NVIDIA GF100 & Friends speculation

[Sontin]

Because the rumor is, the 512SP part is supposed to have 200+ mhz higher core clocks(if that is true) over the 448SP part. Not to mention 48ROPs vs 40 and 384bit bus over 320.

Not really, because this is the GTX 470. The GTX480 would be a 448 part with higher clocks and 48 ROPs. So the difference between the 448 and the 512 chip is only 14,2%.

 

[XMAN26]

Not really, because this is the GTX 470. The GTX480 would be a 448 part with higher clocks and 48 ROPs. So the difference between the 448 and the 512 chip is only 14,2%.

Why would the 448SP part be the GTX480 and the 512SP part be the GTX470? That makes absolutely no sence what so ever.

And initial rumors had the 448 part at around 400 core and the 512 part around 600.

 

[Sontin]

Why would the 448SP part be the GTX480 and the 512SP part be the GTX470? That makes absolutely no sence what so ever.

And initial rumors had the 448 part at around 400 core and the 512 part around 600.

I throught there will be no 512 part as real product. So why should they use all chips only for the GTX470 cards with less than 48 rops and lower clocks? You said they must disable two SM but that has nothing to do with the ROPs and the clocks...

 

[XMAN26]

I throught there will be no 512 part as real product. So why should they use all chips only for the GTX470 cards with less than 48 rops and lower clocks? You said they must disable two SM but that has nothing to do with the ROPs and the clocks...

No, I said they were getting mostly 448 parts of the line and that 512 part were nearly non existent and that they were planning a B1/respin to try and fix that. THEY want to have 512SP part out for sale.

 

[aaronspink]

The thing is, the delays that affected the soon-to-be-released part are unlikely to have effected the refresh parts. It is likely, therefore, that the refresh parts will come in rather short order (either this fall or next spring).

The point being Nvidia doesn't have much room to do a refresh for a big. They are already die size limited and there won't be another process ready. Refreshes tend to consist of things like shrinks or increased die sizes or moderate to significant frequency bumps. Which is also out because they are very power limited.

What I'm saying is they don't have a lot of wiggle room on the engineering side to do a refresh at least until the next process step.

 

[Sontin]

No, I said they were getting mostly 448 parts of the line and that 512 part were nearly non existent and that they were planning a B1/respin to try and fix that. THEY want to have 512SP part out for sale.

You said that 99% of the chips will be 448 parts. So why should they use all of these chips for a lower clocked 448 card with 40 ROPs, when they can use a lot of them for 448 part with clocks like the 512 version, 48 rops and 1,5gb?

 

[XMAN26]

You said that 99% of the chips will be 448 parts. So why should they use all of these chips for a lower clocked 448 card with 40 ROPs, when they can use a lot of them for 448 part with clocks like the 512 version, 48 rops and 1,5gb?

Man qouting myself can be a pain, but here it goes.

http://forum.beyond3d.com/showpost.php?p=1392678&postcount=3835

That is my post and here is what I said. I think you misread or understood what I said.

They plan to go ahead and launch with the 448SP part. The reason for the respin/B1 is because 99% of what they have been getting off the line has been 448SP parts. As it stands right now, if they leave things as they are, the 512SP part will be a PE part only and most likely never see store shelves. This is NOT what they want, so they are going for a B1 sooner than anyone would expect.

"The reason for the respin/B1 is because 99% of what they have been getting off the line has been 448SP parts"

Does not say 99% of teh chips will be 448 parts. It states simply that 99% of what has been coming from TSMC has been 448 parts.

 

[Sontin]

You wrote this:

Because the rumor is, the 512SP part is supposed to have 200+ mhz higher core clocks(if that is true) over the 448SP part. Not to mention 48ROPs vs 40 and 384bit bus over 320.

So the first thing is that nVidia must disable two sm to use 99% of the chips and they will use them only for the GTX470 part.
My question was why would they do that? They can use many of them for a GTX475 with higher clocks and more rops than the GTX460 card.

