New rumor about ATI Southern Islands

[flopper]

If AMD genuinely desires to address the HPC market, they will have to invest more transistors into their designs for dedicated HPC function....which will mirror what Nvidia has done with Fermi.

Of course, the AMD fan boys will herald this as the greatest thing ever even if the power envelope is increased dramatically, much as what happened with Fermi

Fusion as a concept and the aim for power/price/performance strategy will allow them to come up with stuff that suits that concept of a HPC tech card down the line.
I cant think they be skipping a growing new market as we move towards a union between different structural elements, combining for example a videocamera with a normal camera is now on the way, you buy both in the same toy as the technology now have moved far with that.
Like the phone that is now, music, video, chat, and well a phone.

social networking is increasing, at a rate that is astounding, and software and hardware needs to met up there as that continues.
If I be hired by amd I advice them to get going there with their commercials even more.

So, our hardware, video and cpu, and other things is getting soon to a point where we can have a combined element, and fermi is just powerhungry due to bad design, first generation tends to be that.
I expect that Nvidia will improve the power issue a lot in the upcoming revisions they do.

[STEvil]

Hey carfax. Multi-quote.

[Florinmocanu]

It is a failure, compared to the 8000s series. It's a "meh" architecture when gaming is concerned because it takes a huge die and a lot more TDP to surpass the smaller and cooler Cypress chip. that is, perf/mm.

[madcho]

It is a failure, compared to the 8000s series. It's a "meh" architecture when gaming is concerned because it takes a huge die and a lot more TDP to surpass the smaller and cooler Cypress chip. that is, perf/mm.

Cypress is far from full optimised. Cayman will be better to that. So i think too, fermi is a total failure, like was old 512bits HD2900

[spursindonesia]

Darn, as i've said before, incoming FUDs & fanbois talks are soooo predictable, and now the thread is going fubar as it is. Just say i don't fu<kin believe the numbers and moving on, ATi is da suxxorz that incapable of such feats, that's still very much respectable in my view.

[rozzyroz]

Makes my english look fantastic!

Finger's up tho, keep'em comin'

classic

i think you mean thumbs up... fingers up means something completely different

[LightSpeed]

Any long term Nvidia user will tell you that it's nothing for Nvidia to squeeze about 15% extra performance (average) or more out of their GPUs between the introductory drivers, and the fully optimized drivers; which may take about a year or more to realize..

But the current drivers are not introductory drivers. Fermi already received a performance boost with the latest drivers. Also, why are you ignoring the fact the cayman is a new architecture as well, that can be tweaked for more performance with newer drivers too

[spursindonesia]

But the current drivers are not introductory drivers. Fermi already received a performance boost with the latest drivers. Also, why are you ignoring the fact the cayman is a new architecture as well, that can be tweaked for more performance with newer drivers too

For fanbois, the road only goes one way. ......

[Carfax]

So, our hardware, video and cpu, and other things is getting soon to a point where we can have a combined element, and fermi is just powerhungry due to bad design, first generation tends to be that.

Unless you're a semiconductor engineer, how could you possibly know that Fermi is power hungry due to "bad design?"

I don't think most people here understand that Fermi is the first TRULY computational GPU. All those things you said about convergent technologies into one neat package, are made manifest in Fermi....in the sense that Fermi has many CPU like qualities such as being fully IEEE compliant, support for C++, having dedicated caches and support for much greater VRAM.

Hey carfax. Multi-quote.

It is a failure, compared to the 8000s series. It's a "meh" architecture when gaming is concerned because it takes a huge die and a lot more TDP to surpass the smaller and cooler Cypress chip. that is, perf/mm.

The very fact that a GPU like Fermi, which isn't even a pure gamer chip can outperform Cypress (which is dedicated towards gaming) speaks volumes in and of itself.

And as I mentioned before, Fermi's performance is increasing with every driver update. I've noticed a lot of differences between the various driver revisions since I've owned mine.

It wouldn't surprise me if Fermi ends up being 20% faster than it was on introduction, simply due to driver optimizations.

[Carfax]

But the current drivers are not introductory drivers. Fermi already received a performance boost with the latest drivers.

Yes, the current drivers are no longer introductory and have Fermi specific optimizations. But it's only been a few months since the release of Fermi, and they've already gotten large performance increases.

It usually takes a year or more to really milk the architecture for all it's worth, so more performance increases are certainly coming.

Also, why are you ignoring the fact the cayman is a new architecture as well, that can be tweaked for more performance with newer drivers too

I wouldn't say I was ignoring it. It's just that even though Cayman is a new architecture, it will still share very many similarities to it's predecessors.

Since they migrated towards a unified shader architecture, ATI's strategy has been to simply double all the resources on new designs. I think the only thing that ATI didn't double with the R800 was the memory interface, which remained at 256 bit.

