The ATI Radeon 5XXX Thread

[ajaidev]

http://www.semiaccurate.com/2009/09/...-vs-xbox-360s/

"E OF THE FEATURES of the upcoming ATI Evergreen family, also known as the 5-series, is a tessellator. While this might be old news to graphics card enthusiasts, this time it really is different, mainly because Microsoft is finally backing the technology.

ATI has been putting tessellators"

[largon]

GDDR5 uses more pins which caused the bigger die to be pad limited

I thought so, too, until I counted them. Turned out GDDR5 has 62 I/O pins per chip while GDDR3/4 has 69. GDDR5's has more pins total but that's just because it has 100 power/ground pins, while GDDR3/4 has 70.

[LordEC911]

I can't remember the exact reason behind the pad limit then, if it wasn't the too small die. There was a review or an article about it, and RV770 was supposed to have 480 SPs, but was pad limited so they had "free space" to fill up and slapped 800 SPs to it. GDDR5 uses more pins which caused the bigger die to be pad limited, as far as I can recall.

Correct, GDDR3/GDDR5(GDDR4?) memory controllers, more pins for the GDDR5 due to power/ground, sideport taking up extra perimeter, UVD2 is slightly larger than the original. Let's not forget that they slightly increased the transistor density over RV670.

I also recall hearing that RV770 was pad limited in early stages.

[Jacky]

Nobody, had the faintest clue about it having 800 SPs until the day it was launched and its official specs were put up. "NVIDIA are smart, powerful, brilliant, telepathic and knew everything at ATI" is a frail argument, not supported by anything that went on between Q2~Q3 2008.

Just from common sense I'd have to go with Calmatory on this one. You don't need to be ignorant of the other one's chip to be surprised by it, or to state that you were surprised about details of their launch plan ("surprise" has a lot of meainings -- good yields, high clocks and an early launch can be a surprise even if you know the arch very well).
Surely, AMD was surprised by Intel's Conroe, but it doesn't mean they haven't seen the engineering samples long before we did. The design cycles are so long, it really doesn't matter if you get to see the engineering samples, because you can't do much about your own product.

[orangekiwii]

The anticipation!

GTX380 vs 5870(x2)

MUST KNOW!

GTX380 will prevent me from buying a card until I know how it performs... so it is doing its job (in a sense)

[Calmatory]

I thought so, too, until I counted them. Turned out GDDR5 has 62 I/O pins per chip while GDDR3/4 has 69. GDDR5's has more pins total but that's just because it has 100 power/ground pins, while GDDR3/4 has 70.

So in the end, GDDR5 indeed uses more pins no?

Edit: Ah, obviously the power/ground pins don't go trough the GPU. Nevermind.

[Helloworld_98]

The anticipation!

GTX380 vs 5870(x2)

MUST KNOW!

GTX380 will prevent me from buying a card until I know how it performs... so it is doing its job (in a sense)

you'll be waiting a while for a GTX380 at this rate, or you'll be paying more than $1000 for it.

[Your_Boss]

Or you can get one of the 7 they taped out! $10,000 please!! LOL

THE SAGA of Nvidia's GT300 chip is a sad one that just took a turn for the painful when we heard about first silicon yields. Nvidia's execution has gone from bad to absent with low single digit yields.

A few weeks ago, we said that Nvidia was expecting first silicon back at the end of the week, the exact date was supposed to be Friday the 4th plus or minus a bit. The first bit of external evidence we saw that it happened was on the Northwood blog (translated here) and it was a day early, so props to NV for that. That lined up exactly with what we are told, but the number of good parts was off.

The translation, as we read it, says there were nine good samples that came back from TSMC from the first hot lot. That is below what several experts told us to expect, but in the ballpark. When we dug further, we got similar numbers, but they were so abysmal that we didn't believe it. Further digging confirmed the numbers again and again.

Before we go there though, lets talk about what a good die is in this case. When you get first silicon back, it almost always has bugs and problems. First silicon is meant to find those bugs and problems, so they can be fixed in succeeding steppings.

By 'good', we mean chips that have no process induced errors, and function as the engineers hoped they would. In other words not bug free, but no more errors than there were in the design. 'Good' in this sense might never power on, just that the things that came out of the oven were what was expected, no more, no less.

Several experts in semiconductor engineering, some who have overseen similar chips, were asked a couple of loaded questions: What is good yield for first silicon? What is good yield for a complex chip on a relatively new process? The answers ranged from a high of 50% to a low of 20% with a bunch of others clustered in the 30% range. Let's just call it one-third, plus or minus some.

