ATI Radeon - Next Hardware Generation?

Kniteman77

Weaksauce
Joined
Nov 30, 2003
Messages
123
Does anyone know when the next generation of ATI GPU's is slated for?

Looking at sinking a significant amount of money into a card right now and I'm concerned that all the 6990's on Newegg seem to be disappearing. Any ideas on why that might be?
 
People are speculating within the next month or two for the 7xxx series. All speculation, no proof...that I'm aware of.
 
Do you happen to know if it's going to be the same architecture as the 6xxx series, or a whole new chip?

Cause I guess that's the important question.
 
People are speculating within the next month or two for the 7xxx series. All speculation, no proof...that I'm aware of.

They're speculating production within the next few months, not release. With TSMC's 28nm process still being in it's infancy, it would take a bit of time to produce enough chips for a hard launch. Best bet is Q3 for either AMD or Nvidia's 28nm designs.
 
Last edited:
Do you happen to know if it's going to be the same architecture as the 6xxx series, or a whole new chip?

Cause I guess that's the important question.

Should be built like Cayman's VLIW-4 architecture. Mainly being what Cayman was meant to be when designed instead of what was released. There will probably be some more tweaks here and there, hopefully allowing for more primitives per cycle(4 would be nice) and increase tessellation scaling.
 
I remember AMD first talking about it a year ago when they announced northern islands and southern islands. If I recall:

Northern Islands: 6xxx series, 40nm, small design improvements, out in the fall of 2010

Southern islands: 7xxx series, 28nm, large design improvements/overhaul, out in summer 2011 or fall 2011 depending on how fast 28nm fabrication gets nailed down

AMD still hasn't announced a date yet so any specific times other than "sometimes this year" must be taken with a grain of salt.

They were originally going to make southern islands the 6xxx series but due to delays related to 28nm fabrication they ended up focusing on minor improvements to their existing 40nm architectures to get something out the door in the meantime.
 
Should be built like Cayman's VLIW-4 architecture. Mainly being what Cayman was meant to be when designed instead of what was released. There will probably be some more tweaks here and there, hopefully allowing for more primitives per cycle(4 would be nice) and increase tessellation scaling.

I don't know for sure yet, but I would think it would be logical to assume that it will be a continuation of the new VLIW-4 architecture found in the 6950 and 6970 also, I think that's a logical assumption they would move forward on that architecture.
 
in my experience die shrinks are always worth waiting for especially if they are a few months away.

you will always see a bump in performance, power draw, temperatures.
 
Repeating something I posted about 3 weeks ago:

No real next gen until 2012, Southern Islands taped out back in February and is mainly a die shrink of Cayman to 28nm (same VLIW4 architecture but with tweaks. I'm expecting more cores while keeping power consumption and TDP similar to Norther Islands. I wouldn't be surprised if they managed to go lower though.)

If I wanted to know anything about the true next gen AMD GPUs (AMD Graphics Core Next) I would make sure I attended the Fusion Developers Conference this June. https://amdfusion.wingateweb.com/scheduler/eventguide/publicScheduleByType.jsp?ts=13026810674470

103 - Evolution of AMD's Graphics Core, and Preview of Graphics Core Next
GPU shader cores have been evolving frequently and significantly at AMD. We introduced our common shader core in 2007 with the HD 2000 series. This introduced the unified VLIW-5 instruction set that we've had since. In late 2010, we introduced the first significant departure from this core architecture, the symmetrical VLIW-4 used in the HD6900 series of products. In this presentation, we will review that evolution, but also present an overview of the next generation of AMD cores under development. This next generation of cores will propel forward its capabilities and continue this evolution.

Speaker
Eric Demers CTO and CVP, Graphics BU AMD

2620 - AMD Graphics Core Next
At the heart of every AMD GPU is a power aware high performance set of compute units that have been advancing to bring users new levels of programmability, precision and performance. Starting with the introduction of the HD 2000 family of Unified Shader Systems to the PC consumer markets in 2007, AMD has delivered four unique generations. In this presentation, an overview of AMD's Graphics Core Next architecture will be introduced. We’ll talk about the architecture of this new shader system, and how it provides improvements to the necessary software stack and creates user programming flexibility. We believe you will obtain an understanding of this new architecture and how it lays the foundation for future AMD architectures.

