No New GPUs from AMD for the Bulk of 2013

[TheMailMan78]

I think AMD just showed the hand of the industry as a whole. Mobile and console are the future and pursuing a shrinking market (desktop PC's) aggressively isn't a smart move. NVIDIA has to anymore. They hold no real advantage at this point except in the dedicated GPU market.

[Crap Daddy]

Originally Posted by TheMailMan78

I think AMD just showed the hand of the industry as a whole. Mobile and console are the future and pursuing a shrinking market (desktop PC's) aggressively isn't a smart move. NVIDIA has to anymore. They hold no real advantage at this point except in the dedicated GPU market.

I've been preaching this whenever I have the opportunity. NV is testing the wallet of the small niche of enthusiasts to see how many limited Lamborghini type of cards can they sell while trying to get as much money for as long it's possible from their mid and low range cards. AMD has no Ferrari to compete but they have good cards for a fair price plus glorious bundles to keep them in the business until they figure out if the desktop PC is going extinct.

[cadaveca]

Originally Posted by okidna

Since Catalyst 13.2 BETA 3 : http://support.amd.com/us/kbarticles...etaDriver.aspx



TR did a review for this driver : http://techreport.com/review/24218/a...on-frame-times

that's just a fix for those three apps specifically.

[Mindweaver]

I agree with AMD not making a new card until next year, but something's not adding up.. Why are the high end 7xxx cards drying up at Newegg? Has anyone see there stock on any version of 79xx's? They are slim to none..

To me this looks like AMD is getting ready to release the 8XXX cards. This also, looks like something to get everyone that's holding off for the next gen to go ahead and buy now. If you look at the stock of nvidia cards at newegg you can see they still have a ton of cards.. AMD maybe pulling a fast one on Nvidia.. If AMD releases the 8XXX cards and they are a lot faster than nvidia's 6xx cards for the same price, then nvidia will be forced to drop the price on those cards and take a big lose.. With AMD in all of the new consoles they can afford to do this... It reminds me of the sleeping dragon quote.. Only time will tell now.

[TheMailMan78]

Originally Posted by Mindweaver

I agree with AMD not making a new card until next year, but something's not adding up.. Why are the high end 7xxx cards drying up at Newegg? Has anyone see there stock on any version of 79xx's? They are slim to none..

To me this looks like AMD is getting ready to release the 8XXX cards. This also, looks like something to get everyone that's holding off for the next gen to go ahead and buy now. If you look at the stock of nvidia cards at newegg you can see they still have a ton of cards.. AMD maybe pulling a fast one on Nvidia.. If AMD releases the 8XXX cards and they are a lot faster than nvidia's 6xx cards for the same price, then nvidia will be forced to drop the price on those cards and take a big lose.. With AMD in all of the new consoles they can afford to do this... It reminds me of the sleeping dragon quote.. Only time will tell now.

They are drying up because they are rebranding them as the 8xxx cards.

[Xzibit]

Originally Posted by qubit

No, that's not right. The GK110 has 15 SMX units and is used in the K20 card.

The apparently cut down version used in the upcoming Titan has one disabled, making 14 SMX units.

NVIDIA's official whitepaper can be downloaded from here, which shows this.

Might want to read more.

15 SMX was the design the original silicon had. The one Nvidia taunted when it announced GK110. Nothing with 15 SMX has come out not K20X or K20.

Even the white paper you link alludes to that but look up specs for K20X 14 SMX and K20 13 SMX none use 15 SMX. Dont know where you get K20 uses 15 SMX if K20X doesnt.

A full Kepler GK110 implementation includes 15 SMX units and six 64‐bit memory controllers. Different
products will use different configurations of GK110. For example, some products may deploy 13 or 14
SMXs.

There are a few other things in the Whitepaper just like the 15 SMXs that never made it to K20X and K20.

[yogurt_21]

simple reason, AMD has never had to deal with 3 console launches before, they likely don't have the resources to work on those AND release a new gen. SO they'll wait it out while working on the more profitable market.

[happita]

Well, this sucks. I was planning on a whole new build this summer. But I guess since the new consoles are coming at the end of the year and AMD being in both of them, I guess they are focusing more on them versus the PC market for now. Guess I'll have to stick with my 5850 for now, unless I can score a great deal on something else.

[TheMailMan78]

Originally Posted by happita

Well, this sucks. I was planning on a whole new build this summer. But I guess since the new consoles are coming at the end of the year and AMD being in both of them, I guess they are focusing more on them versus the PC market for now. Guess I'll have to stick with my 5850 for now, unless I can score a great deal on something else.

