Predict: The Next Generation Console Tech
- Thread starter Acert93
- Start date
- Status
Dr Evil said:
Wasn't that mostly a result of the ps3 having a outdated gpu?
If you look at the processors made on a given node, there are lots of cases where a more mature node sees processors made that offer bigger leaps in performance over the first chips on that node. For example:
- Athlon 64 X2 90nm to Athlon 64 X2 65nm offered basically no improvement in raw performance, but a small improvement in perf/watt.
- Athlon 64 X2 (65nm) to Phenom (65nm) was almost double the raw performance, despite no change in node.
- Phenom (65 nm) to Phenom 2 (45nm) was about 25% more performance.
- Phenom 2's slowly got faster until later in the node when Phenom 2 X6 (45nm) was released and was about 50% faster than the fastest chips earlier in the node.
In all these cases the biggest jump in performance came with larger and more advances processors later on in the same node. The idea that you have have to be in there day 1 on a new node to get a big jump simply isn't always true. MS managed that with the 360 at great cost - they couldn't test properly, it generated more heat than they predicted and they were probably supply constrained (if not by the gpu then they would have been if they hadn't been more severely constrained by memory or the CPU) .
Recent GPU history shows several similar occurrences with node transitions. Build a bigger, faster, smarter chip on a mature node and you can sometimes get a bigger jump than just going in a new node day 1.
- Athlon 64 X2 90nm to Athlon 64 X2 65nm offered basically no improvement in raw performance, but a small improvement in perf/watt.
- Athlon 64 X2 (65nm) to Phenom (65nm) was almost double the raw performance, despite no change in node.
- Phenom (65 nm) to Phenom 2 (45nm) was about 25% more performance.
- Phenom 2's slowly got faster until later in the node when Phenom 2 X6 (45nm) was released and was about 50% faster than the fastest chips earlier in the node.
In all these cases the biggest jump in performance came with larger and more advances processors later on in the same node. The idea that you have have to be in there day 1 on a new node to get a big jump simply isn't always true. MS managed that with the 360 at great cost - they couldn't test properly, it generated more heat than they predicted and they were probably supply constrained (if not by the gpu then they would have been if they hadn't been more severely constrained by memory or the CPU) .
Recent GPU history shows several similar occurrences with node transitions. Build a bigger, faster, smarter chip on a mature node and you can sometimes get a bigger jump than just going in a new node day 1.
That goes only for NVIDIA until GT200. Fermi/GF1xx was manufactured under TSMC 40G and upcoming Kepler/GK1xx will be manufactured at TSMC 28HP. On the other side of the river bank ATI/AMD is traditionally manufacturing its GPU chips for years now on the smallest available half or full node process.
If you consider that GF100 was a tad short of 530mm2@40nm it isn't too far away from the upper threshold for 40nm. It wasn't obviously possible to "squeeze" those 3 billion transistors into 55nm, let alone 65nm. The G80 times are long gone in the GPU world, especially as chips become consistantly bigger due to scaling demands both in capabilities and in performance.
DuckThor Evil
Legend
ATI wouldn't have had anything fancy over Xenos either. The products that were better were also bigger and not suitable for a console.
DuckThor Evil
Legend
I'm sorry but those examples are if not terrible then not very good either. Yeah If you just transfer your existing design to a smaller process, you'r not going to see huge gains. Also going from dual core to quad core is not hard to see "double the raw performance" likewise 50% with 6 core > quad core...
AMD did see some nice die size shrinks though.
Also MS wasn't supply constrained to any meaningful degree for longer than a month or two, despite their hardships that imo had nothing to do with the chips, but errors in the cooling and manufacturing. Launch PS3 used even more power.
Recent GPU history doesn't really back that up at all. yeah there has been some room for tweaking but for example GTX 580 uses basically the same chip as the one in GTX 480, which is a huge chip and thus running into problems with it was more likely than with a smaller chip. On the AMD side, Cayman 6970 is only a little bit faster than the 5870, which was released two years ago when 40nm was the bleeding edge. Caymans gains also basically came from being bigger. AMD hasn't really been able to get any sort extra performance from tweaks during the 40nm era and the next leap in performance comes due to the 28nm process shrink. Imo the situation has been the same pretty much the whole time. yes there are few examples when the initial design has been so tiny that there has been plenty of room to increase the die size(3870>4870 or G80), but those times are long gone. The gains during the last two years on 40nm have been miniscule.
+ there are countless better examples showing the opposite of what you'r implying without having to resort to doubling the die size etc.
Last edited by a moderator:
Indeed...
As for fitting the 28nm CPU in the ~40watt budget, I'm not seeing how this would be an issue.
There aren't any cpus built on the 28nm process to compare, but there are a few built on 32nm.
