AMD confirms new architecture + launch timeframe for HD6xxx

[happy medium]

Would it really be that surprising for Barts XT to be somewhere between the 5850 and the 5870?

No, I've been saying this for a month now. I would think it would be about 5% faster then a 5850 and cost about 239.00 at launch. That would put it about 13% faster then a gtx460 which easily should be below $200 by then.

 

[tijag]

It is not unreasonable, but the problem is we dont know what GPU Barts XT actually is.

Is it going to be HD6770 priced at $150-200? Or is it going to be HD6870 priced at $200-250? If AMD delivers HD6770 at $150-200 with performance between 5850 and 5870, they hit the jackpot. If, however, they deliver Barts XT as HD6870 and price it at $250 with performance between 5850 and 5870, then it's nothing special.

The key piece of information is price, not just performance. Recall that HD4870 was slower than the GTX280, but priced at $299, it was an unbeatable card.😛

I'm not really too concerned with what they call it.

If it's between 220 - 240 and has better performance than a 5850, or even equal to a 470... that will be a compelling product.

If the Barts XT part is $200 and the Barts Pro part is $160, that would be an amazing value. I see no reason why they should price parts that low that have this kind of rumored performance. I think its more likely that they undercut the current pricing of the 5850 by just a little bit, while beating its performance by just a little bit.

For example, if Barts XT sells for $235 and is between a 5850 and a 5870 in performance, meanwhile with serious improvements to tessellation, that seems like a mainstream winner.

That could leave Barts Pro to $200 or so and be slightly better performance than GTX460.

 

[HumblePie]

I'll be selling my 4870x2 finally and one 5870. Drop the 5870 into my HTPC and pick up two 6970's for my main. Better start saving a few extra bucks to cover the difference from what I get from selling my current cards.

 

[RussianSensation]

For example, if Barts XT sells for $235 and is between a 5850 and a 5870 in performance, meanwhile with serious improvements to tessellation, that seems like a mainstream winner.

That could leave Barts Pro to $200 or so and be slightly better performance than GTX460.

Oh that's true of course; and hence lies the problem. All we are getting are slight performance increases in the last 12 months. However, in the last 7-8 years, we generally got 70-100% performance improvement at the same price every 12-15 months, or a 2x price reduction for the same performance.

Several times this trend didn't hold true, however. This happened when 9800XT replaced 9700Pro, 7800GTX replaced 6800Ultra and again resurfaced starting with HD5800 and GTX4xx series, which both flopped in terms of delivering 2x the performance at the same price or delivering the same performance at 2x lower price. For instance, HD5770 was actually slower than the HD4870 and priced higher when it launched. And frankly, GTX470/480 are nowhere near 2x the performance increase over GTX280/5.

No matter what card ever comes out in the mid-range, it will always be judged against 6600GT in my eyes, a card that beat out a previous high end 9800XT and cost less. So far, neither AMD nor NV have been able to deliver a killer mid-range card just like the 6600GT was back then. Even now 12 months later, GTX460 is hardly a compelling reason to upgrade to for GTX275/4890 owners. Hopefully HD6000 series raises the mid-range performance 2x from HD5770 - it's about time.

 

[HumblePie]

Oh that's true of course; and hence lies the problem. All we are getting are slight performance increases in the last 12 months. However, in the last 7-8 years, we generally got 70-100% performance improvement at the same price every 12-15 months, or a 2x price reduction for the same performance.

Several times this trend didn't hold true, however. This happened when 9800XT replaced 9700Pro, 7800GTX replaced 6800Ultra and again resurfaced starting with HD5800 and GTX4xx series, which both flopped in terms of delivering 2x the performance at the same price or delivering the same performance at 2x lower price. For instance, HD5770 was actually slower than the HD4870 and priced higher when it launched. And frankly, GTX470/480 are nowhere near 2x the performance increase over GTX280/5.

