No New GPUs from AMD for the Bulk of 2013 [TechPowerUp]

[ShintaiDK]

My God, man! There were never going to be any 20nm GPU's in '13. What are you on about? My point is Apple and 20nm are irrelevant to the situation now.

AMD can say that they never announced a release date for the next series, so it's not delayed. It's the same excuse we heard from nVidia last year. If AMD don't release their new desktop cards before quarter 4 2013, or until 2014, that's late.

A refresh doesnt mean a node change.

But the nodes do dictate the timeline between cards. Meaning they stretch the lifespan of exsiting cards. So it doesnt matter what AMD and nVidia announced or didnt announce. They both changed schedule from what they used to do. The ~2 year cycle is history and the new cycle is ~3 years. But still only 2 actual releases in between.

 

[CakeMonster]

So.. the interesting part is what AMD will be able to do on 28nm later this year. The 20nm delay shouldn't necessarily mean that the refresh on 28nm would be delayed, even though its understandable if they choose to delay it too, so that the 20nm wait won't be closer to two years after the 28nm refresh. I'm anxious to see what they have...

It kind of seems like both AMD and NV have a "truce" regarding the 28nm refreshes... they both delay them to milk the market. You can pretty much disregard Titan in this.

 

[blackwhiskers]

This almost registers as a good thing in my book. Developers of both games and drivers can sit down and squeeze the real juice out of the cards, and we know there's plenty in these things we're running these days.

of course, whether they actually try this squeezing (and proper one, not new, even more expensive forms of AA that use all resources for tiny improvements) is a different question. but at least driver-wise I can't help but be excited about what's in store for existing cards.

 

[CakeMonster]

This almost registers as a good thing in my book. Developers of both games and drivers can sit down and squeeze the real juice out of the cards, and we know there's plenty in these things we're running these days.

of course, whether they actually try this squeezing (and proper one, not new, even more expensive forms of AA that use all resources for tiny improvements) is a different question. but at least driver-wise I can't help but be excited about what's in store for existing cards.

I disagree, this sounds too much like the console excuses that are thrown around. It assumes that the developers are so lacking in talent that new hardware holds them back from making big driver improvements on the previous generation. The amount of extra juice that we get out consoles very late in their lifetimes is very small compared to what new hardware would bring. And similarly for graphics cards, which do not get major architecture overhauls with every generation, this can simply not be counted as a "good thing" in any context. A new release from AMD, which would simply be a refresh of 7xxx still on 28nm, would not require a driver overhaul that would take away significant resources for driver improvements of the current 7xxx series.

 

[fleshconsumed]

Look on the positive side guys, you'll be getting much better longevity out of your cards than you thought you would. The 6xx and 79xx series are turning out to be like the 8800 of the old times.

 

[Homeles]

Look on the positive side guys, you'll be getting much better longevity out of your cards than you thought you would. The 6xx and 79xx series are turning out to be like the 8800 of the old times.

Slowed technological progress isn't a good thing. Here's a better idea: stop being insecure. Who cares if there's faster hardware out there? Assess your own needs.

 

[tviceman]

I think it was already a foregone conclusion that there would be no 20nm GPU's until 2014, but in the interim I'll be shocked if Nvidia does not start refreshing the Kepler lineup by June. Refreshing / respinning a chip takes way less resources than significant architecture changes and has obvious benefits of increased perf/watt and better yields. Even the 200 series cards (which ran for a considerable amount of time) had new products with the gtx275, gtx285, gtx295 and also experienced a die shrink.

 

[moonbogg]

Some do, some don't. Those that don't, probably wouldn't be in the market for a new flagship AMD card either.

The bleeding-edge game isn't for blue-collars with 3 kids and a wife who likes designer bags and watches.

PC gaming isnt that expensive no matter what rig you like to have. Its more about if you are willing to pay than if you can pay, or if you think its worth it to pay.
For instance, I would not pay $25 for a hamburger. Could I? Sure, why not. But I don't because its a bad value in my eyes. Just because you CAN buy something doesn't mean you should go full retard and spring for it.