Speaker
Mike Houston Fellow AMD
Mike Mantor Senior Fellow AMD
 
So essentially the 'next gen' of cards is just going to be a die shrink on the 6xxx series cards?

And the REAL next major architecture switchover won't be until 2012.

Correct me if I'm wrong on that.

Repeating something I posted about 3 weeks ago:

No real next gen until 2012, Southern Islands taped out back in February and is mainly a die shrink of Cayman to 28nm (same VLIW4 architecture but with tweaks. I'm expecting more cores while keeping power consumption and TDP similar to Norther Islands. I wouldn't be surprised if they managed to go lower though.)

If I wanted to know anything about the true next gen AMD GPUs (AMD Graphics Core Next) I would make sure I attended the Fusion Developers Conference this June. https://amdfusion.wingateweb.com/scheduler/eventguide/publicScheduleByType.jsp?ts=13026810674470
 
So essentially the 'next gen' of cards is just going to be a die shrink on the 6xxx series cards?

And the REAL next major architecture switchover won't be until 2012.

Correct me if I'm wrong on that.

Pretty sure that's the plan. After all Cayman was supposed to have 1920 SP's and that didn't happen because of the 32nm skip.

Cayman crunched down to 28nm with additional SP's and potential VLIW4 arch from top to bottom sounds good to me.

Can you say 6990+ single card performance for $300-$400. 28nm is exciting, we've been stuck on 40nm for far too long. Huge increases in GPU performance are just around the corner.
 
So whats going to happen when we reach 0nm... are we going to to PM??? :D (picometer)
 
So whats going to happen when we reach 0nm... are we going to to PM??? :D (picometer)

Well a silicon atom is 220 picometers in diameter.

28 nanometers / 0.22 nano meters = 127.27

So in the 28 nanometer fab process we're talking 127 atoms width for each gate.

I'm not a physicist, but I did study chemical engineering and I'm pretty sure that in the 2 to 1 nanometer range, (in this case 10 to 5 atoms in width) you're going to start meddling with quantum mechanics and the electrons might not always go where you want them to go.

So my guess is physics will stop us at Nanometers :)
 
There is a physical limit, that much I know, forget what it is though.

I've heard that 6nm is the about the smallest we can expect to get without quantum interference, which means in the next decade we might see an indefinite period of relative stagnation in terms of performance increases.
 
I've heard that 6nm is the about the smallest we can expect to get without quantum interference, which means in the next decade we might see an indefinite period of relative stagnation in terms of performance increases.

Unless someone comes up with a better way to do things :)
 
Thanks to all who posted on this thread, much needed info. I was deciding on whether to upgrade now or wait a few months. If it may be as much as 6 months I think I'll upgrade now as my GPU is failing fast and I do not want to have to count on a bake to keep me functional.
 
when 28nm becomes available, amd and nvidia should refresh today's gpu's.

think about it, Fermi with more shaders at 1ghz would be sick, and they would be cheap too.
 
when 28nm becomes available, amd and nvidia should refresh today's gpu's.

think about it, Fermi with more shaders at 1ghz would be sick, and they would be cheap too.

ATI is currently king of the shader hill. Not sure Nvidia can take that away from them.

Also, correct me if I'm wrong but more shaders = larger die size = less likely to be able to preform at higher clock speeds.
 
when 28nm becomes available, amd and nvidia should refresh today's gpu's.

think about it, Fermi with more shaders at 1ghz would be sick, and they would be cheap too.

Fermi could really stretch it's legs on 28nm and make-up it's few short comings in the process. Should be interesting to see who takes the crown with the 28nm shrink. AMD seems to be the people's choice these days.
 
ATI is currently king of the shader hill. Not sure Nvidia can take that away from them.

Also, correct me if I'm wrong but more shaders = larger die size = less likely to be able to preform at higher clock speeds.