No matter what card ever comes out in the mid-range, it will always be judged against 6600GT in my eyes, a card that beat out a previous high end 9800XT and cost less. So far, neither AMD nor NV have been able to deliver a killer mid-range card just like the 6600GT was back then. Even now 12 months later, GTX460 is hardly a compelling reason to upgrade to for GTX275/4890 owners. Hopefully HD6000 series raises the mid-range performance 2x from HD5770 - it's about time.

The problem is GPU's I think are finally hit the asymptote of performance much like CPU's have been. Moore's Theorem I always knew was a bunch of baloney. All performance items are based off a 1/x curve. Eventually every innovation tapers off.

 

[Illyan]

This is why I hope Barts comes out as 67XX rather than 68XX.

 

[nitromullet]

Even now 12 months later, GTX460 is hardly a compelling reason to upgrade to for GTX275/4890 owners.

That's also not really the target market for the GTX 460, so it really shouldn't be a compelling reason for GTX275/4890 owners. GTX 470/480 and 5850/5870 are the cards targeted towards these people. If you "upgrade" from last gen's high end to this gen's mid range you shouldn't expect much in the way of a performance increase.

I don't see many 5870/5850 or GTX 480/470 owners picking up a Barts card either. They will wait for Cayman.

edit: I've always felt that a good mid-range card should be along the lines of last gen's high end in terms of performance, but at a mid-range price.

 

[Daedalus685]

The problem is GPU's I think are finally hit the asymptote of performance much like CPU's have been. Moore's Theorem I always knew was a bunch of baloney. All performance items are based off a 1/x curve. Eventually every innovation tapers off.

Moore's 'Law' has nothing to do with performance, only transistor density. Which has been increases at the expected rate. We have a decade or so before we stop meeting Moore's law. By then hopefully we perfect other means beyond 2D ICs.

Beyond that, doubling performance would be unheard of without a shrink or die size increase.

 

[tijag]

Oh that's true of course; and hence lies the problem. All we are getting are slight performance increases in the last 12 months. However, in the last 7-8 years, we generally got 70-100% performance improvement at the same price every 12-15 months, or a 2x price reduction for the same performance.

Several times this trend didn't hold true, however. This happened when 9800XT replaced 9700Pro, 7800GTX replaced 6800Ultra and again resurfaced starting with HD5800 and GTX4xx series, which both flopped in terms of delivering 2x the performance at the same price or delivering the same performance at 2x lower price. For instance, HD5770 was actually slower than the HD4870 and priced higher when it launched. And frankly, GTX470/480 are nowhere near 2x the performance increase over GTX280/5.

No matter what card ever comes out in the mid-range, it will always be judged against 6600GT in my eyes, a card that beat out a previous high end 9800XT and cost less. So far, neither AMD nor NV have been able to deliver a killer mid-range card just like the 6600GT was back then. Even now 12 months later, GTX460 is hardly a compelling reason to upgrade to for GTX275/4890 owners. Hopefully HD6000 series raises the mid-range performance 2x from HD5770 - it's about time.

The 4890 was an unusually good deal. I think it could be had for around $200 dollars. Generally though, high end cards sell for 400 bucks or so, and in this case if you can buy a barts XT for 130+ less than the 5870, but meet it in performance, I think you should feel very happy. Thats essentially bringing last generations 'high end parts' down to 'every day prices.'

That sounds like an acceptable situation to me.

 

[nitromullet]

The problem is GPU's I think are finally hit the asymptote of performance much like CPU's have been. Moore's Theorem I always knew was a bunch of baloney. All performance items are based off a 1/x curve. Eventually every innovation tapers off.

Well, the price has dropped as well. It used to be that a high end single gpu card was ~$600 at launch, but now these cards are $350-500. People paying $800+ for video cards is not very common anymore, but I remember paying $1100 for dual 7800GTXes just a few years ago. Today you could buy three 5870s for that much. What you get for your money these days is staggering compared to what it used to be.

