ATI RV770 to have 900 stream processors

GermanPC Games Hardware writes the the ATI RV770 Radeon 4850 and 4870 actually has 900 stream processors but that only 800 of them are currently activated: The RV770 is supposed to have 900 stream...

More...

 

well this is something i didn't know.

i bet people will try and get them to work now.

 

cannot see it myself , but it would be nice if they released a bios to unlock these extra stream processors

 

And ATI woud have lot of problems with "broken" GPU´s =/
No way...

 

RIght, additional 100 stream processors would not make any difference.
I find that hard to believe. More always means better in graphic cards world.

 

haha i think someone will come up with bios to unlock if its possible at all.

 

I hope someone will. I'm willing to give it a try. FSAA hit would certanly be lower with extra 100 streams. This means either slightly faster Crysis or Crysis with FSAA hehe

 

There is likely a defect in one of those 100 SPs simply using a bios to 'hack' it could result in some serious problems. The only thing that could work is a very sophistocated system of running tests on the clusters and activating the ones that work. However, I would assume these on current cards are lazer locked.

What AMD can do is cherry pick/bin parts that are 100% working and have something to go against Nvidia. I don't see AMD doing this outright, but it may be somethign they give an option to their partners. They need to focus on getting the 40nm chip (RV740) out the door so they can roll out the RV870.

 

Oops ATI fixing excesive performance in their cards ...

 

That's a load of bollocks.

ATI cards don't have as many shaders as they advertise.

They use the same 16 "cluster" based design that nvidia use, and in direct comparison actually use less of them.

If it did have 900 of them, in actual fact it would have 182 shader processors, which is impossible if they are clustered in groups of 16, because the 4870 as i said in another thread has 160. You can't make 180 from 160 in groups of 16.

 

Hold on. Each shader(Cluster of 5 > the 180 you gave) is able to do 5 threads. So that's why it acts as 900 shaders. I wonder how this evolves

 

No, each shader is clustered in a group of 16 and each shader can operate 5 threads.

In hardware, they use the same grouping technique as Nvidia.

Each shader can operate with 5 threads, but those 5 threads arent actual shaders on hardware.

If they did in actual fact have the amount of shaders they said they did, then the card would have 182, not the 180 (or 900) they are saying. It would be a stupid figure likw 910. Yet the say they count 900... which is impossible...

 

Ah, you have a point there. I didn't know that they were actually clustered @ 16. I always thougt they were clustered with 5. Hence my confusion, srry. But then again. Every extra activated shader cluster helps.

 

aye, the only problem is, the more shaders you have onboard, the more you'll need to increase performance there after.

It's true what they said though if you look at the GTX260, you got on average a 5% increase in game based performance. IF they did indeed have a dormant cluster, you'd be looking at 10% or there abouts.

Still, i think they're clutching at straws though tbh From what i can see, theres nothing there to activate.

 

Impossible like 800 stream processors on HD4800 cards?

 

Hows that impossible though?? :grin:

In ATI's wierd way of calculating their hardware, it does indeed have 800 stream processors.

160 x 5 threads per shader = 800

900 shaders calculated by the same equasion can't be done in a 16 shader cluster.

 

It's marketing gimmick.

 

The R700 cards are all based on the previous R600 architecture, where each cluster was really a superscalar unified shader cluster that consisted of 5 stream processing units, therefore they basically are right when they say that their cards have more than 800 stream processing units.

In case you're wondering why they're not so powerful, it's because the flaw with the architecture is that it's based on VLIW architecture, and the chip can't co-issue instructions when one is dependent on the results of the other. Therefore the overall performance is highly dependant on the mixture of instructions (vector-specific mostly, which most programmers don't really bother coding for, so the full power of the card cannot be drawn upon) being used by the application and how well the real-time compiler in the driver can organize said instructions.

 

Believe it when i see it tbh :biggun:

 

One thing are stream processors, one thing are shader cores...