 

[XMAN26]

You wrote this:



So the first thing is that nVidia must disable two sm to use 99% of the chips and they will use them only for the GTX470 part.
My question was why would they do that? They can use many of them for a GTX475 with higher clocks and more rops than the GTX460 card.

Dont know. All I can say is what I have heard. And what I've heard is 448 part are a plenty and the 512s are scarce and they are not the least bit happy over it.

 

[dnavas]

What I'm saying is they don't have a lot of wiggle room on the engineering side to do a refresh at least until the next process step.

And if the double-precision limiting is a driver hack in search of a hardware reflection? I don't know if they could save trannies by removing Tesla-specific features, and I'm not sure it would be wise if they did, but would that not give them some room?

FWIW, I would be surprised to see a refresh prior to the money-makers shipping.... Shipping their high-end is only the start of this release cycle, no?

 

[trinibwoy]

Good to see we're helping to ask those tough GF100 questions

All of the questions below come from AMD user forums such as Rage3D, Beyond3D and so on. We've forwarded the link to nVidia and expect to get an answer during this week.

http://www.brightsideofnews.com/new...-has-29-tough-gf100-questions-for-nvidia.aspx

 

[Groo The Wanderer]

A3 went to the fab according to rumors in early December; it seems like they started mass production within February which makes sense. A hypothetical release in March makes sense for all the above and there doesn't seem to be any time for another metal spin (what for anyway?) let alone a full re spin which you'd rather expect somewhere in mid summer and not as early as March.

Production silicon has been at NV for a few weeks. It is in very bad shape, awful in fact.

-Charlie

 

[Groo The Wanderer]

April is when they are sending it in, again, they are "NOT" happy wth the number of working 512SP parts they are getting currently and want to improve those numbers. B1 stepping is, from my understanding a test run to see if they fix what is causing the 512SP parts to fail and become 448SP parts. They dont want to go all summer without their top part out at all or as a PE part only til they get it coming off the line in numbers that they can sell.

They are not happy with the number of 448SP parts either, but that is not going to stop them from a triumphant launch.

-Charlie

 

[green.pixel]

All of the questions below come from AMD user forums such as Rage3D, Beyond3D

Somewhat poorly worded IMO.

 

[Groo The Wanderer]

Yeah, it's called "it's hard to get a 3+ billion transistor chip pushing the reticle limits of the process to yield significant quantities of chips without disabling some functional units".

It is even worse when your architecture is broken on top of crap yields.

-Charlie

 

[Richard]

Good to see we're helping to ask those tough GF100 questions

http://www.brightsideofnews.com/new...-has-29-tough-gf100-questions-for-nvidia.aspx

Where's my AMD CPU and ATI video card?!? All reputation, no gain.

 

[John021]

Smelling a top of the line GTX580 by august. Man that thing is gonna ROCk!

 

[mczak]

Does not say 99% of teh chips will be 448 parts. It states simply that 99% of what has been coming from TSMC has been 448 parts.

But assuming somewhat normal error distribution, if 99% of the working parts only qualify as 448 shader parts, I would expect that something around 90% doesn't work at all...
Unless there's some very specific problem only affecting some simds (e.g. like bogus clock routing or whatnot).

 

[trinibwoy]

Really? You said the same about the gt200b.

http://www.theinquirer.net/inquirer/news/1018019/nvidia-270-290-deep-trouble

Hmmm well let's hope he's even more wrong this time too, eh?

 

[XMAN26]

They are not happy with the number of 448SP parts either, but that is not going to stop them from a triumphant launch.

-Charlie

Dont know where you are getting that from, but your wrong. Sure the yields aren't what they'd like, but they are happy with the chip overall. They are pisses that 512SP parts aren't coming like they like/want but, but they hope to have a fix for that by mid summer with 512 and 448 parts, improved yields and less issues aswell.