28nm Next-Generation GPUs 45% Faster

[The_Knight]

Source : 28nm Next-Generation GPUs 45% Faster - Softpedia

Since TSMC has begun to manufacture 28nm chips, people interested in what future GPUs will accomplish might be interested in learning of the most recent report regarding the 28HP.

28HP is the first process that TSMC will make with the High-k Metal Gate (HKMG) technology, instead of the silicon oxynitride (SiON) that 40nm GPUs employ.

They should have much lower gate leakage and, thus, a lower power draw and higher clock speeds.

TSMC managed this through a material that features a high dielectric constant, compared to silicon dioxide gate dielectric.

This report says that, though AMD refused to speak on the matter, ‘sources’ within TSMC itself stated that the 28HP HKMG can enable performance boosts of up to 45% over the technology used in the last two generations of graphics cards.

Needless to say, this is a substantial progress, even knowing that smaller process geometries usually do make possible faster, smaller and cheaper chips.

Of course, it isn’t a given that the upcoming graphics processors from NVIDIA and Advanced Micro Devices will actually sport such a big performance jump.

The boost would, after all, imply a fairly high power consumption, so the two companies may favor power efficiency first.

It is also worth noting that the high-end graphics cards of today aren’t really challenged overmuch by the latest games, so trying to make the next batch of cards as strong as can be could prove to be a rather unnecessary effort.

Then again, even if it does involve a higher power consumption than usual, some of AMD’s and NVIDIA’s partners might decide to make that sort of beast anyway.

Now the only possible problem that could strike both AMD and NVIDIA in equal measures is that of low chip yields.

TSMC has problems with its 40nm process for a long time and there is a real worry that the same will happen with 28nm.

[baccilus]

The big question is how will they be priced. The wait will be worthless if either it doesn’t perform great (like BD) or if it is priced too high as compared to similar performance from the older generation.

[Devan232]

Hope 7x isn’t a fail like Bulldozer :\

[The_Knight]

NVIDIA Kepler GPUs Mass Produce Shortly

With its latest Tegra out and about, NVIDIA is now getting ready to release its newest collection of GPUs, the so-called Kepler, which will, apparently, enter mass production shortly.

The Kepler graphics processing units have been spoken of before, but they are only now turning into a physical reality.

They aren’t going to start shipping yet, though, since that will only happen next year (2012).

What is, on the other hand, about to happen is the beginning of the mass production stage of these next-generation GPUs.

“There is a portion [of operating expenditures] that is related to 28nm tape-out of new processors that will go into production shortly,” said Jen-Hsun Huang, chief executive officer of Nvidia.

According to the latest conference call with financial analysts, the Santa Clara, California-based corporation already has a fair number of design wins.

“The vast majority of the increase is coming from the very significant increase in design wins that we have in several areas,†said Mr. Huang.

“We have more notebook design wins for the Ivy Bridge cycle than we ever had in notebooks; this is likely the most successful notebook cycle we have ever experienced. And so we have got a lot of engineers dedicated to getting those notebooks into production.â€

Kepler GPU are based on the 28nm manufacturing process and is supposed to make graphics cards not only more powerful, but also more flexible in terms of programmability.

GPGPU (general purpose processing on GPU) application development should accelerate and NVIDIA will also implement virtual memory space (lets CPUs and GPUs to use the “unified” virtual memory), pre-emption and a higher autonomy from the CPU.

“The reason for our [design win] success, I believe, is because our Kepler generation of GPUs was designed for intense energy efficiency. With energy efficiency, we were able to translate that to simultaneously higher performance, as well as longer battery life,” explained Mr. Huang.

Source : NVIDIA Kepler GPUs Mass Produce Shortly, Many Design Wins - Softpedia

[The_Knight]

28nm NVIDIA GPU Roadmap Revealed

Leaks and rumors about what NVIDIA has in store have not been absent, but one has finally arisen that provides the exact plans of the company in regards to the 28nm manufacturing process.

28nm is the technology that both Advanced Micro Devices and NVIDIA rely on for their new series of graphics processing units (GPUs).

This report has now shown precisely how NVIDIA plans to have its Kepler advance.

The high-end chip will have a smaller die than the massive Fermi and will bear the name GK104.

With 384-bit memory support, it can churn out 2 TFLOPS (30% more than GTX 580) and will be ready by the middle of next year.

Unfortunately, the prowess comes at the price of a high power consumption, 250W, which makes one wonder how much energy the GK110, a dual GX104, will need (will appear soon after).

Much sooner than GK104 will arrive the GK107, the first model as it were, which is a notebook part, with a memory interface of 128 bits and PCI Express 2 support (other GK chips have PCI Express Generation 3 support).

The GK106 will come second, a mid-range GPU, featuring 256-bit memory. Along with the GX107, it will show up at some point during the second quarter of 2012.

So far, the Santa Clara company seems to have similar designs as AMD, but the similarities end at the end of 2012 / early 2013, when NVIDIA will again launch a GPU with a massive die: GX112, with 512-bit memory.

So far, things sound optimistic, but there is always a catch and, in this case, it is TSMC that might cause disruptions to this plan, as well as AMD’s own.

Many fear that the nightmare behind the 40nm low chip yields, which hit both companies very hard last year, will repeat with 28nm.

The first signs are already showing in fact, as Taiwan Semiconductor Manufacturing Company is already experiencing a high demand while producing only 20,000 units per month.

Source : 28nm NVIDIA GPU Roadmap Revealed - Softpedia

[neojjjk]

Re: 28nm NVIDIA GPU Roadmap Revealed

pci e 2.0 support only for gk 107 ? that’s too bad .