Intel has a few, but I'm not going down that road again. Matter of fact forget I even mentioned it!
AMD has only the Fusion line to compare (I refuse to look at bulldozer any more than I have to!)
This leaves only the Llano to compare to.
Llano is a 1.45 billion transistor chip. CPU and GPU combined so it is very difficult to draw a direct comparison, but AMD does have one of these 1.45billion transistor (3 times my target size xb720 cpu) that operates at under 45watts and on 32nm to boot.
Granted, it's binned for mobile, but it is 3 times(!) larger and on 32nm instead of 28nm.
Other comparison:
http://www.anandtech.com/show/3774/welcome-to-valhalla-inside-the-new-250gb-xbox-360-slim/3
XB360s on 45nm.
Total system power draw 90watts at full load.
90watts x 80% = 72watts total system budget. Minus out the disc drive (13watts) and "other" power draws (9 watts?) puts the xcgpu at 50watts
Taking the traditional approach: 50watts @ 45nm = 25watts @ 32nm
Granted, it isn't likely to be perfectly linear, but it is also a smaller node (28nm vs 32nm) which should get us pretty darn close.
Conclusion:
Current xcgpu (~500 million transistors) power draw is 50watts@45nm.
Projected power draw 32nm = 25watts
I think it's safe to assume they can (comfortably) hit under ~40 watts for their next-gen 500million transistor cpu at 28nm.
Last edited by a moderator:
...see above
Indeed!
I've been pitching this idea for years.
MS has been taking baby steps in this direction lately, but I'm not seeing enough effort/investment on their part to really take this opportunity and run with it.
Maybe this will change now with Kinect and xb720 around the corner.
Exactly.
I don't think anyone would argue against the notion that a process node usually improves over time which allows for higher clocking, or lower power at the same clock.
However, it would be the first time in history (that I'm aware of) if roughly the same design (trans count) was lower power on the old node vs the newer one.
It's all about what is available in the short term.
2012-2013 is the time-frame.
28nm is as good as it gets for MS/Sony in that time frame. Teething problems (low yield) are expected at the introduction of a new process node, thus allowing for "later gains on the same process node" which is usually just taking advantage of the node's full capability.
Chips shipping now on 28nm will probably not be able to hit target spec for either speed, or TDP. But the process will get better.
AMD and Nvidia are both expected to launch new GPU's early next year (Q1) using 28nm.
As the process matures, better yields should surface (~mid 2012), just in time for mass production of ps4 and xb720 to launch late 2012.
If there is a problem with the 28nm process, obviously the launch window(s) will be pushed back, but things seem to be going rather smoothly for TSMC as they claim to be shipping all variants of 28nm for over a month now.
exactly. Especially about the part where real-world performance was compared
The general conversation was about CPUs. That was AMDs almost complete recent CPU history. AMDs fab is now Global Foundries, a foundry that TheChefO has been pimping heavily. Those example are the best and most appropriate ones available IMO. If you have any better or more appropriate examples then hey, I'd love to see them.
To round out AMDs entire recent lineup of CPU (just to make absolutely sure that under no circumstances can I be accused of cherry picking):
- Athlon 64 X4 (45nm) to Llano (32nm) showed a small improvement in perf/watt, no real increase in performance, and massive, massive yield issues that are unresolved over half a year later (after an initial delay of over half a year).
- Phenom 2 X6 (45nm) to Bulldozer (32 nm). OH. DEAR.
The jumps in performance across node transitions have been smaller than the increases that occurred during the lifetime of the node.
Oh yeah, WiiU CPU: 45nm. There's got to be a reason for this!
And yet this is what is commonly done. You want to ignore practice in favour of your theory - a theory which wants us ignore the reality that clocks and chip sizes typically increase as the node progresses.
I can't understand for the life of me why you'd claim that isn't relevant to this discussion. It's entirely relevant. It goes to the very heart of discussion.
And for AMD this has always happened on a mature node. Until Bulldozer.
Now lets look at Intel (from memory, tell me if I'm wrong):
- Single core to dual core: done on the same node (65nm)
- Dual core to quad core: done on the same node (45nm)
- Quad core to hex core: done on the same node (32nm)
Just like AMD (Bulldozer excepted). The real leaps in CPU power come on the same node, which is completely the opposite of what TheChiefO claimed. And we're not even counting the clock speed increases - the raw performance jump from the first Phenom 2 X4 to the last Phenom 2 X6 is really quite big.
This isn't true. MS launched with a piddling amount of consoles instead of millions they could have sold on day one. Supply was constrained in at least the UK and America until after Christmas.
So you're more likely to run into problems with a big chip early on in the node, but this is ... somehow ... not ... relevant? C'mon man, think about it.