670 sir. Don't look back.

[okidna]

Originally Posted by cadaveca

that's just a fix for those three apps specifically.

Yup, sadly you're right

[cadaveca]

Originally Posted by okidna

Yup, sadly you're right

And with that said, I'd much rather see more drivers, than more cards from AMD right now.

[Ravenas]

Why should they release them before then? Console games are the driving force of GPU power requirements. The consoles will probably be available by the end of this year with next generation games. Those next generation games require better GPUs and the developers were be building engines around that. So need to rush these things... These comments about AMD being cash strapped ect... of course they are, but they are still going to be putting high quality affordable GPUs on the market. There is no reason to rush a product out when the consumer doesn't have a high demand for it yet.

[Frick]

Originally Posted by qubit

Good for you.

I wish I could buy it - I'd seriously geek out with a card like this.

I am indifferent.

[GC_PaNzerFIN]

Originally Posted by HumanSmoke

And you know this how? Do you work for Nvidia or TSMC ? There have been precisely two SKU's based on the GK 110...one has 93% functionality, and the other 86.7%. Using your logic, Tahiti's yields must also be poor, since the split is 100%, 87.5%, and 75%.

Now now don't start putting words in others mouths. You got that logic part from your own head. Since this subject obviously needs clarifying...

I know it because I have been studying computer architecture and processor design as in HW for a very long time.

Do you have any idea how hard it is to manufacture a 7 billion transistor GPU which is well over 500mm2 in die area? Especially how the defect rate/yield relates to the die area? Yeah I guess you don't. You don't have to work for NVIDIA for that btw...

There is no fully enabled GK110 out and there won't be for a long time. Why? Because the amount of fully working chips from these pre-launch risk wafers is too low to launch a product with mass availability without making unacceptable trade of amount of GPUs you can sell and the cost of manufacturing them. Even GK104 and Tahiti get a lot partially defective GPUs from the Fabs just look at the amount of them being sold. And those are more than third smaller than GK110 die. As die size increases not only are you getting less chips out of the wafer in best scenario but the change there are multiple defects on a single chip goes up too and fast.

After a long period of time which is many months the yields will improve and stabilize as the manufacturing process matures for that exact product. This is especially true for gigantic ICs. At this point you'd generally start selling the stockpile of partially working ICs too because there is plenty in storage from manufacturing up to this point.

What I am in fact saying is quite the opposite you thought as my logic. Tahiti, and GK104 for that matter, are less prone to multiple defects per chip resulting in better yields.

[Wrigleyvillain]

Just bought 7950 so all fine and dandy with me (and I agree with Dave re. driver work instead).

[badtaylorx]

im ok with this so long as AMD puts out more cards along the line of the Tahiti LE class cards. I was quite impressed with that lil' guy. I'd really like to see an HD 7890 Tahiti LEx2 put out.

3072 Stream Processors at a price-point under $500...

anyone else up for that???

[Casecutter]

Originally Posted by Sasqui

They either landed on a leap-frog development that will really impress by late '13 or early '14 or…

Originally Posted by TheMailMan78

I think AMD just showed the hand of the industry as a whole. Mobile and console are the future and pursuing a shrinking market (desktop PC's) aggressively isn't a smart move. NVIDIA has to anymore. They hold no real advantage at this point except in the dedicated GPU market.

Originally Posted by Crap Daddy

I've been preaching this whenever I have the opportunity. NV is testing the wallet of the small niche of enthusiasts to see how many limited Lamborghini type of cards can they sell while trying to get as much money for as long it's possible from their mid and low range cards. AMD has no Ferrari to compete but they have good cards for a fair price plus glorious bundles to keep them in the business until they figure out if the desktop PC is going extinct.

With this news of waiting it out, I kind of want to think AMD made GCN and Boost work really super together, something they may partially implemented in console parts and why they got the contracts. While game developer release new titles for those console, AMD knows they'll have a leg up with GCN drivers and buying-time for those releases to port to the PC platform. While I think AMD realizes Kepler (GK104) is at this point fairly tapped out. Nvidia can pick-up modest gains with a re-spin, but without an overall revamp (major investment) and perhaps even die size bump they won’t achieve the next hurdle. If AMD released in say 3 months they’d take the lead price/performance, but why? This all may have to do that they are hope to "dig-up" some information as to what Nvidia could do with Kepler, it been fairly quiet on that.