The first hot lot of GT300s have 104 die candidates per wafer, with four wafers in the pod Nvidia got back a week and a half ago. There is another pod of four due back any day now, and that's it for the hot lots.

How many worked out of the (4 x 104) 416 candidates? Try 7. Yes, Northwood was hopelessly optimistic - Nvidia got only 7 chips back. Let me repeat that, out of 416 tries, it got 7 'good' chips back from the fab. Oh how it must yearn for the low estimate of 20%, talk about botched execution. To save you from having to find a calculator, that is (7 / 416 = .01682), rounded up, 1.7% yield.

Nvidia couldn't even hit 2%, an order of magnitude worse than the most pessimistic estimate. Ouch. No, just sad. So sad that Nvidia doesn't deserve mocking, things have gone from funny to pathetic.

At this point, unless there's a massive gain in yields on the second hot lot, there might not be enough chips to do a proper bring up and debug. This stunningly bad yield could delay the introduction of the chip, adding to the current pain and bleak roadmap. If there aren't enough 'good' parts from the second hot lot, that might require running another set, adding weeks to the total. Q1? Maybe not.

It is going to be very interesting to see what Nvidia shows off at 'Not Nvision' in a couple of weeks. Will it give the parts to the engineers to work on, or show them off as a PR stunt? We will know soon enough. In any case, the yields as they stand are sub-2%, and the status of the GT300 is far worse than we had ever imagined.

http://www.semiaccurate.com/2009/09/...eilds-under-2/

[zerazax]

I can't remember the exact reason behind the pad limit then, if it wasn't the too small die. There was a review or an article about it, and RV770 was supposed to have 480 SPs, but was pad limited so they had "free space" to fill up and slapped 800 SPs to it. GDDR5 uses more pins which caused the bigger die to be pad limited, as far as I can recall.

As largon wrote, GDDR5 has fewer pins

That said, that reason was in hindsight. If you recall, at the time before the actual release of the RV770, people were wondering whether ATI could even fit 2.5x more shaders in ~30% more space - pad limitations weren't an issue. For RV740 though... pad limiting was definitely an issue brought up

[Lightman]

Correct, GDDR3/GDDR5(GDDR4?) memory controllers, more pins for the GDDR5 due to power/ground, sideport taking up extra perimeter, UVD2 is slightly larger than the original. Let's not forget that they slightly increased the transistor density over RV670.

I also recall hearing that RV770 was pad limited in early stages.

It was pad limited with planned 8 clusters (so not 480SP but 640SP were designed at first). This also answers only 32 Texture Filtering units and 40TU, originally planned was 32/32 and not 32/40 for RV770. All this because some brilliant engineers managed to redesign/optimize SPU so they were taking a lot less space than original RV670 ones and still provided more capabilities.

If you dig Beyond3D you will find Dave himself told this, but I can't be bothered to look for quotes now (too late)

[Chumbucket843]

you'll be waiting a while for a GTX380 at this rate, or you'll be paying more than $1000 for it.

if g300 yields are 2% then rv870 is not too much better off. its tsmc's fault here considering g300 is smaller than gt200.

[Calmatory]

if g300 yields are 2% then rv870 is not too much better off. its tsmc's fault here considering g300 is smaller than gt200.

Oh please

What are the die sizes, and sources for them? AMD has aleady plenty of experience with TSMC's 40 nm process, thanks to RV740.

More than likely that the 2 % yield number is FAR off, but it indicates that yields are LOW. Most probably the chip design by Nvidia has flaws, which they need to sort out and fix, then re try and see how it turns out. Or fix the issues and take a risk; mass produce without testing.

[003]

if g300 yields are 2% then rv870 is not too much better off. its tsmc's fault here considering g300 is smaller than gt200.

Correct, G300 and RV870 are on the same process. And G300 isn't that much bigger than RV870.

This is of course going on that assumption that the 2% yield garbage spewing from Charlie is true, which it probably isn't.

[~CS~]

if g300 yields are 2% then rv870 is not too much better off. its tsmc's fault here considering g300 is smaller than gt200.

Alot of people have no clue what was the puprose of RV740 , AMD needed to know how many transistors they can safely jam into 40nm TSMC process while having good yields , RV870 is coming , AMD done the homework that NV did not .

[003]

Alot of people have no clue what was the puprose of RV740 , AMD needed to know how many transistors they can safely jam into 40nm TSMC process while having good yields , RV870 is coming , AMD done the homework that NV did not .

Nvidia has other GPUs being produced on the 40nm process but they are sold in the OEM channels mainly as mobile parts.

[Final8ty]

if g300 yields are 2% then rv870 is not too much better off. its tsmc's fault here considering g300 is smaller than gt200.