Well, that also depends of which vendor are you talking about. AMD's shaders are so small (Much smaller than nVidia's shaders) that they can fit lots of them, clock them high and have power consumption in check while having great performance. nVidia shaders are much bigger and will use more power, so they have to be more careful in the TDP department.

Here's a post of mine on another thread here. Here's what I think about next generation of AMD's GPU's and its issues.

"That prediction is a fad. AMD's VLIW underutilization always has been an issue since its induction on the HD 2x00 series. A good example of this. The Radeon HD 6870 has 1120 stream processors which in reality, are 224 stream processors in which every processor in there, is capable to execute up to 5 instructions per clock, but as far as it is from the same thread, and that's the catch.

You compare it to the scalar architecture of the for example, the GTX 460 which has 334 stream processors, and each processor there is capable to execute one instruction per clock regardless if its from the same thread or not, which shows that nVidia's approach in terms of performance is quite predictable and it's based on Thread Level parallelism, AMD's approach of the VLIW5 is torward Instruction Level Parallelism which means that it requires of compiler tricks to maximize its execution resources. So, sounds quite outstanding that the GTX 460 is almost able to keep up with the HD 6870 and its sheer amount of 1120 stream processors, but in the other hand you could also say that its a feat seeing the HD 6870 with its 224 stream processors being able to outperform slightly the GTX 460 with its 334. AMD's VLIW approach under maximum utilization is able to smoke anything that nVidia currently offers, but that's something that only would happen in rare ideal circumstances, I will explain that below.

nVidia's approach usually is better as it requires of little software optimization, but also means that the chip die will be much larger as their shader processors are much fatter and consumes far more power. AMD's approach is to accomodate much smaller stream processors to increase parallelism that it is easy in terms of hardware implementation, but will require good software to work, but graphic rendering is so parallel that it explains why AMD had been trading blows for a while with its much smaller chip, specially since 2008.

But even with that, AMD knew that their VLIW5 performance wouldn't scale linearly forever with the increase of stream processors. So they did two little experiments. One of them is the HD 6870 has lots of tweaks at the hardware level that allows their Dual Command Queue processor to take more control of the shader resources compared to the HD 5000 series, which also explains why the HD 6870 performs so close compared to the HD 5870 while having 34% less stream processors and with a smaller die, it is a feat.

Their second test is their Cayman GPU, which moved their design from VLIW5 to VLIW4, which is a move that clearly shows that they're moving torward a more oriented TLP design than with previous generations which was more oriented to ILP. It was because according to AMD, they only saw an average of 3/4 or 60%-75% of utilization on their VLIW5 design in the best case scenario, something that only happened on 60% of the time, showing that a lot of hardware was wasted idle on the die. So instead of adding 4 little stream processors and one fat processor for special tasks which certainly used a lot of space and idled a lot of time, made more sense removing that fat processor and increase the computing performance of the remaining 4 making them equal on everything.

So in some circumstances where AMD can achieve maximum execution resources with a VLIW5 design (Something very rare), it should be faster than the VLIW4 as it happened on Civilization 5 compute tests. But if you couple the VLIW4 design with Barts tweaks, means that you can achieve a much better utilization of the VLIW4 resources that AMD was able to achieve on the HD 5000 series. That explains why the HD 6970 while it only has 1536 stream processors (As it has 24 SIMD engines), it is faster than the HD 5870 that has 1600 stream processors (20 SIMD engines). The issue here is that Cayman was supposed to be a 28nm product, so other performance enhancing features weren't added as it would make the chip bigger than it already is. So think of this, if Cayman used the VLIW5 design, it would have 1920 stream processors, that also explains why it isn't much faster than the HD 5870.

So the secret sauce would be mixing Barts optimizations (95% of the HD 5870 performance with 34% less stream processors), increased processor count (From HD 5870's 20 SIMD to 24 SIMD's from the HD 6970), Tessellation enhancements, bigger VRAM size and some core clock bumps and can give you between 20%-60% higher performance depending of how much advantage the software can take from Cayman's approach. So I do think that in the future, both GPU vendors with their different approaches will remain close for a while, nVidia's path is safer but also more expensive, but they have deeper pockets than AMD. I can't wait to see what nVidia/AMD will offer on their 28nm process!!"