 

[Absolution75]

No matter what card ever comes out in the mid-range, it will always be judged against 6600GT in my eyes, a card that beat out a previous high end 9800XT and cost less. So far, neither AMD nor NV have been able to deliver a killer mid-range card just like the 6600GT was back then. Even now 12 months later, GTX460 is hardly a compelling reason to upgrade to for GTX275/4890 owners. Hopefully HD6000 series raises the mid-range performance 2x from HD5770 - it's about time.

Bolded for emphasis 😛

Yeah, its even harder when you get a HD 4890 for $128 over a year ago. . . Given, I could have made a profit on the card. . . but nothing really interests me atm - as far as a replacement.

How about a midrange card that is at least as fast as the HD5850 for $180-240ish. That will attract my attention.

 

[tijag]

Bolded for emphasis 😛

Yeah, its even harder when you get a HD 4890 for $128 over a year ago. . . Given, I could have made a profit on the card. . . but nothing really interests me atm - as far as a replacement.

How about a midrange card that is at least as fast as the HD5850 for $180-240ish. That will attract my attention.

That sounds exactly like Barts XT.

 

[happy medium]

How about a midrange card that is at least as fast as the HD5850 for $180-240ish. That will attract my attention.

Sounds like a factory overclocked gtx460, you can have that now.

 

[bryanW1995]

yeah, gtx 460 1gb has been spotted in the wild below $180 recently in fact. you should set your sights higher, maybe shoot for 5870 performance for $200 or less.

unfortunately, we're likely going to end up with a mixed bag. figure a barts xt will be more like a 5850 in older games and more like a 5870+ in newer games.

 

[at80eighty]

I can only think of 2 games where HD6800/GTX5xx series will be truly appreciated - Metro 2033 and Crysis 2. Unfortunately one of these games is delayed until March 2011.

the upside being that the drivers will have matured a solid deal by the time Crysis2 is out.

im really looking forward to the 6890 or 6990 - cant wait either 😛

 

[bryanW1995]

i don't think that anybody has talked about a 6890. current rumors for barts xt are 6770 or 6870, current rumors for cayman xt is 6870 or 6970, and antilles looks to be 6970 or 6990.

 

[bunnyfubbles]

Why is there even mention of a 6890?

So many people clung to the idea of a 5890 and that never came to be. The 4890 was an oddity of special circumstances that broke the norm, not establish it.

 

[at80eighty]

Yeah i was thinking 6970 and went ahead and typed 6890 - mea culpa - I posted that way too soon after waking up 🙂

 

[blastingcap]

That sounds exactly like Barts XT.

A 5850-caliber card at anything over $199 would be a failure in my book, as it would barely keep pace with NV's midrange price/perf ratio. GTX460-1GB's have already been on sale for as low as $180AR, and they may go even lower given another 5 weeks (the time between now and Oct. 25). I'm guessing $170AR by Oct. 25.

In short, if AMD wants to keep its market share in the midrange segment, it needs to BEAT the GTX460, not just run alongside it and hope (in vain?) that single-card Eyefinity beats PhysX and CUDA.

A 5850-caliber card at ~$180 should be enough to win back price/perf from NV in the midrange. But >$200 = fail. A 5850-caliber card at ~$160 would be a win and force NV to slash prices even more to compete.

Of course, there are rumors that Barts XT will end up as a 5870 caliber card and not 5850 caliber, in which case a somewhat higher price might be justified. Either way, I hope AMD goes for body blows and long-term marketshare rather than short-term profits. Every AMD card sold could mean one less user in the NV CUDA/PhysX ecosystem, and in the meantime, consumers win.

 

[RussianSensation]

Either way, I hope AMD goes for body blows and long-term marketshare rather than short-term profits. Every AMD card sold could mean one less user in the NV CUDA/PhysX ecosystem, and in the meantime, consumers win.

