At the Hot Chips conference today, AMD CTO Mark Papermaster revealed details about the company's next-gen Steamroller CPU architecture. Steamroller promises to address Bulldozer's shortcomings, and AMD expects a 30% increase in instructions per clock.
A lot of it is down to the Pentium 4 turning out to be a dead end. Intel designed it to scale to 10ghz, with a massively long pipeline. That turned out not to be possible, so all the design sacrifices that were made (primarily a ridiculously long pipeline) turned out to be bad bets.
When Intel introduced Core 2 Duo, performance per clock in many cases doubled, on the same socket and process node. I'm unaware of a precedent for that, at least in recent history.
Then Intel a couple of years later rolled out Nehalem, with an integrated memory controller and hyperthreading, cementing their advantage in the server market. AMD has been playing catch up ever since.
If Intel's chips were half the performance today, AMD would be winning; though not by quite as much.
The big one is they fell behind on process technology. AMD (now Global Foundries)'s fab operations are at this point a full process node behind what Intel is doing. So as they release their bleeding-edge 32nm parts, Intel has long since worked out the kinks and is selling 32nm CPUs at budget prices. And simultaneously Intel is selling better-performing 22nm parts into the high end at much better margins.
During the Athlon/Opteron days, the silicon capabilities were much closer (does everyone remember the "paper launches" of various PIII speed bins, and of course the recall when they pushed a little too far?).
At the time ATI was 1/2 AMD's market cap. Too much of their capital got tied up in the acquisition and R&D fell off.
The overall vision is still very, very compelling (integrated FP co-processing). At the time, AMD figured it could beat Intel to the punch by buying one of the best FP companies out there. Instead they both(AMD/ATI)fell behind as merging the two companies took far more time, attention and capital than was originally anticipated.
At this point, it just comes down to Intel spending more in R&D (http://newsroom.intel.com/community/intel_newsroom/blog/2011...) than AMD makes in revenue (http://www.anandtech.com/show/5465/amd-q411-fy-2011-earnings...) means Intel is at this point just throwing money at rapid release fabrication tech combined with their tick tock being so effective (they had to be after getting wasted with the Pentium 4 stagnation). Even if AMD didn't blow tons of market cap on buying up ATI, they would still be looking at Intel just being an order of magnitude larger than them outright. They struck gold with the Athlon 64 line and that is the only reason they ever overtook Intel, which like other commentors mentioned, was because of Intels own hubris.
2) Memory controller/north bridge into CPU - correct
3) lower price point than intel
They had a better product, at a lower price == market share.
Mobile:
They missed (as did intel) the ultra-low power/mobile market
1) ARM owns the market here
Bulldozer:
Made their bet on a high throughput/high latency micro-architecture, which didn`t pay off. So their now stuck with un-competitive technology at a lower price point which to the market == cheap knock off.
To make it worse, their process technology is a generation behind intel which increases costs thus eats further at any price discount.
GPU:
Meanwhile, Intel is eating their "Fusion" style cpu/gpu integration with every release they make. Can`t remember the last time some buzz about AMD fusion but intel HD 2k/3k/4k is in the news all the time.
They should have made Pat Gelsinger CEO when he was turned down for the same job at intel.
Looks like too little too late. If Steamroller is still a year or more out, that will put it in competition with Haswell (DDR4 in the server part, AVX2 with FMA ops, etc.). Considering how much Ivy Bridge is already dominating and those kinds of memory bandwidth and FP throughput increases, Haswell looks like it will be a monster.
I don't follow AMD's stuff much beyond knowing that Bulldozer was not very competitive beyond the $100-120 price point. Is Steamroller set to change that or is AMD dying a long, slow death?
Also, is AMD going to do anything in mobile or is that all ARM moving forward?
ARM is introducing 64 bit chipsets in less than a year or so. The power savings will be worth the switch for cloud hosts. Most devices sold already run ARM.
When you say "mobile", what segment do you refer to?
AMD has a pair of processors, one named "bobcat" and one on the way that was just announced named "jaguar", which are supposedly aimed at light laptops and tablets. I haven't heard of any AMD products aimed at cell phones, though.
Intel, on the other hand, does appear to be looking to get into the smartphone market, so you may see some Atoms in future smartphones trading paint with ARM.
