"nVidia Denies Plans to Ship Kepler in 2011"

[thilanliyan]

http://www.xbitlabs.com/news/graphi...nies_Plans_to_Release_Kepler_GPU_in_2011.html

Guess this was wrong?
http://www.xbitlabs.com/news/graphi...o_Start_Shipping_Kepler_GPUs_by_Year_End.html

 

[bryanW1995]

That's surprising. If they expected anything at all this year then they could have let that statement slide. We're probably looking at mid 2012 for kepler at this point, and I wouldn't be surprised if AMD misses 2011 at least, either.

This is all a big fail for both amd (graphics) and nvidia b/c with BD coming out this year and SB-E early next year they could get plenty of sales if they just have something, anything, new to offer.

 

[BenSkywalker]

Seems to me that it would make a lot more sense to use Tick-Tock this generation then any of the previous, this is a full node drop which I don't recall ever having seen before in the graphics industry. I see anything outside of Tick-Tock being very dangerous for smooth launches no matter what company is making it.

I haven't seen anything, not even so much as rumors, stating that is what nVidia is doing, but to me it seems the only purely logical path to take.

 

[Idontcare]

Seems to me that it would make a lot more sense to use Tick-Tock this generation then any of the previous, this is a full node drop which I don't recall ever having seen before in the graphics industry. I see anything outside of Tick-Tock being very dangerous for smooth launches no matter what company is making it.

I haven't seen anything, not even so much as rumors, stating that is what nVidia is doing, but to me it seems the only purely logical path to take.

That is a remarkably astute observation there BenSkywalker :thumbsup:

I had not contemplated this fact before. You are right, all the way up to 40nm the foundry customers, as well as the foundry's fab engineers, had the benefit of half-node incremental changes to help them climb up the requisite learning curve.

(55nm to 40nm was really a half-node as TSMC's "40nm" was little more than a relabeling of their late and delayed "45nm" node...the actual 40nm half-node that was on the books became the new 32nm full-node, and the 32nm that was on books got shuffled down and relabled as the new 28nm half-node, then they canceled 32nm, which meant they cancelled the old 40nm)

So this newest node transition, the 40nm -> 28nm, is really the first time TSMC and their customers have had to climb up the technological complexity ladder with a rung-spacing being a whole entire node. Companies like AMD and Intel have been doing full-node strides for decades, so no wonder that AMD is able to take the 40nm->28nm step in one bound without a hitch.

But this is new territory for Nvidia in terms of developing the risk management tools that are as robust as the ones AMD has had for decades.

I had not thought about this at all until I saw your post Ben, nicely stated!

 

[Arkadrel]

That's surprising. If they expected anything at all this year then they could have let that statement slide.

^key point above. Wont be any kepler shipped in 2011,
and it takes a few months after things are shipped before you can launch a product.

Basically this is Nvidia saying:
You wont be able to find Kepler in a shop, until 2012 Feb or lateron.

*IF*

AMD managed to keep their launch dates, it means they ll have a good 5-6 months lead again, just like back with the 5xxx series.

 

[Keysplayr]

^key point above. Wont be any kepler shipped in 2011,
and it takes a few months after things are shipped before you can launch a product.

Basically this is Nvidia saying:
You wont be able to find Kepler in a shop, until 2012 Feb or lateron.

*IF*

AMD managed to keep their launch dates, it means they ll have a good 6 months lead again, just like back with the 5xxx series.

I don't see what Nvidia's denial of launch in 2011 has to do with AMD.

 

[Grooveriding]

^key point above. Wont be any kepler shipped in 2011,
and it takes a few months after things are shipped before you can launch a product.

Basically this is Nvidia saying:
You wont be able to find Kepler in a shop, until 2012 Feb or lateron.

*IF*

AMD managed to keep their launch dates, it means they ll have a good 5-6 months lead again, just like back with the 5xxx series.

Looks like it will be AMD all alone again with 28nm if all goes according to their plans. While nvidia is saying they can't make a 2011 launch, AMD is going ahead with a launch this year :thumbsup:

28nm Radeon Chips This Year? AMD Says Yes


Saddle up and if you're in the market for a next gen 28nm part later this year, looks like AMD will be the only game in town.