Source: http://hardforum.com/showpost.php?p=1037111496&postcount=5
 
@evolucion8: That was incredibly insightful, thanks!

^This + 1

I was actually wondering the same thing. As the 6000 series isn't a giant bump from the 5000 series, but also uses less to make more. I'm very excited. I need to start saving.
 
I'm pretty sure the next gen will get 3 or 4Gb of VRAM, to cancel out any ''advantages'' the still elusive and hard to find 580 3Gb will get for a couple of months.

Nvidia made some pretty bad decisons regarding VRAM, and they are now trying to fix that problem. 580 with 1.5Gb are having a hard time with high resolutions with high AA, and multiple screen set-ups. That's why they are trying to get more 580 3Gb out really fast.

So AMD will also react, and use more VRAM.
 
So essentially the 'next gen' of cards is just going to be a die shrink on the 6xxx series cards?

And the REAL next major architecture switchover won't be until 2012.

Correct me if I'm wrong on that.

I'm not sure I buy that speculaton. The 6xxx series was rumored to be the same core with different 'improved' shaders. But I think there were fundamental improvements made that had little to do with the shaders. In fact one could argue that on a per shader basis, that the new VLI4-W shaders in the 69xx series are not as powerful overall as the older style VLI5-W shaders used in the 6850/6870.

I think AMD needs to continue to tweak the efficiency of VLI-4W shaders or even go to a different improved design because they don't peform as well as the old style shaders on benchmarks such as Heaven Uni-Engine and 3DMark. Tessallation peformance also needs improvement. It's almost on par with Nvidia now, but still about a tier lower. I know Heaven and 3DMark aren't games, but I'm sure they don't like coming in 2nd to Nvidia on those universal benchmarks.
 
3D Mark 11 is much better now with 11.4, but Heaven is still a ''problem''.

But since both Nvidia and AMD can ''cheat'' in Heaven, I wouldn't put too much weight on that synthetic benchmarks. Both side can disable tessalation and still get the ''Extreme Tessalation'' showing. I don't even use it anymore.

And for the GPU score in 3D mark 11, I can do 23K, just like almost every ''normal'' 580 Quad-SLI set-up with same CPU and speed. And I'm using a 6990+6970+6970.
 
I'm not sure I buy that speculaton. The 6xxx series was rumored to be the same core with different 'improved' shaders. But I think there were fundamental improvements made that had little to do with the shaders. In fact one could argue that on a per shader basis, that the new VLI4-W shaders in the 69xx series are not as powerful overall as the older style VLI5-W shaders used in the 6850/6870.

I think AMD needs to continue to tweak the efficiency of VLI-4W shaders or even go to a different improved design because they don't peform as well as the old style shaders on benchmarks such as Heaven Uni-Engine and 3DMark. Tessallation peformance also needs improvement. It's almost on par with Nvidia now, but still about a tier lower. I know Heaven and 3DMark aren't games, but I'm sure they don't like coming in 2nd to Nvidia on those universal benchmarks.

I'm just interested in raw calculation throughput :) Video performance is me 5 years ago.
 
I'm not sure I buy that speculaton. The 6xxx series was rumored to be the same core with different 'improved' shaders. But I think there were fundamental improvements made that had little to do with the shaders. In fact one could argue that on a per shader basis, that the new VLI4-W shaders in the 69xx series are not as powerful overall as the older style VLI5-W shaders used in the 6850/6870.

I think AMD needs to continue to tweak the efficiency of VLI-4W shaders or even go to a different improved design because they don't peform as well as the old style shaders on benchmarks such as Heaven Uni-Engine and 3DMark. Tessallation peformance also needs improvement. It's almost on par with Nvidia now, but still about a tier lower. I know Heaven and 3DMark aren't games, but I'm sure they don't like coming in 2nd to Nvidia on those universal benchmarks.