And since Cayman won't be getting a die shrink, it seems like it will be a more refined version of the R800, rather than a next generation architecture.

Anyway, driver optimization is an ongoing affair with no end in sight. Cayman will undoubtedly receive performance improvements via driver tweaks, but I doubt it will get as much as Fermi, since Fermi was a completely new design.

[Jowy Atreides]

But the current drivers are not introductory drivers. Fermi already received a performance boost with the latest drivers. Also, why are you ignoring the fact the cayman is a new architecture as well, that can be tweaked for more performance with newer drivers too

Don't forget all the "ATI bad drivers hurrrrrrr" comments.

If nvidia drivers are unoptimized and inefficient for a whole year and carvidia himself said it.
Yup, rub in face time

[Manicdan]

i feel like carfax is going to cause another thread to be locked

my opinion is that nvidia is not a graphics card company anymore
they are HPC
Physx
and drivers

they simply buy games with TWIMTBP, and force you to use their products to enjoy it

i think it would have been easier and cheaper to just make a good graphics card from the start, instead of having to expand your business just to keep your customers you already had.

its time they get back to their roots and build something for gamers

[Frontl1ne]

its time they get back to their roots and build something for gamers

Didn't you know? The future is CUDA. Who cares about gamers when our graphics cards have 512 CUDA CORES!!!!

[Carfax]

i feel like carfax is going to cause another thread to be locked

And how is that? I've been nothing but civil in this thread.. It's not my fault if other people get bent out of shape

its time they get back to their roots and build something for gamers

I don't know what you're talking about. 480 SLi is the fastest gaming setup I've ever owned, and along with 480 Tri SLi, it is the only setup that can comfortably handle the XD resolutions ie 2560x1600 and above.

Ironically, your pure gaming chips stumbles miserably when it comes to the extreme gaming

[Mechanical Man]

Didn't you know? The future is CUDA. Who cares about gamers when our graphics cards have 512 CUDA CORES!!!!

Cut down to 480 CUDA cores by design...

[LightSpeed]

I wouldn't say I was ignoring it. It's just that even though Cayman is a new architecture, it will still share very many similarities to it's predecessors.

Since they migrated towards a unified shader architecture, ATI's strategy has been to simply double all the resources on new designs. I think the only thing that ATI didn't double with the R800 was the memory interface, which remained at 256 bit.

And since Cayman won't be getting a die shrink, it seems like it will be a more refined version of the R800, rather than a next generation architecture.

Anyway, driver optimization is an ongoing affair with no end in sight. Cayman will undoubtedly receive performance improvements via driver tweaks, but I doubt it will get as much as Fermi, since Fermi was a completely new design.

Yes, Cayman will be a more refined version of the R800. Think RV670 > RV770. Along with increasing the resources, efficiency was increased as well.

Also, since Fermi already got a performance boost with the new driver, and Cayman isnt a completely new architecture, Id say performance boost would be similar between the two with new drivers. Which leaves Cayman with a higher performance and lower power consumption as well (most likely)

The Nvidia exclusive features dont matter enough to the majority of people to warrant a card thats slower and consumes more power IMO, but to those who think the features are that important, thats perfectly fine. To each, their own

[Manicdan]

And how is that? I've been nothing but civil in this thread.. It's not my fault if other people get bent out of shape



I don't know what you're talking about. 480 SLi is the fastest gaming setup I've ever owned, and along with 480 Tri SLi, it is the only setup that can comfortably handle the XD resolutions ie 2560x1600 and above.

Ironically, your pure gaming chips stumbles miserably when it comes to the extreme gaming

your in here trying to preach about how your gaming machine is the fastest in the world

the discussion started because we say that perf per watt and perf per mm2 is way in favor with AMD vs the gf100, which is a fact.

phsyx being awesome, is an opinion
nvidias driver team is going to get you 15% more perf in the next year, is an assumption
nvidia moving to cuda cores rather than gaming perf will earn them more money in the long run, is an assumption.

and all 3 of those things have nothing to do with rumors for the 6000 series cards, but instead push people into an argument very similar to those in the last physx thread that was closed

so please either offer something useful without starting up a my card is bigger than your card war

[flopper]

Unless you're a semiconductor engineer, how could you possibly know that Fermi is power hungry due to "bad design?"

.

are you one?
If you aint how would you even consider my answer to be appropiate as you wouldnt understand it?

Its called 460 since Nvidia went over the top with their first generation Fermi.
its bad design, simple as that.

amd on the other hand, well...their 3 last designs and implementations seems to be flawless. The jury still out for the 6000 series.

[Calmatory]

Ironically, your pure gaming chips stumbles miserably when it comes to the extreme gaming

Extreme gaming.

Honestly.