 

[happy medium]

Looks like it will be AMD all alone again with 28nm if all goes according to their plans. While nvidia is saying they can't make a 2011 launch, AMD is going ahead with a launch this year :thumbsup:

28nm Radeon Chips This Year? AMD Says Yes


Saddle up and if you're in the market for a next gen 28nm part later this year, looks like AMD will be the only game in town.

Where does it say it will have a product in retail in 2011? I read this..

"AMD claims it is on track to introduce a revamped lineup of GPUs codenamed “Southen Islands” using the new process later this year"

Am I missing something? Does" introduce" mean cards in the retail channel or paper launch?

 

[Vesku]

Perhaps this is extra caution after being so optimistic regarding TSMC's '40nm' node.

 

[thilanliyan]

Saddle up and if you're in the market for a next gen 28nm part later this year, looks like AMD will be the only game in town.

Even if AMD launches a 28nm part in 2011, it doesn't mean it will be high-end, which is what most people here would be wanting to happen (ie. they want to see huge performance gains). It could just be a tester card to get kinks worked out on the new process a la 4770 on 40nm.

 

[SirPauly]

To be fair, these times coming up is sort of a silly season to me as next generational hardware is getting close and gamers getting excited. What PC gamer in their right minds don't get excited about what the engineers of nVidia and AMD create for their customers?

Not knowing is tough at times!

 

[nemesismk2]

it's a shame nvidia will have no new video cards this year but they have great video cards already available

 

[gorobei]

the amd announcement is vague on what "unveil" in 2011 means, but that's still somewhat better than "not shipping".

the more interesting part was that amd is using 28nm from both GF and TSMC. since the processes aren't interchangeable it does suggest a probable 28nm dieshrink of vliw4 from GF, and that they'll use TSMC's 28nm HP on GCN later on. I wonder if the dual development on both processes are responsible for the operating losses reported from the amd video division.

 

[ch33kym0use]

News is too jerkish about it. Apart from that, my computer is fast enough. Don't need new generation yet.

 

[Firestorm007]

I don't see what Nvidia's denial of launch in 2011 has to do with AMD.

Maybe the AMD 7 series scared them into a delay...

P.S.

I'm just busting your chops.

 

[AnandThenMan]

I don't see what Nvidia's denial of launch in 2011 has to do with AMD.

It certainly appeared that the initial Nvidia announcement on Kepler was in response to AMD.

 

[notty22]

I'm pretty sure AMD's quarterly report on the state of their graphics division combined with GPR shipping data was the impetus for AMD to issue its press releases. To appear optimistic about future products/revenue.

 

[XX55XX]

Whatever. Today's cards provide more than enough horsepower anyways.

I do hope Nvidia improves on the efficiency front, though.

 

[Firestorm007]

I'm pretty sure AMD's quarterly report on the state of their graphics division combined with GPR shipping data was the impetus for AMD to issue its press releases. To appear optimistic about future products/revenue.

You sir are downright hilarious! I must admit. I guess you are really sore that AMD will once again beat Nvidia to the market; so, please stop with the revenue report each time your team gets a smack down. It's getting really, really old....Plus, they don't have to be optimistic...they already dominate with the 6990....and before you say they cant be found, Tigerdirect has an XFX currently in stock; and before you repeat yourself again, just because the 590 beats it in Crysis 2...it doesn't mean its faster....

 

[Idontcare]

the amd announcement is vague on what "unveil" in 2011 means, but that's still somewhat better than "not shipping".

the more interesting part was that amd is using 28nm from both GF and TSMC. since the processes aren't interchangeable it does suggest a probable 28nm dieshrink of vliw4 from GF, and that they'll use TSMC's 28nm HP on GCN later on. I wonder if the dual development on both processes are responsible for the operating losses reported from the amd video division.

I know its popular in the "media" to state that the foundry processes are not "compatible" but I can say from firsthand experience that you can have the foundries align the processes such that your design is portable-enough to be produced on two, or more, foundries at the same time.

