G80 Already in the works for 2006.

[Seph gPu]

As I stated before, the 7800GTX is a "hold-em over" card, the real meat and potatoes is in 2006 with the G80. THAT will have enough for UT2K7, Prey, etc. (If you want max detail max settings)

NVIDIA's Chief Architect David Kirk has said in the recent interview that they will do a unified architecture in hardware when it makes sense and when it is possible to make the hardware work faster unified. It will be easier to build in the future, but for the meantime, there's plenty of mileage left in this G70 architecture.

http://www.vr-zone.com/?i=2492&s=1

For the lazy....

"VR-Zone came to know that the future is actually pretty near where NVIDIA G80 design is based on unified shader architecture slated to appear in 2006. ATI has already taken the unified approach for their R500 Xenon within the XBox 360 and has 48 unified pipelines. As we know, in an unified shader architecture, there are no dedicated vertex and pixel shader engines but unified shader engine capable of executing both types of instructions.

G70 already supports Longhorn WGF 1.0 API so most likely G80 will support WGF 2.0 with improved virtualization techniques and new pipeline stages. The likely process technology choice for G80 is 90nm and the architecture is made for high core speed (~1GHz). G80 design is completed and waiting for ATI R580 when the time comes. "

[mattyo]

!@#$cough!@#$cough!@#$

I asked DK about unified architectures when I was given my very initial NV40 breifing he argued quite vehemently against them then. He has pretty much been consistently on that path since then, up until fairly recently where be has been making more conciliatory noises about it. Given that DK is working one or two architecture down those types of thoughts are probably about the types of things he's actually working on. Given the recent noises I think its almost certain they will go the unified route at some point, but I personally don't expect it for G80 given the design of this thing is probably in its final stages (i.e. the high level "architecture" choices were set down a long time ago), but possibly for the architecture after - this type of timing would also fit a lot better with the timing for WGF2.0.

[jAkUp]

Hmmm.. and I thought NV swore against unified shaders..

[MUYA]

They havent sworn off....just saying the time isn't right just now....translating that until "the time is right" they will do things the traditional way. Maybe g80 is with unified architecture....or not

[Banko]

I would think the G80 will release when Longhorn does.

[killahsin]

And another gpu will become bottlenecked severely by the cpu. If it were released today that is. The g70 can handle quite well the games you mentioned. The problems lay mainly with the cpus. This is why the most fundemental change even in the console department has been going to multithreaded. I suspect as those multithreaded games (oblivion ut2k7 unreal 3 based games etc) come out people with the gtx's and x2 64's are going to be very very happy.

[Riptide]

By the time UT2K7 comes out most high end users won't be playing on a 7800GTX. The G80 or whatever will be out before then. X2's though will probably be more widespread. Hopefully prices will come down on them by Q2 of 06. I'd like to get a 4600+ X2 for less than $500.

[jAkUp]

Yea im waiting on a 3GHz+ X2 :D

[Riptide]

Same here I think. Unless the 4600+ gets real attractive pricewise I'm waiting for a bigger bump in clock speeds over what I have now. 3GHz would be nice. :)

[rohit]

And another gpu will become bottlenecked severely by the cpu.
I doubt this when longhorn would release. longhorn, as stated by many sites ""will have a different approach towards HW and the way to utilize them for a particular app".
also i agree to, "dual-core will be widely spread by the time longhorn is out. longhorn will benefit most from it."

[ricercar]

I'm unsurprised that the G80 design is inflexibly complete. Contemporary GPU design starts 2+ years before they tape out.

Consider: NV47/G70 was the first consumer release that I never worked on, and I was with NVIDIA until September 2003. That's ~18 months before an "unknown" product came to fruition. G70 is merely NV47, a speed bumped NV4x. The rearchitected NV50 series has been in design much longer.

[shabby]

And another gpu will become bottlenecked severely by the cpu.

Blame canad... err intel :)

[Graphicmaniac]

for Ricercar

could u explain what Nvidia president means when he said U3 engine will be integrated in the hardware of nv50?

rsx is more g70 or more nv50?

[Razor1]

when was this stated? I think the g70 and later cards were designed to work better with the Unreal 3 engine not that the engine is actually going to be integrated. Just like the Fx line and gf6's were designed to work better with the Doom 3 engine.

[Acid Rain]

And another gpu will become bottlenecked severely by the cpu. If it were released today that is. The g70 can handle quite well the games you mentioned. The problems lay mainly with the cpus. True with some games, but even 7800GTXSLI can become bogged down a bit with extremely high IQ settings, (8xAA at 1920x1440 and beyond, etc.) and if FEAR shows anything, it's still all about graphics horsepower. In many benches, the bottleneck is clearly v-ram bandwidth, something demonstrated several times by Ratchet's Rage3d review.

Also, this is plainly seen in SLi benchmarks, where, with the same "lowly" CPU, performance in many games at very high settings is nearly doubled. If it was a CPU bottleneck, there would be practically no difference, in fact, results might even be worse due to the additional CPU usage required by SLi.

We really need a 512bit bus to the v-ram. That will truly crack open a new can of performance. Sure, faster CPUs will help, but certainly not enough to overcome the on-card bandwidth bottleck.

