AMD Bulldozer Core Patent Diagrams

swaaye said:
I want to see a per clock comparison against Kentsfield.;)
Click to expand...

it's slower than yorkfield probably slower than kentsfield too in gaming..

img0033850nuhi.gif

Indice 100 = Deneb

Efficiency is worse than kentsfield

http://www.hardware.fr/articles/842-12/consommation-efficacite-energetique.html
 
Last edited by a moderator:
That is some truly horrid performance per clock on games. It's like the K6 days are back.
 
I'm doubtful that it would be for anything like the current BD.
For a high-performance CPU process, GF is still the only game in town for AMD.

Perhaps a lower end APU with a BD successor on it could target a TSMC process someday.
It wouldn't be for a while, though. AMD couldn't promise volumes on leading edge with TSMC, as AMD learned with 28nm, and BD would probably be too big to port to the volume 40nm process.
 
This is at least the third time Maier's old posts have come up in this thread. Note the actual date of the source posts, and if we go back through this thread we can see those claims discussed multiple times.

There are likely good reasons in those posts.
However, he left quite some time ago. While he claimed to have further gossip since then, he represents a somewhat bitter snapshot of a time long since past.

I also question his context. He mourns the loss of heavy custom design and the shift to a more ASIC-type design process.

What we don't know is whether AMD could have afforded to maintain that level of design effort as time went on, nor do we know how much of that old school of thought would have benefited at 32nm and below. As the physical realities of manufacturing have changed, some of the old tradeoffs no longer apply.

That being said, I believe even AMD has admitted some loss of circuit performance due to their new design flow.
I think BD still has custom paths in critical parts, just a heavier bias towards automated design than prior.
 
Where does AMD go from here? They will only get further behind with each release from Intel. I don't see how they can survive this in the long term. Who in their right mind would hold on to AMD stock?

If Intel decided to squash AMD it would take maybe a couple of months. But of course they won't do that since it helps them legally if AMD gives the illusion of being a competitor.
 
I don't really know.
They will probably have to stick with the direction they have, with incremental improvements to x86, more Fusion APUs, and praying they get a process update every 2-3 years.
If they manage to trim down their costs to fit into a niche as a 2nd-tier value alternative, they could persist for a while since they no longer worry about funding process development.

I think that's a driving impetus for much of what they've done so far, from the adoption of ASIC-type design flows and going fabless. BD is a philosophy with an eye to answering the question "how do we design a CPU when we're pretty sure we won't be too good at building/paying for it?"
The pipeline was lengthened to reduce complexity per stage, its threading model reduces validation effort, it has features meant to make it more tolerant of process variation (not unique to them, but probably more critical than ever), it has a nortbridge and L3 arrangement that iterates what was done before, and it has created a slim core that will hopefully not lose too much performance on a weaker process.
That maybe sort of worked.
 
But they don't even have "oh well just wait for Bulldozer and everything will be good" any more. There is nothing to look forward to. They are screwed.

I knew BD would be bad, but I didn't expect it to be wreck the company bad.
 
3dilettante said:
I don't really know.
They will probably have to stick with the direction they have, with incremental improvements to x86, more Fusion APUs, and praying they get a process update every 2-3 years.
If they manage to trim down their costs to fit into a niche as a 2nd-tier value alternative, they could persist for a while since they no longer worry about funding process development.

I think that's a driving impetus for much of what they've done so far, from the adoption of ASIC-type design flows and going fabless. BD is a philosophy with an eye to answering the question "how do we design a CPU when we're pretty sure we won't be too good at building/paying for it?"
The pipeline was lengthened to reduce complexity per stage, its threading model reduces validation effort, it has features meant to make it more tolerant of process variation (not unique to them, but probably more critical than ever), it has a nortbridge and L3 arrangement that iterates what was done before, and it has created a slim core that will hopefully not lose too much performance on a weaker process.
That maybe sort of worked.
Click to expand...

Their saving grace for desktops will come from their ownership of ATI; if they sell a well balanced APU, they can sweep the fact that they don't have the best performing CPU under the rug. Well integrated GPUs which will eventually share CPU memory space will definitely net you wins for supercomputer designs that seems to be increasingly GPU oriented. The rest comes from having a primary design path forward in the high margin / highly threaded server space that, as you say reduces the costs in R&D + validation and yields very serviceable trickle down desktop parts.
 
I'm not sure HPC is going to be that big a win. That market is more price-sensitive than server.
AMD almost completely screwed Cray with its delayed shipment of BD. Oddly enough, Cray put itself in this position again after AMD did something similar with Barcelona.

Intel's graphics are not too great, but it is working on a lot of projects to fix other bottlenecks like bandwidth, programming tools, and communication between CPU and GPU. AMD is behind on all of these things, and it all costs money. Maybe if BD helps it in the server market, there could be more cash to spend. There are downsides to BD there as well, however.

AMD's reputation is also damaged. The problems with meeting demand or contractual obligations are a repeat of earlier woes.
 
3dilettante said:
I'm not sure HPC is going to be that big a win. That market is more price-sensitive than server.
AMD almost completely screwed Cray with its delayed shipment of BD. Oddly enough, Cray put itself in this position again after AMD did something similar with Barcelona.

Intel's graphics are not too great, but it is working on a lot of projects to fix other bottlenecks like bandwidth, programming tools, and communication between CPU and GPU. AMD is behind on all of these things, and it all costs money. Maybe if BD helps it in the server market, there could be more cash to spend. There are downsides to BD there as well, however.

AMD's reputation is also damaged. The problems with meeting demand or contractual obligations are a repeat of earlier woes.
Click to expand...

This is all very true; what I described was AMD's plan to continue hobbling along as opposed to seize 50% market share. As usual, AMD will have to undercut Intel's price to survive.
 
fellix said:
AMD Bulldozer Registry Fix
Click to expand...

Sorry but that link doesn't have any sort of proof. Even people commenting there are asking for proof. And supposedly, he is deleting other people's comments. Even people from a site(Kubuntu) he mentioned, where he got the patch from have no idea what he is talking about. Yet I still hope for a miracle that it is true.. with bated breath...
 
Considering people from AMD working on the kernel patch were talking about 3% improvement I'd take that 40-70% with a truckload of salt.
 
Rootax said:
Phenom I were at least competitive against older AMD generation, no ?
Click to expand...
Mostly - yes, but there were quite few instances where Phenom slightly lagged behind the top A64 X2 models or even outright took the last place. Later on, the TLB bug patch sliced off a couple of percents on top of this.
 
Well I thought Phenom I was terrible too. ;) Phenom II was acceptable but still sometimes slower per clock than Kentsfield. Their pricing saved the day obviously.
 
You must log in or register to reply here.
Back
Top