Yes, It may come down to slowing development, resources, manpower, tied to a shrink market. IDK It could be that pulling back is related to both Volcanic Islands, and Nvidias’ Maxwell and 20Nm manufacturing. But overall AMD probably knows Nvidias' path is 8+ months out, so why not slow it down. Last time they moved early and it brought more grief on them.

Lastly this has almost nothing to do with Titian, except Nvidia (and AMD) wants to see how many they can sell, and can they elevate the market in future go arounds. Proving there might be enough takers in a $600-750 range to bump normal "enthusiast" price point fo Maxwell in 2014-2015. This will permit AMD to see if Nvidia might bring other lower derivatives from further gelding GK110’s. That could give AMD an idea if Nvidia could in fact bump the GK104 die with a slight revamp while hitting good gains? I really believe AMD priced the 7970 because they really thought Nvidia would be onboard, but they pulled a fast one on AMD by getting to work from the GK104.

This is AMD being smart... for all the right reasons.

[HumanSmoke]

Originally Posted by GC_PaNzerFIN

There is no fully enabled GK110 out and there won't be for a long time. Why? Because the amount of fully working chips from these pre-launch risk wafers is too low to launch a product with mass availability without making unacceptable trade of amount of GPUs you can sell and the cost of manufacturing them.

Nvidia began supplying ORNL with GK 110 based Tesla's (93% functional) in the first week of September. Nine weeks later ORNL completed a Linpack run using 18688 of these GPUs (2000+ per week). This in an atmosphere where Nvidia's wafer starts were primarily aimed at fulfilling OEM mobile ( the GT 625M, GT 645M, GTX 670MX, GTX 675MX and GTX 680MX were all released during the same timeframe), as well as consumer GK 106 and GK 104 variants.

While a larger die invariably will contain more defects than a smaller one, no one here- not even you - know what the defect rate is, and more importantly, how much redundancy is built into the GPU die to allow for defects. So, while I would be agreement with you over the likelihood of a lower yield as die size increases, I am not in agreement with absolute statements that aren't supplied with corroborative proof....

Originally Posted by GC_PaNzerFIN

I can guarantee you there will be like a ton of not-good-enough-for-Titan GK110s. Its massive chip with poor yields

I'd also contend that "massive" is a relative term. I don't think the GK 110 would be appreciably larger than GF 110 - and if anyone has experience in producing large complex GPUs, it's probably Nvidia...the 20,000+ units shipped from the first A1 revision are probably testimony to that.

[GC_PaNzerFIN]

Originally Posted by HumanSmoke

Nvidia began supplying ORNL with GK 110 based Tesla's (93% functional)...

Again you prove my point. None of sold was fully enabled GK110.

[erocker]

Originally Posted by GC_PaNzerFIN

Again you prove my point. None of sold was fully enabled GK110.

I think I lost what the point was.

[GC_PaNzerFIN]

Originally Posted by erocker

I think I lost what the point was.

That the 1000 dollar Titan is not the only GK110 to get into markets this year. And that you only disable parts when you have yield problems.

[erocker]

Originally Posted by GC_PaNzerFIN

That the 1000 dollar Titan is not the only GK110 to get into markets this year. And that you only disable parts when you have yield problems.

Ah, I see. Many of us believe that the current cards (6 series) are the derivative of GK110 already. But really this is for a different thread. Discuss here: http://www.techpowerup.com/forums/sh...d.php?t=179919

This thread is about discussing what AMD is going to do this year (or not do, lol).

Carry on.

[Goodman]

That doesn't surprise me as AMD are pretty low on cash (for a company that size) best thing for them to do is use what they already have make it better & still sale it at better profit

The other thing is i don't see the point of new GC's either from AMD or Nvidia not until we see screens/monitors getting mainstream with higher resolution then 1080 , no really don't we have this resolution for long enough already? (for PC)

[Lionheart]

Either AMD is too busy with the next gen consoles or they're chucking a Sony

Sony CEO Kazuo Hirai - "Why go first, when your competitors can look at your specifications and come up with something better?”

Week later, Sony show's off a teaser trailer of the most likely PS4 announcement Feb 20th trolling Microsoft & surprising everyone

[TheMailMan78]

Originally Posted by Lionheart

Either AMD is too busy with the next gen consoles or they're chucking a Sony

Sony CEO Kazuo Hirai - "Why go first, when your competitors can look at your specifications and come up with something better?”

That didn't help them with the last generation at all.