The article mentions "feeding the cores" several times; possibly Bulldozer's cores were sitting idle too much, and this is how AMD plans on improving the performance.
I'm not aware of any AMD ARM products. I don't know if they're working in that space. I've heard that Intel is making ARM chips, even though they still have the x86 crown.
I'm not aware of any ATI mobile GPUs either, while Tegras have been prominent in recent products.
It doesn't matter much if they can keep making incremental improvements at this point. They really need to increase their iteration speed. This update should have been out last year.
[+] [-] incision|13 years ago|reply
1: http://en.wikipedia.org/wiki/Athlon#Athlon_.22Classic.22
2: http://semiaccurate.com/2011/10/17/why-did-bulldozer-underwh...
[+] [-] reitzensteinm|13 years ago|reply
When Intel introduced Core 2 Duo, performance per clock in many cases doubled, on the same socket and process node. I'm unaware of a precedent for that, at least in recent history.
Then Intel a couple of years later rolled out Nehalem, with an integrated memory controller and hyperthreading, cementing their advantage in the server market. AMD has been playing catch up ever since.
If Intel's chips were half the performance today, AMD would be winning; though not by quite as much.
Core 2 Duo review (compare it to the Pentium D, which is dual core): http://www.anandtech.com/show/2045/11
i7 3770k review (the FX 8150 is AMD's flagship 8 core Bulldozer CPU): http://www.anandtech.com/show/5771/the-intel-ivy-bridge-core...
[+] [-] ajross|13 years ago|reply
During the Athlon/Opteron days, the silicon capabilities were much closer (does everyone remember the "paper launches" of various PIII speed bins, and of course the recall when they pushed a little too far?).
[+] [-] noobface|13 years ago|reply
At the time ATI was 1/2 AMD's market cap. Too much of their capital got tied up in the acquisition and R&D fell off.
The overall vision is still very, very compelling (integrated FP co-processing). At the time, AMD figured it could beat Intel to the punch by buying one of the best FP companies out there. Instead they both(AMD/ATI)fell behind as merging the two companies took far more time, attention and capital than was originally anticipated.
[+] [-] zanny|13 years ago|reply
[+] [-] pheon|13 years ago|reply
Athlon era:
1) Got HyperTransport - correct
2) Memory controller/north bridge into CPU - correct
3) lower price point than intel
They had a better product, at a lower price == market share.
Mobile:
They missed (as did intel) the ultra-low power/mobile market
1) ARM owns the market here
Bulldozer:
Made their bet on a high throughput/high latency micro-architecture, which didn`t pay off. So their now stuck with un-competitive technology at a lower price point which to the market == cheap knock off.
To make it worse, their process technology is a generation behind intel which increases costs thus eats further at any price discount.
GPU:
Meanwhile, Intel is eating their "Fusion" style cpu/gpu integration with every release they make. Can`t remember the last time some buzz about AMD fusion but intel HD 2k/3k/4k is in the news all the time.
They should have made Pat Gelsinger CEO when he was turned down for the same job at intel.
[+] [-] wmf|13 years ago|reply
[+] [-] tmurray|13 years ago|reply
[+] [-] programminggeek|13 years ago|reply
Also, is AMD going to do anything in mobile or is that all ARM moving forward?
[+] [-] gauravk92|13 years ago|reply
[+] [-] sliverstorm|13 years ago|reply
AMD has a pair of processors, one named "bobcat" and one on the way that was just announced named "jaguar", which are supposedly aimed at light laptops and tablets. I haven't heard of any AMD products aimed at cell phones, though.
Intel, on the other hand, does appear to be looking to get into the smartphone market, so you may see some Atoms in future smartphones trading paint with ARM.
[+] [-] stephengillie|13 years ago|reply
I'm not aware of any AMD ARM products. I don't know if they're working in that space. I've heard that Intel is making ARM chips, even though they still have the x86 crown.
I'm not aware of any ATI mobile GPUs either, while Tegras have been prominent in recent products.
Edit: I did find this: http://www.amd.com/US/PRODUCTS/NOTEBOOK/APU/Pages/tablet.asp...
[+] [-] dkhenry|13 years ago|reply