 

[Spjut]

Wasn't planning on upgrading this year, but not fun if we'll get new cards less often.

Can be a positive effect in terms of longer driver support though, and since the studios can't count too much on hardware advances, perhaps some will spend more time on optimisation.

Would be cool if we got DirectX12 soon and games could gain 20% or so performance from that

 

[pakotlar]

What's the point of a new gpu generation when your target consumer is running games designed for hardware that is 8 years old?

 

[blastingcap]

My God, man! There were never going to be any 20nm GPU's in '13. What are you on about? My point is Apple and 20nm are irrelevant to the situation now.

AMD can say that they never announced a release date for the next series, so it's not delayed. It's the same excuse we heard from nVidia last year. If AMD don't release their new desktop cards before quarter 4 2013, or until 2014, that's late.

Chill out. Okay?

As someone else said, there was no 8xxx on the boards in the first place except the mobile part. I'm just speculating that they tried to get more out of 28nm but there is only so much you can do at the margins without significantly increasing die size (thus lowering profits and moving TDP up which may not be feasible), changing architecture (but GCN is pretty set), or going for a die shrink (not going to happen). Driver tweaks and such can only get you so far, and you probably don't need to launch a refresh if all you are going to get is something like a 10% increase in speed, if that.

So at the end of the day with a declining PC market, and with consoles commanding a lot of attention right now (AMD engineers working with the consolemakers a lot, no doubt), AMD may have figured that it just wasn't worth it to issue another 28nm GPU series.

Also, the timing of GPUs-on-20nm will likely be delayed thanks to Apple, so there is not as much time pressure to squeeze in a 28nm GPU series in-between the 7xxx series and the 9xxx series, assuming 9xxx is 20nm.

 

[Silverforce11]

I think Apple had something to do with this. Demand for the latest TSMC 20nm process is huge and it looks like GPU makers refused to pay the asking price. I would not be surprised if NVDA also stuck to 28nm through most of 2013, possibly ALL of 2013. Without a die shrink and with just software/driver tweaks and hardware arch revisions and a maturing 28nm, I don't think AMD could do much better than what they already have, so they probably said screw it, we'll punt until TSMC 20nm becomes more affordable.

This. Smartphones dominate the market and they are all using TSMC. High end GPU market is not even comparable to the SoCs of smartphones in terms of $ or volume. Plus they are already taking most of the capacity on 28nm as it is.

 

[CakeMonster]

Also, the timing of 20nm will likely be delayed thanks to Apple, so there is not as much time pressure to squeeze in a 28nm GPU series in-between the 7xxx series and the 9xxx series, assuming 9xxx is 20nm.

Wouldn't that be the opposite? More than enough time to push out another 28nm, since 20nm will be even more delayed than 28nm was. They did a 40nm refresh with 6970, and they would have even more time for a 28nm refresh before 20nm.

 

[blastingcap]

Possibly. Other factors are in play though. Consoles demanding time, a declining PC market, wafer availability and costs at 20nm are a lot different if Apple steps in, than in previous years, and AMD being more on the ropes than before regarding profitability. New CEO too. AMD can't afford missteps at this point. Perhaps they feel that squeezing in a quick refresh at 28nm would be a misstep... like adding unnecessary expense.

 

[VulgarDisplay]

What if the reason AMD is delaying the refresh is because the memory management driver offers a big enough performance increase that they don't feel pressured at all by nvidia to release cards?

It would extremely cool if that's the case, but we'll just have to wait and see. I would imagine it will improve performance in some way. How much is the question.

 

[RussianSensation]

- Will Titan's release change anything about HD7000 vs. GTX600 series offerings in terms of price/performance? Maybe for the 1% of users going HD7970GE CF / GTX680 SLI. Otherwise, not really.

- Isn't GTX700 also rumoured to be delayed (losing the word loosely here since it was never announced on NV's roadmap officially) to Q4 2013 from the same sources? I am pretty sure I read that it was.

