Computex 2009: AMD Demonstrates working 40nm DirectX 11 Silicon

[zerazax]

I've got to say, either AMD is into the PR games or is very very confident that the 40nm DX11 chips will come out soon for them to actually display a wafer of it out there. Its not something that happens often

[tajoh111]

I've got to say, either AMD is into the PR games or is very very confident that the 40nm DX11 chips will come out soon for them to actually display a wafer of it out there. Its not something that happens often

i think AMD trying to mislead NV again. how many shaders can you fit on this? consider rv740 is 640 and you have 35% about more area, I am thinking 960 or 1000 shaders might be the specs. Unless they do something magical, I have a feeling this part they are displaying going to be as fast as 100% scaling with 4770. 1000 shaders with a 925 mhz clock or something.

At that speed the gtx 385 has a good chance to be able to be faster than even the x2 version of this chip. I don't think AMD would thus be that arrogant and this should be a 150-200 dollar chip, with an larger higher end chip still not shown yet.

[Jowy Atreides]

4890 performance with DX11 and lower power consumption?

Yes Please!

[Manicdan]

but did they fix the memory controller to allow downclocking the ram without screen flickers? thats the real question

[Shintai]

but did they fix the memory controller to allow downclocking the ram without screen flickers? thats the real question

They surely must have. I dont think we could expect anything else since its a new design.

[Manicdan]

They surely must have. I dont think we could expect anything else since its a new design.

no ones perfect, and they didnt do it with the 4770, what a year after they couldnt do it with the 4870. i wouldnt put it past them if they somehow skipped the minor details trying to rush this out the door

[Krelin]

Where can I get a macro shot with that epic wafer for my desktop?

[Manicdan]

Where can I get a macro shot with that epic wafer for my desktop?

that wont give you any win 7 features you know

[madcho]

if not posted ..

[Shintai]

no ones perfect, and they didnt do it with the 4770, what a year after they couldnt do it with the 4870. i wouldnt put it past them if they somehow skipped the minor details trying to rush this out the door

4770 wasnt new. Just shrinked.

[hurleybird]

Hard to tell, but the layout looks totally different from RV770/790... kind of like a checkers board. Modular design perhaps?

[Krelin]

..if not posted ..

Well, have that on the screen ATM. Still, would love to have a close-up pic.

[ubuntu83]

if not posted ..

Lolz AMD the co owners of GF the rivals of TSMC

[hurleybird]

I'm surprised no one has leaked an up close die shot...

[zerazax]

i think AMD trying to mislead NV again. how many shaders can you fit on this? consider rv740 is 640 and you have 35% about more area, I am thinking 960 or 1000 shaders might be the specs. Unless they do something magical, I have a feeling this part they are displaying going to be as fast as 100% scaling with 4770. 1000 shaders with a 925 mhz clock or something.

At that speed the gtx 385 has a good chance to be able to be faster than even the x2 version of this chip. I don't think AMD would thus be that arrogant and this should be a 150-200 dollar chip, with an larger higher end chip still not shown yet.

Yeah 35% more, but bear in mind that RV670 was 192mm^2 and RV770 was 265 mm^2, roughly 38% more and we all know how the shaders turned out

I dont think we can go by just die size to determine how many units it will have but I'm sure this is exactly what ATI wants people to belive

[Helmore]

Hard to tell, but the layout looks totally different from RV770/790... kind of like a checkers board. Modular design perhaps?

Or just a test chip. Nothing on that board refers to that wafer as having ATI's next chips on it.

@zerazax: We're not even sure if RV740 only has 640 shaders, it might actually be a salvaged part to improve the meager yields of TSMC's 40 nm node. I doubt it though, but you never know.

[Nedjo]

4770 wasnt new. Just shrinked.

shrink of what? I'm not aware of any 55nm GPU out there with 640 SPs and 128bit mem. BUS... please point me where to find one?

[LordEC911]

4770 wasnt new. Just shrinked.

Yeah, 4770 wasn't new...
It is quite a modified RV770.

Or just a test chip. Nothing on that board refers to that wafer as having ATI's next chips on it.

@zerazax: We're not even sure if RV740 only has 640 shaders, it might actually be a salvaged part to improve the meager yields of TSMC's 40 nm node. I doubt it though, but you never know.

Actually we are pretty sure that it is 640 shaders, thanks to die shots.
If they are still only getting, supposedly, ~20% yields on a salvage part then TSMC's problem is much worse than we thought.

A test chip? What is the point of designing and developing a GPU for 40nm and then not using it?
Talk about a waste of time and resources.

[Tim]

Hmm, I have been out of the loop for a little while, but I thought showing working silicon wasn't just going to be a wafer lol. That was my own mistake btw, by reading the information in a totally different (read, wrong) way. My bad.

[informal]

Hmm, I have been out of the loop for a little while, but I thought showing working silicon wasn't just going to be a wafer lol. That was my own mistake btw, by reading the information in a totally different (read, wrong) way. My bad.

There was also a custom DX11 ATI demo running ..

[Tim]

There was also a custom DX11 ATI demo running ..

Ah my bad, I was right then, but wrong....erm. Yes.

That's quite impressive then, wonder what card/gpu it is, non qualification sample ofcourse, but very interesting.

[madcho]

on the screen i see 25 core wide. We suppose the wafer is 300mm wide, then this gpu is 144mm².

I must be wrong i think ..., i belive it's more near 250mm².

Anyone can calc it better ?

[Helmore]

Actually we are pretty sure that it is 640 shaders, thanks to die shots.
If they are still only getting, supposedly, ~20% yields on a salvage part then TSMC's problem is much worse than we thought.

A test chip? What is the point of designing and developing a GPU for 40nm and then not using it?
Talk about a waste of time and resources.

By a test chip I meant a chip that might not even be a GPU, just a chip to test how the 40 nm process node is doing (a bit like the SRAM chips they usually produce to master a process). This way they won't be giving NVIDIA any ideas about what ATI is going to come up with. I'm only saying this because the chips on the wafer don't seem to have some kind of structure that seems to be from a GPU.
Just look at this picture: http://www.pcgameshardware.com/aid,6...&show=original
To be honest though, I do think it is a wafer full of Evergreen GPUs and I can't wait to see how it will perform.

@madcho: it's around 180 mm^2.

[perkam]

Next part of AMD's Computex Event starts in 3 and a half hours (11PM Eastern) and there are events schedules from 2 AM Eastern to 5 AM Eastern ranging from Notebook and Desktop segments to Graphics.

So, when we wake up tomorrow, we should have a LOT more information on our hands

Perkam

[madcho]

By a test chip I meant a chip that might not even be a GPU, just a chip to test how the 40 nm process node is doing (a bit like the SRAM chips they usually produce to master a process). This way they won't be giving NVIDIA any ideas about what ATI is going to come up with. I'm only saying this because the chips on the wafer don't seem to have some kind of structure that seems to be from a GPU.
Just look at this picture: http://www.pcgameshardware.com/aid,6...&show=original
To be honest though, I do think it is a wafer full of Evergreen GPUs and I can't wait to see how it will perform.

@madcho: it's around 180 mm^2.

a bit too small for RV870, do you think ?