AMD launches its Quad-Core Opteron; Phenom expected in December

It has taken years to get some desktop-level adoption of x64. And even today, 4 years after it was made available in Opteron, we're not even close to it being widespread.

I wouldn't hold out hopes of sudden amazing applications for some crazy hybrid CPU. If something tries to break the x86 mold, it takes years and that's even if it gets attention. And if both AMD and Intel don't standardize something, it may be never. Or you get pathetic levels of support where the only notable use is sponsored (ala 3DNow). To me, this Fusion thing seems like something that wouldn't be possibly mainstream until like a decade or more from now.

 

http://www.dailytech.com/Stumbling+in+the+Aisles+Barcelona+Thoughts/article8804.htm

 

LoL, so now barcelona gains 5%+ performance from a silicon revision? WTF? I could see voltage / power performance changes...

Now I think they're just making shit up.

 

You think they could come back somewhere once the whole AMD/ATI synergy kicks in? Or is that unlikely ...

 

They'll just silently over clock the chips without anyone noticing.

 

I haven't seen any transistor breakdowns for the design, and it might be against AMD's best interest to give that kind of data away (not that there aren't reverse-engineering companies that do just that).

Nonetheless, GPUs compared to CPUs are extremely packed with execution units.
R600 is a big chip, with ~720 million transistors.
Barcelona, with four cores, is ~463 million.
I'm just going to make a number up and say that 1/4 of R600 is devoted to the stream processors.

If they are integrated into Phenom like all the other units, as you seem to describe, then they will be wired into each core.
If it is 1 R600's worth of SIMDs per core, that would almost triple the size of a Phenom processor.
If it is 1/4 of R600's processors per core, it would bump Phenom to 600 million transistors.

The R600 per core would be massive, and there are other considerations, such as how one would hook 320 stream processors to an x86 register file that tops off at 40 integer and 120 SSE/FP non-speculative and rename registers for K8 (Phenom should be the same for integer, not sure about FP).

The big version would have 4 SIMDs with 3 16-wide read and 1 16-wide write port each.
That's 12 read and 4 writes.
K8 had 5 and 5 for its fp register file.
I'm not sure what Barcelona's count is at at the moment.
Regardless, the width of the ports would also be larger.
Barcelona would have 128-bit ports, while the R600 ports would be 16*32 bits wide.

However, without a revamped decoder and additional instructions, the VLIW instruction packets would not exist to be decoded, and the streaming units would be saddled with a register file so small they would likely be able to call almost as many software-visible registers per clock as there are physical registers on the core.

The full R600 units per core would also require the widening of Phenom's issue rate. R600's 4 SIMDS operate on their own instructions, so the 3-wide decode rate of K10 would hinder it.

The 1/4 R600 per core could be folded into the standard issue width.

To sum up, any such change would range from insane to huge, especially for AMD, which is struggling to fab a chip as large as Barcelona.

 

It probably depends on your environment, but in our HPC applications I'd guess that the number x86_64 codes exceeds those run 32-bit mode by a fair margin (and for quite some time, too). If you're talking about running a 64-bit version of Windows on the average desktop, then you're probably right.

 

What is going on with the stepping numbers?

Rumors are that more than one stepping has been shown in preview, with the 2 GHz being B1 and that the 2.5 GHz model is a B2.
And yet others are saying that neither stepping is what we'll be seeing in retail.

On top of that, there is nothing in the way of availability in retail for any stepping.
Which big customers are snapping the chips up, and what kind of discount can they wrangle if they are buying up the older steppings?

 

A couple of guys posting at Extreme Systems with actual hardware are suggesting the B2 stepping corrects errata which currently require BIOS work arounds. It's reasonable to accept having the chip function as intended would gain performance over a defective stepping.

While like most I was a little disappointed with the Barcelona launch I'll reserve judgement till we have Phenom out an thoroughly tested before (finally) upgrading.

 

Going from an article on the Barcelona-architecture (and comparing it to the K8 and Core2) in the new issue of the German c't magazine, it seems that the K10 is not as much of a dud as most people seem to make it out — for its intended market that is.

In any case, I'll report back in a few months with comparisons between a node of our Barcelona-cluster and my Xeon 5150 workstation...

 

What is it that AMD needs right now to stay competative?

 

Access to Intel's fabs and a few billion dollars capital

 

Well, Barcelona 4S CPUs are definitely keenly priced, so that might at least interest mainframe/large-scale server customers.

Now, will Agena (Phenom) be an interesting deal for the middle-range/high-end desktop consumers? That remains to be seen.

Note that I don't even discuss whether these parts would be interesting deals for AMD from a net margin standpoint. Because, things dosn't look all that great, to say the least, on that front.

 

Sounds to me as if they have to go back to the drawing board and create a truly revolutionary processor and not an evolutionary one to get ahead of Intel. Question is, do they have the capital? Living off of cheaper solutions is not a very cost effective method in the long run, unless they will hence forth be targeting budget consumers.

 

Agreed 100% AMD needs a revolution, and native quad-core with relative parity to Core 2 (assuming 2.5GHz+ before year-end) ain't it.

 

Well, a revolution would be nice, but not that it's that easy to accomplish. I don't even think we can expect Nehalem to be revolutionary, despite Intels deep pockets. And being tied to the x86 architecture also make a revolution harder.

No, I still think AMD should be able to improve along the evolutionary path as well. Intel has already proven that it's possible with Core 2, which is more of an evolution from the Pentium 3.

Per

 

Very true. Sandy Bridge may very well be revolutionary though (at least from a performance perspective) 200 DP GFLOPs is nothing to sneeze at in a commodity processor.

Please. Core 2 is such a departure from the PIII it's not even a valid comparison. You may as well be comparing the Pentium to the 8086. Sure, Core has its roots in P6, but it's sooooooo much more than that.

 

Comparing the Core 2 as an "evolutionary step" to the Pentium 3 would be somethink akin to comparing the X2 to the K6-3 of yesteryear.

In other terms, no, they don't work like that. The original Banias (Pentium-M) architecture was "linked" to the P3, but only in terms of IPS throughput rather than any architectural relation. The Banias was it's own design, built of components that were primarly targetted for power efficiency per-computation first, then reduction of heat dissapation second; clockspeed was not the main goal. This was more than certainly not the aim of the P3.

The P3 ran the entire core at full-voltage full-time, the Banias ran multiple voltage domains, and was able to fully disable sections of the core while not in use -- to save both power and reduce heat. It also reduced those voltage domains in relation to clockspeed, while not affecting bus speed, which is yet another set of features missing from the P3. From an IPS perspective, Banias did better than the P3 at integer performance, but due to it's mobile focus did worse at floating point. Banias also featured micro-ops fusion for grouping of SIMD-like capabilities with it's extra register space.

So no, contrary to popular misconception, the Banias was not a direct descendant of the P3 architecture. It certainly had influences, but it was not the same.

As for AMD getting back in the game? They need a whole new design from the ground-up, just like Intel's Isreali fab did with the Banias core. That, or a miracle.

 

Ok, then I've learned something today... thanks!

Not that the modern human has that much in common with our 1 million year old anscestors, but still we're only an evolution over her... not a revolution!

 

Quite the contrary; most indications suggest that we're very similar. The differences from then until now are primarily "societal", such as eating habits, bathing habits, education, and a bit more grey matter. DNA-wise, we're likely less than 1% different than we ever were. Less than 10% of your DNA has to change for you to pop out as any of the hundreds of thousands of mammals on this planet; you know this right?

No, the Banias core is still not a simple evolutionary step away from the P3.