At Texas Instruments we did this for years, across many nodes, simultaneously producing the exact same mobile phone chips from three different foundries as well as in-house at the same time. Sometimes it was TSMC, UMC, and SMIC, others it was TSMC, UMC, and Chartered.

So don't be too quick to just assume AMD is not having both GF and TSMC produce the same product. It's not guaranteed but it is also not nearly as impossible as the rumor/news sites try to make it out to be.

 

[gorobei]

I know its popular in the "media" to state that the foundry processes are not "compatible" but I can say from firsthand experience that you can have the foundries align the processes such that your design is portable-enough to be produced on two, or more, foundries at the same time.

At Texas Instruments we did this for years, across many nodes, simultaneously producing the exact same mobile phone chips from three different foundries as well as in-house at the same time. Sometimes it was TSMC, UMC, and SMIC, others it was TSMC, UMC, and Chartered.

So don't be too quick to just assume AMD is not having both GF and TSMC produce the same product. It's not guaranteed but it is also not nearly as impossible as the rumor/news sites try to make it out to be.

i thought mask costs comprised a large portion of the manufacturing budget (aside from wafers). for large production runs where you need that volume i can see contracting with multiple foundries and duplication of masks, but wouldnt video card runs be significantly smaller that multiple duplicate sets of masks would send it over budget?

 

[Idontcare]

i thought mask costs comprised a large portion of the manufacturing budget (aside from wafers). for large production runs where you need that volume i can see contracting with multiple foundries and duplication of masks, but wouldnt video card runs be significantly smaller that multiple duplicate sets of masks would send it over budget?

Obviously there is a cross-over point, a volume threshold, where it makes economic sense to pursue a diversified foundry strategy. There is also a risk-mitigation reason to do it as well.

However there is also a cost-savings benefit. Early on it helps to have access to multiple foundries in terms of getting all the more wafers from the limited bleeding-edge capacity. Two 28nm fabs, albeit with 1000wfr/mnth capacity each, is still better than one.

But later on as capacity builds out in both foundries it also helps to have them negotiate off of each other, a sort of reverse auction if you will.

Without question our wafer prices at TSMC were low(er) because we brought in the fact that UMC (or other) had capacity for us to use if we wanted.

Its just business. But you need to be a large(r) player to do this, you can't be ordering 500 wfrs per month and expect to command the kind of resources necessary to align process technology across foundries. Nvidia probably doesn't have those kinds of volumes, maybe they do, they are right at the edge IMO.

But AMD certainly does, not just because of GPUs but also because of CPUs as well as their APUs. AMD could easily drive an alignment requirement.

(mind you, by alignment requirement I don't mean the two foundries are driven/guided to offer the exact same process tech, rather they are guided to meet the exact same electrical spec as needed by the customer, Idrives, metal resistance, current capacity, etc...we refer to it as "electrical equivalence")

 

[SirPauly]

I'm pretty sure AMD's quarterly report on the state of their graphics division combined with GPR shipping data was the impetus for AMD to issue its press releases. To appear optimistic about future products/revenue.

nVidia is solid with discrete but in over-all shipments, AMD ships many more GPU's than nVidia. The APU is formidable, imho.

 

[notty22]

nVidia is solid with discrete but in over-all shipments, AMD ships many more GPU's than nVidia. The APU is formidable, imho.

Are you forgetting Nvidia's workstation gpu's? AMD is actually fighting back in this area, but Nvidia still ships over 83% share.
As to actual #'s who is shipping more, when adding mobile,desktop, workstation, I'm not sure. Its estimated over 1/3 of Nvidia worth is from workstation gpu's.
http://www.trefis.com/company?hm=NVDA.trefis#
http://ucvox.wordpress.com/2011/05/19/workstation-graphics-2/

 

[tviceman]

AMD has not announced what 28nm parts will be delivered - and judging by their history they tend to release smaller, less powerful chips ahead of high parts. So while it's certain they'll be beat Nvidia to 28nm, it's speculation and only speculation at this point to talk about with what products.

Also, Nvidia "shipping" Kepler in 2011 to me does not mean the same thing as releasing the part. To me, when I see shipping it means that the product is finished and is being shipped to AIB's. Released means it's available to buy.