- If initial 28nm node prices were so expensive, wouldn't that suggest that early 20nm node prices would also be very expensive? Then if 20nm GPUs might only see the light of day by Q3-4 2014, if AMD/NV release 28nm refreshes now, people will complain just the same that it's taking too long between 28nm refreshes and 20nm GPUs. Releasing 28nm GPUs in Q4 2013 would allow NV/AMD about a year to get their 20nm designs in order for launch in the second half of 2014.

- Even if NV refreshes GTX680, since HD7970GE is already in the lead, how much faster can GK114 get without expanding the bus width? 10-15% faster than HD7970GE? But if NV bumps the speed of 680 by 20-25%, the Titan's 60% advantage for $900 would shrink to just 28-33% faster over GK114. That would make the Titan's price nearly impossible to justify if GTX680's GK114 successor is only $499. This scenario would piss off Titan's early adopters. Even if GK114 is 10-15% faster than HD7970GE and launches at $499, AMD could just drop the price of HD7970GE's from $430 to $399. Remember that HD7970GE is already way under $499 price level. If GK114 is 15% faster at $499, a $399 HD7970GE would still have superior price/performance.

 

[dangerman1337]

- If initial 28nm node prices were so expensive, wouldn't that suggest that early 20nm node prices would also be very expensive? Then if 20nm GPUs might only see the light of day by Q3-4 2014, if AMD/NV release 28nm refreshes now, people will complain just the same that it's taking too long between 28nm refreshes and 20nm GPUs. Releasing 28nm GPUs in Q4 2013 would allow NV/AMD about a year to get their 20nm designs in order for launch in the second half of 2014..

Assuming this is all true (seriously, it is looking very vague from AMD's response and that slide looks very fake to me since the logos are behind the charts) then a Q4 2013 release for a 28nm referesh would likely imply a H1 2015 release more so than anything else IMO due the gap between the HD7000 series and HD8000 series. This similarly happened with the HD5000 to HD6000 which was 14 months and another 14 months later the HD7000 series began to release.

 

[tviceman]

- Even if NV refreshes GTX680, since HD7970GE is already in the lead, how much faster can GK114 get without expanding the bus width? 10-15% faster than HD7970GE? But if NV bumps the speed of 680 by 20-25%, the Titan's 60% advantage for $900 would shrink to just 28-33% faster over GK114. That would make the Titan's price nearly impossible to justify if GTX680's GK114 successor is only $499. This scenario would piss off Titan's early adopters. Even if GK114 is 10-15% faster than HD7970GE and launches at $499, AMD could just drop the price of HD7970GE's from $430 to $399. Remember that HD7970GE is already way under $499 price level. If GK114 is 15% faster at $499, a $399 HD7970GE would still have superior price/performance.

My guess is that one of GK110's cut down SKU's will become the new "gtx780." I think GK114 will have probably only a 5-10% increase in performance over GK104 (at the same TDP) and will be designated "gtx770" if it is priced $399 or something unusual like "gtx770 ti" if it's priced ~$450. If GK114 has higher yields than GK104 to the point where they only need to bin for 1 less part than GK104, then Nvidia can get away with dropping the price $50 over what GK104 is selling at now without a hit to margins.

 

[tviceman]

What if the reason AMD is delaying the refresh is because the memory management driver offers a big enough performance increase that they don't feel pressured at all by nvidia to release cards?

It would extremely cool if that's the case, but we'll just have to wait and see. I would imagine it will improve performance in some way. How much is the question.

If AMD is delaying the release for non-technical reasons (other than overstock of existing product) it's a bad move. They did that during the Athlon days and once Intel passed them up (in part because of that) they were never able to recover.

 

[RussianSensation]

If AMD is delaying the release for non-technical reasons (other than overstock of existing product) it's a bad move. They did that during the Athlon days and once Intel passed them up (in part because of that) they were never able to recover.

