[Cloudfire777]

Here is the rumored spec for GTX 750 TI

Its from early september last year

and it looks awefully like this Kepler mobile GPU

But the specs for GTX 750 Ti could be fake. I don`t know.
:hmm:

 

[ShintaiDK]

Its Kepler, not Maxwell.

Device 10de:11c7
Name: GK106 [GeForce GTX 750 Ti]