The 480 was late and hot with poor yields. The 580 was a very similar chip but came when the process was more mature. The 580 was 10 - 20% faster in games while using 10 - 20% less power and yields were vastly better. All on a process where there was apparently no difference start to finish!
A chip like the 480 (late, hot, slow) would be an unmitigated disaster for a console vendor.
A bigger chip later on?!?
The gains in the first 12 months were fairly big.
March 2009: Mobility Radeon 4830
April 2009: Radeon 4770
September 2009: Radeon 5870
January 2010: Mobility Radeon 5xxx
So small chips to start, then six months in the first performance part, then 10 months in the mobile, power efficient performance parts. That is not a surprising pattern.
Oh man!
Later in the node you can get acceptable yields on bigger chips, and run the same chips faster within the same power budget. This is relevant to consoles. That, in a nutshell, is it.
Seriously?
I made mention that if MS/Sony were to launch in 2012, they would have the choice of GF 32nm or TSMC 28nm.
Lately the 28nm of TSMC has been getting more positive news reflecting the possibility of 28nm in 2012.
The only thing I've been "pimping" is that 2012 has a very real shot at introducing real next-gen consoles.
Where did I claim this?
DuckThor Evil
Legend
That sort of comparison is flawed by design. Chip companies like AMD and Intel have different motivations and timetables for their decisions than a console manufacturer trying to pick the optimum timeframe for launch. Often the first products that come out of a new node is a pipe cleaner that helps to ramp up the production line for other products. This saves money for them and they can use these products for something. I don't see any point in using a cost cutting measure or a pipe cleaner product as some sort of base here. I'm sure they could make a Pentium 1 on 32nm and then the jump during that node would be amazing...
You can't use something like "Athlon 64 X2 90nm to Athlon 64 X2 65nm offered basically no improvement in raw performance, but a small improvement in perf/watt."
as an argument here, because improvement in raw performance was never intended in that shrink. It was purely a cost cutting measure. No-one can expect a huge gain in performance in situations like that. Perhaps better overclocking a bit. Architectures don't usually scale in a way that sees huge improvements there. The 65nm chip was about half the size and had less cache...Raw performance was not the target there and most of your other examples are similar or flawed in some other way (2x bigger die giving 2x performance only due to the fact that the initial die was tiny). The optimal strategy for chip making is not the same as launching a game console.
My quess would be that the CPU in WiiU wont be a very aggressive design and therefore it doesn't require smaller process than that. New process also will take time to become cheaper than the one that still has high volume. Nintendo's design might fit nicely to older process and get costs benefit that way, it also puts a tighter ceiling on what you can put there.
I don't quite understand what you are saying here. Intel launched Hex core on 32nm (i7 980x) in March 2010 before any quads on that process if I'm not mistaken, but there were dual core Clarkdales slightly before that. Sandy Bridge was the first 32nm quad early 2011. Intel is making anything between 2-8 cores on 32nm currently, although two cores are disabled on the native 8 core chip (3980X) I'd definitely say that Intel's Tick Tock strategy has always brought major architectural changes on a new node. Core 65nm, Nehalem 45nm, Sandy Bridge 32nm.
In any case Intel's doings aren't really relevant either, because they are playing different game also.
360 launched November 18 in NA and early December in Europe. By February it was quite easy to find a unit. PS2 for example had far bigger shortages. 360 supply was able to meet the demand quite quickly. I'm still happy with my 1-2 months, but if it helps add one or two more as it doesn't change anything.
I brought that up, because that example is the best and proper argument for your angle and actually what you should have brought up in the first place instead of something like Athlon 64 X2 90nm vs 65nm.
My point at the same token was that, as that truly is the best case for your angle. Those percentages happened on a absolute monster 500mm2 chip that would never have any business inside a console. Smaller mid range-ish chip is not going to run into problems of that magnitude and therefore in the console realm the problem is not going to be as big as in that example.
A bigger chip later on?!?
It's not that much bigger, it was quite a modest improvement in every way, that was just to get a new product out there and it would have closed the performance gap AMD was behind to GTX 480, enabling AMD to rise their prices to higher level, but then came the GTX 500-series and AMD was back to square one or actually a bit worse.
Those are not gains. Different products for different purposes/goals
Later in the node you get better yields yes. How much that matters if you compare console life cycles of 2012-2020 vs 2013-2020 is debatable. Especially if the earlier launch helps you in other ways and the chips wont be that big in the first place. 10-20 % percent faster console one year later is probably not going to light the world on fire and that boost would probably drown to the fact that in the alternate scenario the devs have had more time with the tiny bit weaker hardware.
Last edited by a moderator:
As I've said.
Hence the reason why I've been repeating the idea of using multiple GPU's having merit.
Splitting the gpu budget to two smaller chips for ... wait for it ... increased yields on a newer process.
...
- Status
Similar threads
- Locked