[Carfax]

your in here trying to preach about how your gaming machine is the fastest in the world

Talk about exaggerating. You guys were the ones bragging as to how the Cayman would own Fermi and blah blah blah. I merely advised to be more skeptical of the benches, and as it turned out, they are likely fake.

the discussion started because we say that perf per watt and perf per mm2 is way in favor with AMD vs the gf100, which is a fact.

It's only a fact if you hold to your ignorant view that Fermi is a dedicated gaming processor, when the "fact" is, it's not.

Fermi is much more than just a gaming GPU, and as such, you can't hold it to the same standard as you do Cypress.

phsyx being awesome, is an opinion

I agree and even said so. PhysX is an extra optional feature, that not everyone will care for.

nvidias driver team is going to get you 15% more perf in the next year, is an assumption

It's an assumption based on recurrent past experience, so it's not as though it's groundless.

nvidia moving to cuda cores rather than gaming perf will earn them more money in the long run, is an assumption.

You act as though every assumption is bad, or has no element of truth.

It's a FACT that the HPC market is more profitable than the gamer market.

As I already mentioned previously, a Quadro video card can cost as much as 5,000 USD! With that kind of price tag, is it a groundless assumption to think that Nvidia investing more resources in HPC won't earn them more money?

You think Nvidia is stupid? Why do you think they designed Fermi in the first place?

and all 3 of those things have nothing to do with rumors for the 6000 series cards, but instead push people into an argument very similar to those in the last physx thread that was closed

Seriously, take your blinders off. You act as though you guys have been discussing nothing but 6000 rumors, when in reality, you've been comparing it to Fermi and slamming Fermi while touting the 6000 series as the greatest thing ever when the specs aren't even known!

so please either offer something useful without starting up a my card is bigger than your card war

I'll do that the moment you guys stop acting as though the 6000 series is going to own Fermi.

[Calmatory]

I do think Nvidia is stupid. If they weren't, Fermi wouldn't have been a failure for them. It didn't meet their expectations, hence why it was failure for them. Saaya's sig sums it up.

[Carfax]

Also, since Fermi already got a performance boost with the new driver, and Cayman isnt a completely new architecture, Id say performance boost would be similar between the two with new drivers. Which leaves Cayman with a higher performance and lower power consumption as well (most likely)

Cayman could very likely outperform the 480, while having lower power consumption I agree. I guess we'll just have to wait and see..

The Nvidia exclusive features dont matter enough to the majority of people to warrant a card thats slower and consumes more power IMO, but to those who think the features are that important, thats perfectly fine. To each, their own

Well there's more to it than that. A lot of people go with Nvidia because of their driver support.

Whether you agree with that or not, you have to admit that there's a perception that ATI's drivers are inferior to Nvidia's.

As long as I've been a computer enthusiast (which is slightly more than a decade now), that stereotype has persisted.

are you one?
If you aint how would you even consider my answer to be appropiate as you wouldnt understand it?

I'm not, but you don't see me making comments as to whether Fermi had a bad design or not.

That kind of question warrants indepth knowledge that I don't possess.

Its called 460 since Nvidia went over the top with their first generation Fermi.
its bad design, simple as that.

Fermi was designed to tap into BOTH the HPC and gaming market. As such, how can you say it's a bad design when Fermi is leading the 5000 series in gaming, and utterly destroying ATI's FirePro series in the professional market.

amd on the other hand, well...their 3 last designs and implementations seems to be flawless. The jury still out for the 6000 series.

So flawless they need to use "brilinear" filtering

[Carfax]

I do think Nvidia is stupid. If they weren't, Fermi wouldn't have been a failure for them. It didn't meet their expectations, hence why it was failure for them. Saaya's sig sums it up.

Yep, Fermi is a failure when it's leading the 5000 series in most gaming benchmarks (especially DX11), and destroying the FirePro series in professional applications..

Not to mention xtreme high resolution gaming, and 3D gaming where again, Nvidia dominates ATI.

[Iconyu]

I do think Nvidia is stupid. If they weren't, Fermi wouldn't have been a failure for them. It didn't meet their expectations, hence why it was failure for them. Saaya's sig sums it up.

Actually, I think nvidia didn't want to use Fermi for the gaming market. It's powerful but it's not a gaming card really, it's just got so much grunt it can pretend that it is.

Charlie said they had planned a different card but scrapped it and only had Fermi in reserve so they had to make it fit. And I agree with him, Fermi could not really be nvidia's first choice for the gaming market, they would have known it would be a bastard to make but when you're planning to sell at 5k each it's not a issue.

[LightSpeed]

Cayman could very likely outperform the 480, while having lower power consumption I agree. I guess we'll just have to wait and see..

Well there's more to it than that. A lot of people go with Nvidia because of their driver support.

I have to agree there. AMD's driver team seriously needs to get their act together, especially the stuff that get broke after being fixed before.