While I hope that AMD strongly improves the performance, I would not want it to blow nV out of the water. History has shown us that when 1 firm dominates (G80 series, HD5000 series) and there is little competition in the marketplace, we get both performance and price stagnation. So certainly we would need both firms to compete well, not one blow the other away like happened this round. Stronger and much closer offerings from both camps such as 6800GT/X850 Pro, X1800XT/7800GT, GTX260/4870 result in better pricing and superior options for consumers.

 

[Will Robinson]

A 5850-caliber card at anything over $199 would be a failure in my book, as it would barely keep pace with NV's midrange price/perf ratio. GTX460-1GB's have already been on sale for as low as $180AR, and they may go even lower given another 5 weeks (the time between now and Oct. 25). I'm guessing $170AR by Oct. 25.

In short, if AMD wants to keep its market share in the midrange segment, it needs to BEAT the GTX460, not just run alongside it and hope (in vain?) that single-card Eyefinity beats PhysX and CUDA.

A 5850-caliber card at ~$180 should be enough to win back price/perf from NV in the midrange. But >$200 = fail. A 5850-caliber card at ~$160 would be a win and force NV to slash prices even more to compete.

Of course, there are rumors that Barts XT will end up as a 5870 caliber card and not 5850 caliber, in which case a somewhat higher price might be justified. Either way, I hope AMD goes for body blows and long-term marketshare rather than short-term profits. Every AMD card sold could mean one less user in the NV CUDA/PhysX ecosystem, and in the meantime, consumers win.

Yup I agree with this,HD5870 level performance going up against GTX460 would be a great move by AMD.
I hope they can pull it off.😛

 

[badb0y]

Sorry about that happy, guess I got carried away there 😛 .

Anyways that's great news for us that the HD 6k is coming sooner than many expected.

 

[blastingcap]

While I hope that AMD strongly improves the performance, I would not want it to blow nV out of the water. History has shown us that when 1 firm dominates (G80 series, HD5000 series) and there is little competition in the marketplace, we get both performance and price stagnation. So certainly we would need both firms to compete well, not one blow the other away like happened this round. Stronger and much closer offerings from both camps such as 6800GT/X850 Pro, X1800XT/7800GT, GTX260/4870 result in better pricing and superior options for consumers.

I see NV as the stronger GPU company over the last several GPU generations. AMD should be very afraid of NV's next generation (Fermi 2 or whatever) and should thus push hard with its advantage at 40nm to gain parity in overall market share while it's possible. I do mean OVERALL market share, not DX11. Sure, 5xxx dominates DX11 so far, but that's still a fraction of the market, as you can see from Steam Hardware Surveys and the like. There is a lot of NV inertia which bankrolls and makes plausible PhysX and CUDA and software developers using only NV stuff (F@H, and some graphics programs).

Once there is overall market share parity, perhaps NV will back off the proprietary stuff and wholeheartedly go OpenCL, which benefits everyone except NV. More importantly, with parity comes better competition and lower prices.

So overall I think we are in agreement here.

 

[T2k]

As someone buying dozens of CUDA-exclusive cards I still find it funny how much nonsense NV's PR machine can sell while literally seeing their market share disappear in less than a couple of years... NV bankrolls what? And "plausible PhysX"...?
I wrote it here a year ago and people didn't believe me: NV is doomed if they cannot convince enough companies to switch to CUDA - they lost their entire chipset business, their usual fake speccing did not work in mobile business so Tegra is a disaster so far and as stronger and stronger x86-based integrated CPU-GPU chips will arrive NV's income will fall even lower... in less than a year they lost market leader position to ATI in both discrete and integrated markets.
They need CUDA to survive but I'm highly skeptical about its potential market, regardless all the BS they try to sell to their investors.

 

[HurleyBird]

I see NV as the stronger GPU company over the last several GPU generations. AMD should be very afraid of NV's next generation (Fermi 2 or whatever) and should thus push hard with its advantage at 40nm to gain parity in overall market share while it's possible.

They outshipped Nvidia last quarter, although Nvidia still has it's quadro stronghold.