Look at the last 2 quarters of AMD's GPU division's earnings. Even if AMD didn't sell a single GPU in all of 2013, it wouldn't kill the company. And even if HD8000 was 10x faster than GTX700, it wouldn't save the company by any measures. The key markets where AMD is losing $ (CPUs, APUs, servers, etc.) are actually what can kill the company because the losses there are major and they seem never-ending. What AMD needs to execute well on are Richland, Kabini, Piledriver, etc., not HD8000 GPU that might make them $25 million in profits every 3 months. Even if AMD could launch a GPU as fast as the Titan for $1K, it wouldn't even make a dent in their financials.

 

[ocre]

You can pretty much disregard Titan in this.

why? wth? that doesnt even make sense! Titan is a new GPU coming out in from nvidia in 2013. but you think everyone should pretend it doesnt exist?

Also i think a lot of people arent considering that titan is launching without a huge stock. This naturally will keep the price high of course but over time there will be larger quantities for the geforce brand. Then you have to know there will be the salvaged chips. Castrated gk110s should launch as geforce cards.

And then there is the gtx700 series. Even if AMD decides to put off a refresh in 2013 I think nvidia is still moving along in their plans for the 700 series refresh. I mean, they would be stupid not to. I expect by the end of the year we will see official gtx 700 series cards too.

For all we know, this announcement from AMD may just be a strategy. A lot of people have assumed the bundles were to reduce stock piles to ready a new series launching. This can actually prevent people from buying because they are waiting for the 8000 series. An announcement like this is strange and perhaps their is a motive behind it. It could help push unsure people to but their bundles. Another may be to throw the competition off. Make nvidia think they have nothing to worry about and bam, AMD launches when nvidia's back is turned.

That may be far fetched, but i do wonder about such a statement and why AMD thought it was important to announce something in the beginning of 2013. We still dont know much at all. It could be AMD will launch a 7900 refresh in early Jan 2014. It doesnt have to mean there is no refresh at all, just maybe no hurry for one.

 

[BallaTheFeared]

Look at the last 2 quarters of AMD's GPU division's earnings. Even if AMD didn't sell a single GPU in all of 2013, it wouldn't kill the company. And even if HD8000 was 10x faster than GTX700, it wouldn't save the company by any measures. The key markets where AMD is losing $ (CPUs, APUs, servers, etc.) are actually what can kill the company because the losses there are major and they seem never-ending. What AMD needs to execute well on are Richland, Kabini, Piledriver, etc., not HD8000 GPU that might make them $25 million in profits every 3 months. Even if AMD could launch a GPU as fast as the Titan for $1K, it wouldn't even make a dent in their financials.

I'm reading this backwards, you're saying there is no money in GPU's which doesn't seem correct. The problem is that AMD's Q3 revenue from GPU's was less than half of Nvidia's, which isn't nearly enough cash flow to fix the other side of AMD, which is bleeding money like a head wound.

GPU revenue is not only enough cash to sustain a company, but it's enough to make everyone in it rich. AMD paid a lot of money for ATI, their just bleeding it from everywhere, and their GPU sales have been lackluster to say the least... $342 mill in Q3, vs Nvidia's $740 mill.

 

[piesquared]

AMD promises to squash confusion and quickly clarify 2013 Radeon plans

More concrete information should be coming down the pipeline in the near future, however. The AMD representative I spoke to indicated that the 4Gamer article lacks proper context, and said the company will be clarifying its 2013 strategy for the Radeon brand in a conference call later this week.

Will be interesting to see what AMD is up to..

 

[BallaTheFeared]

Sweet jesus thank you gpu gods!!!!!!!!!!!!!!!!!!!!

 

[RussianSensation]

Sweet jesus thank you gpu gods!!!!!!!!!!!!!!!!!!!!

AMD will announce that PS4 will have swappable AMD GPUs for the rest of its useful life. Sony will buy AMD and PS4 will become the first upgradable console running Linux. They will also announce a partnership agreement with Valve where if you buy a new AMD GPU and a PS4, you'll get lifetime 10% off all Steam games as long as you remain a PSN+ member. And Sea Islands was really a codename for PS4 on AMD's roadmaps.