[The_Knight]

28nm NVIDIA Kepler GPUs Have Support for DirectX 11.1

As the time for launching next-generation GPUs draws near, NVIDIA can’t seem to hide the information on them, with bits and pieces making their way to the web, this time regarding the DirectX support.

The GeForce GTS 400 and 500 series served their buyers well, with their support for pretty much everything, including DirectX 11 graphics.

Now, thanks to the roadmap we spoke about two days ago, it was revealed that the Kepler GPUs would possess next-generation DirectX 11.1 features.

Add to that PCI Express Generation 3 compatibility and you have just what next-gen PCs are expected to offer, especially with Windows 8 closing in.

We know that Intel Ivy Bridge CPUs will support this too, but they definitely won’t have the same prowess as NVIDIA or AMD cards, so there’s no real danger on that front yet.

Link: 28nm NVIDIA Kepler GPUs Have Support for DirectX 11.1 - Softpedia

[ALPHA17]

Re: 28nm NVIDIA GPU Roadmap Revealed

^^ Doesn’t seem to be a great issue, I think the PCi-E leap frog from v1.1 to v2.0 wasn’t really necessary but then lets wait and watch what this system (PCi-E 3.0) offers [apart from bandwidth increase].

[The_Knight]

Nvidia GTX 780 Performance Estimates Surface

**Many suppositions regarding Nvidia’s next-generation Kepler GPUs have come our way during this last week, but now an even more startling rumor has started circulating on the Web.

**

According to this, Nvidia’s upcoming solutions will be released into the GTX 700-series and, the top part, named the GTX 780, will be able to deliver a twofold increase in performance over the current GTX 580.

At the heart of these rumors, stands a slide that was recently published on a Chinese forum, which pretty much suggests that the GTX 780 will blow all its competitors out of the water.

It goes without saying that we doubt the authenticity of this slide as the performance increase seems too high from one generation to the other, but if indeed true than Nvidia has a winner on its hands.

For now, we can just wait and see how will things evolve, but I have a feeling that this graph will be proven to be fake in the not so distant future.

Nvidia GTX 780 Performance Estimates Surface - Softpedia

[user67]

Re: Nvidia GTX 780 Performance Estimates Surface

this is fake. REASON:

nVidia JUST GOT BACK 28nm circuits from production. They havn’t had time to run tests, it’s IMPOSSIBLE. Currently the chips are in testing, not a final product.

I find this a cheap stunt from nVidia to steal away some of AMD’s HD7900 glory.

[Tao]

Re: Nvidia GTX 780 Performance Estimates Surface

@comp@ddict

this is fake. REASON:

nVidia JUST GOT BACK 28nm circuits from production. They havn’t had time to run tests, it’s IMPOSSIBLE. Currently the chips are in testing, not a final product.

I find this a cheap stunt from nVidia to steal away some of AMD’s HD7900 glory.

Its hard to believe but certainly not impossible - if Kepler has twice the number of CUDA cores (which it likely does). And what makes you think they don’t have working in-house samples - if they are going to launch it next year they are bound to have got the samples, tested them and decided if they need to go for another revision more than once by now. They wont be able to mass-produce the chip because of its large size and problems with the 28nm HP process - but that has nothing to do with its design specs and consequently, performance figures.

[muzux2]

Re: Nvidia GTX 780 Performance Estimates Surface

@comp@ddict

this is fake. REASON:

nVidia JUST GOT BACK 28nm circuits from production. They havn’t had time to run tests, it’s IMPOSSIBLE. Currently the chips are in testing, not a final product.

I find this a cheap stunt from nVidia to steal away some of AMD’s HD7900 glory.

It might be possible that they have working samples in house but yes those graphs looks highly fake to me as well.

[devvfata1ity]

Re: Nvidia GTX 780 Performance Estimates Surface

@comp@ddict

this is fake. REASON:

nVidia JUST GOT BACK 28nm circuits from production. They havn’t had time to run tests, it’s IMPOSSIBLE. Currently the chips are in testing, not a final product.

I find this a cheap stunt from nVidia to steal away some of AMD’s HD7900 glory.

as if they told you when they got the chips back from the fab

[mrcool63]

Re: Nvidia GTX 780 Performance Estimates Surface

one more thing.. where can you find metro last night

[vivek.krishnan]

Re: Nvidia GTX 780 Performance Estimates Surface

Quite possible, but then, I dont think these will be out for quite a while. IMO, 1.5 yrs at least, 2 yrs is my estimate.

[asingh]

Re: Nvidia GTX 780 Performance Estimates Surface

Yea, probably correct. Only if sub stations transformers have a PCI.E connector.

[vivek.krishnan]

Re: Nvidia GTX 780 Performance Estimates Surface

@asingh

Yea, probably correct. Only if sub stations transformers have a PCI.E connector.

That, IMO, pwnd the whole thread!

[user67]

Re: Nvidia GTX 780 Performance Estimates Surface

@devvfata1ity

as if they told you when they got the chips back from the fab

umm really? that was lame. read around the net a bit more and you’ll definitely keep up to date buddy

[Zloyd]

Re: Nvidia GTX 780 Performance Estimates Surface

Those numbers are a little optimistic but i think if you scale them back to 80% of the numbers they are claiming its whats expected at minimum. Im expecting a big jump this time … ala another 8800gtx 280 480 type card. Power hungry hot but also beastly beastly performance. Also I think these card are still 6 months away.