[Graphicmaniac]

Well now is a bit a prblem find where i have read it, but in gf3 or gf4 period i have read an article where that president said they were already working on nv50 and this would have integrated U3 engine. really i have not dream lol

now maybe nv50 is changed, cancelled or maybe he wanted mean optimized for U3 but what he said in interview was about integrate in hardware .. he said nothign about D3 anyway.

[Graphicmaniac]

is not the article i was talkign about but is always better than nothing, here it says they will use specific optimizations

http://www.************/read.php?cID=363&pageID=529

"Rein also mentioned that NVIDIA is Epic's premiere partner. He said the two companies have a great relationship. He accidentally put something about NV50 on one of his slides and joked that he would get in trouble for it. It said that NVIDIA and Epic are collaborating on next-generation hardware so that NV50 will have UnrealEngine3 specific optimizations."

[Subtestube]

"Rein also mentioned that NVIDIA is Epic's premiere partner. He said the two companies have a great relationship. He accidentally put something about NV50 on one of his slides and joked that he would get in trouble for it. It said that NVIDIA and Epic are collaborating on next-generation hardware so that NV50 will have UnrealEngine3 specific optimizations."

Yeah - that doesn't really mean that UE 3.0 will be 'integrated onto the chip', but that, as he says, the NV50/G80 (I'm assuming here) will have certain optimisations that are put in place to make the UE 3.0 run somewhat better. Like, for example, how Ultrashadow 2.0 was specifically designed to make stenciled shadows go faster - that was in many ways a concession to Carmack and the D3 engine. In this case we might expect some nice Shadow Buffer specific hardware, or, really, anything else. As I can't think of the UE 3.0 features off the top of my head, I can't think of what other opts might be in place, but basically all this means is that the chip will have special functionality (like Ultrashadow 2.0) that will be targeted at the new Unreal Engine's feature set.

[walterman]

We really need a 512bit bus to the v-ram. That will truly crack open a new can of performance. Sure, faster CPUs will help, but certainly not enough to overcome the on-card bandwidth bottleck.

Totally agree. At least, they should have put those 1.6ghz Samsung chips in this new gen card.

[killahsin]

True with some games, but even 7800GTXSLI can become bogged down a bit with extremely high IQ settings, (8xAA at 1920x1440 and beyond, etc.) and if FEAR shows anything, it's still all about graphics horsepower. In many benches, the bottleneck is clearly v-ram bandwidth, something demonstrated several times by Ratchet's Rage3d review.

Also, this is plainly seen in SLi benchmarks, where, with the same "lowly" CPU, performance in many games at very high settings is nearly doubled. If it was a CPU bottleneck, there would be practically no difference, in fact, results might even be worse due to the additional CPU usage required by SLi.

We really need a 512bit bus to the v-ram. That will truly crack open a new can of performance. Sure, faster CPUs will help, but certainly not enough to overcome the on-card bandwidth bottleck.


I'm not talking about speed bottlenecks. i'm talking about performance. the reality is that there isn't only a bandwith bottleneck as we all love to prescribe to. We can push millions of poly's np. but as soon as you add materials to them you start to run into the bandwith wall. What i am saying is that multi core is going to facilitate more bandwith in a different manner of speaking. The extra performance(not really power) will allow the cpus to handle a more diverse set of things. For instance games being filled with statics(brushes whatever) on the fly mathmatically via pre defined specs. Character model faces being designed on the fly and so forth. Things that were previously being done not only by designers but by the gpu only can now be pushed by the cpu. It also depends fully on the engine. so a game like fear of course that is insanely shader bound will of course push the gpu as much as the cpu if not more. All i am saying that game progression right now is severelly being limited by the cpu rather then the gpu. You could take quake 3 lol and soup it up to the point where it would again become gpu bound by introducing tons of shaders. I was more or less just refering to the fact that the main difference between engine design as of yesterday and engine design as of today/tommorow is the multi core cpus, and what they will offer. There will be alot of overhead freed up. Even texturing is going to be generated by pre defined specs in the future on some games (lol). And no lets not have that debate, again. The procedural debate is an evil one.

[Zelda_fan]

Anyone who has seen the FX57 benchmarks would know that games like Splinter Cell Chaos theory are NOT cpu bound. Bring on the G80! To be honest, this is probably when I'll update my computer when the G80 comes out. I want my next computer to be fully longhorn compliant.

[killahsin]

I'm looking more forwards to the x2's and multi threaded engines. Rather then the g80 to be honest. Although the g80 is the next chip i'll be getting.

[Zelda_fan]

it will take a while for the who "dual core" idea to catch on in the software, but now that both AMD and intel are pushing in that direction, you can expect software to come multithreaded as a standard. As of now, the FX57 is the fastest chip for a single threaded application, but if I were buying for the future, I would choose the X2 series without a doubt.

[Saintster]

That's the card I'm waiting for to make move to pci express!

[borntosoul]

That's the card I'm waiting for to make move to pci express!


i second that! i dont wonna spend a heap of money to buy new MB and pROC and then get a pciE card that wont make a huge different to the performance i have already. if im gonna upgrade im gonna go big this time around.