Correct, G300 and RV870 are on the same process. And G300 isn't that much bigger than RV870.

This is of course going on that assumption that the 2% yield garbage spewing from Charlie is true, which it probably isn't.

Assuming that all designs shrink just as well & as easily as each other on each process, as your saying the design has no bearing on the shrinking process.

[zerazax]

Correct, G300 and RV870 are on the same process. And G300 isn't that much bigger than RV870.

This is of course going on that assumption that the 2% yield garbage spewing from Charlie is true, which it probably isn't.

It's probably not true, but I have to correct one point: You can't assume that the yields are going to be about the same just based on being on the same process, especially when you don't have any idea what GT300's die size is, much less the design and procedure that Nvidia gives to TSMC to produce things (yes, it's on Nvidia to give them the layers etc.) nor the testing procedures behind it

Remember folks... increased die size doesn't increase defects linearly... it increases it exponentially

[Chumbucket843]

Oh please

What are the die sizes, and sources for them? AMD has aleady plenty of experience with TSMC's 40 nm process, thanks to RV740.

More than likely that the 2 % yield number is FAR off, but it indicates that yields are LOW. Most probably the chip design by Nvidia has flaws, which they need to sort out and fix, then re try and see how it turns out. Or fix the issues and take a risk; mass produce without testing.

sorry about that i didnt realize that was a demerijan article aka garbage.why do news sites allow that crap to even be published? lets hope the 40nm mobile graphics helped nvidia out with the real yield numbers.

[zerazax]

Nvidia has other GPUs being produced on the 40nm process but they are sold in the OEM channels mainly as mobile parts.

The same GT218 that has poorer performance than the RV710 while consuming more power on a smaller process?

You ought to look at this post by neliz on B3D, who has been 'in the know' for this past generation as he's called just about everything on the RV870:

http://forum.beyond3d.com/showpost.p...postcount=2151

If what a lot of B3D posters say are true, and a lot of them have been right time after time, then yes yields are very bad for Nvidia for a plethora of reasons from bad practices to design issues - probably not the 2% that Charlie is spouting but the issues *are* there

[LordEC911]

Since the same mis-info and assumptions are being posted in here I will quote myself...

#1- Charlie is NOT the source. A different source leaked some info and a very good informant even posted up the numbers Charlie was going to quote yesterday. This informant quoted similar troubles but slightly higher yields, still under 10%.

#2- G300 never was getting 20-25% yields, that was a compeltely made up rumor based on RV740 yields.
G300 was either just taped out or not even tapedout 3 months ago...

#3- G300 is supposedly around G200 size, even though it is on 40nm vs 65nm.

[Chumbucket843]

Assuming that all designs shrink just as well & as easily as each other on each process, as your saying the design has no bearing on the shrinking process.

coming from the guy who thinks cuda doesnt count as an api. nvidia has a 40nm chip too. we already know that gt200 could not scale down to 40nm so nvidia probably has scaling in mind with g300.with that said there still could be issues with the chip.

[Final8ty]

coming from the guy who thinks cuda doesnt count as an api. nvidia has a 40nm chip too. we already know that gt200 could not scale down to 40nm so nvidia probably has scaling in mind with g300.with that said there still could be issues with the chip.

So your making just as many assumptions as the next guy.

[LowRun]

sorry about that i didnt realize that was a demerijan article aka garbage.why do news sites allow that crap to even be published? lets hope the 40nm mobile graphics helped nvidia out with the real yield numbers.

Maybe because he was so right last time about Nvidia's defective chips for example

[Jamesrt2004]

sorry about that i didnt realize that was a demerijan article aka garbage.why do news sites allow that crap to even be published? lets hope the 40nm mobile graphics helped nvidia out with the real yield numbers.

Originally Posted by LordEC911 View Post
#1- Charlie is NOT the source. A different source leaked some info and a very good informant even posted up the numbers Charlie was going to quote yesterday. This informant quoted similar troubles but slightly higher yields, still under 10%.

#2- G300 never was getting 20-25% yields, that was a compeltely made up rumor based on RV740 yields.
G300 was either just taped out or not even tapedout 3 months ago...

#3- G300 is supposedly around G200 size, even though it is on 40nm vs 65nm.

[cegras]

coming from the guy who thinks cuda doesnt count as an api. nvidia has a 40nm chip too. we already know that gt200 could not scale down to 40nm so nvidia probably has scaling in mind with g300.with that said there still could be issues with the chip.

Rather naive of you to invalidate his perfectly logical statement based upon what he previously said. Do you not agree with the logic? You're saying that as long as die size is similar on the same process, yields should be uniform across both architectures?