AFAIK, the only benchmark that shows a VLIW5 outperforming the VLIW4 design is the Civilization 5 Compute benchmark. Remember than AMD switched from Assymetric VLIW5 to Symmetric VLIW4 so they can keep increasing the shader count and increase the performance without going overboard with the chip's size. Plus like I posted previously, the VLIW5 has some serious underutilization issues, with the Vec4, the execution engine usage is much higher, but Cayman wasn't able to shine as it was a product intented to be made on a 32nm process, when it was scrapped. AMD had to create Cayman and wasn't able to put all the research and optimizations on the chip, plus they had to keep the die size on check. The HD 6970 performs like an optimized VLIW5 1920 stream processor design, that explains also why it isn't much faster than the HD 5870, but it is far more efficient now and will scale better with shader count increase, something that was close to the limits with the VLIW5 design.
 
Does anyone know when the next generation of ATI GPU's is slated for?

Looking at sinking a significant amount of money into a card right now and I'm concerned that all the 6990's on Newegg seem to be disappearing. Any ideas on why that might be?

RUMOR

http://www.digitimes.com/news/a20110425PD218.html

In addition to Llano, AMD's 28nm products including Krishna- and Wichita-based APUs and Southern Island GPU are already under tape-out.

RUMOR

I guesstimate that their super high end will come around the christmas holiday season. Waiting for Southern Island and Kepler is killing me inside. Battlefield 3 you better be worth it!
 
Well a silicon atom is 220 picometers in diameter.

28 nanometers / 0.22 nano meters = 127.27

So in the 28 nanometer fab process we're talking 127 atoms width for each gate.

I'm not a physicist, but I did study chemical engineering and I'm pretty sure that in the 2 to 1 nanometer range, (in this case 10 to 5 atoms in width) you're going to start meddling with quantum mechanics and the electrons might not always go where you want them to go.

So my guess is physics will stop us at Nanometers :)

This is exactly right. We will eventually hit a wall that die shrinks will no longer be feasible. Actually just finished a course in VLSI that explains what happens on the atomic level for CPUs/GPUs...kinda sad but we will hit a wall in about 10 years if not less. Then all we can do is start making the footprint larger :p
 
What about pci-e 3.0? Has there been any news of it yet?


PCI-E 3.0 has been finalized for a while, but it'll still be some time before we see motherboards(and chipsets) that support it. It's just an another extension of PCI-E 2, so not having it won't have a performance penalty if the next gen have support(which I don't really know if they do or not).
 
This is exactly right. We will eventually hit a wall that die shrinks will no longer be feasible. Actually just finished a course in VLSI that explains what happens on the atomic level for CPUs/GPUs...kinda sad but we will hit a wall in about 10 years if not less. Then all we can do is start making the footprint larger :p

I disagree.. there's a lot to be said for architectural efficiency improvements. After all Intel and AMD hit a cpu clock wall years ago.. but they still manage to release and sell new faster cpus (P4 vs. Core 2.. they actually *decreased* clocks for the same performance.. getting some of their clock headroom back).

Its just a bit harder and you'll see smaller gains.. no gains come "for free" from process.

Imagine Intels Tick-Tock strategy without the Tock... speed increases will definitely slow.
 
I disagree.. there's a lot to be said for architectural efficiency improvements. After all Intel and AMD hit a cpu clock wall years ago.. but they still manage to release and sell new faster cpus (P4 vs. Core 2.. they actually *decreased* clocks for the same performance.. getting some of their clock headroom back).

Its just a bit harder and you'll see smaller gains.. no gains come "for free" from process.

Imagine Intels Tick-Tock strategy without the Tock... speed increases will definitely slow.

Oh yeah definitely! I meant that the wall will be in die sizes rather than clock speeds. We can only go so small before the atoms themselves start compounding on each other and then poof your transistors don't work anymore :p Although there is supposed talk of 3-D transistor layouts! Gets me all excited about the new technology :D
 
Who knows though- by the time we get to that wall in 10 years, we may have found a way to make things smaller- not using transistors anymore maybe, but some other property of atoms or something. Remember- technology is not limited to what we know today. What we know TODAY is limited to a certain minimal size, but who would have thought we could have the performance that we do now 10-15-20 years ago? Maybe they'll just find a way to make clock speeds way higher.
 
Back
Top