Bigger hardware flop

[Bjorn]

If you're shooting for a quiet system, then how about waiting for a 90 nm Athlon 64 ? (Which is rumoured to run extremely cool)

AMD has specified a TDP (Thermal Design Power) of 89 Watts for its 0.13u Hammer parts; IIRC this number is increased to 103W for its 0.09u parts. So if noise is a concern, I wouldn't hold my breath for AMD's 0.09u chips.

I was under the impression that the current 0.13 u Athlon 64 doesn't dissipate anywhere near 89W. And that the spec's are there just for future upgrades. Same goes for the 0.09u parts i guess.

http://www.aceshardware.com/read_news.jsp?id=75000419

And i think there were some rumours of 0.09u samples of the Athlon 64 the dissipated around 45W. Don't know at what clock frequency though. It was only a rumour though so i might not matter that much. But i think that it's a good guess to say that similarily performing Athlon 64's will dissipate much less power then their Prescott counterparts.

 

[arjan de lumens]

OK.. the TDPs appear to be far above what the chips actually draw, probably set that way to prevent people from making bad/underdimensioned cooling solutions. Still, that they increased it indicates that they do expect at least some future 0.09u parts to exceed the current 89W figure.

As for the 0.09u Athlon64 samples, the rumors I've heard indicate that they so far run at only 800 MHz. If an 800 MHz version dissipates 45W, that does not bode well for cool processors. (Although most of it is probably leakage current, which is independent of clock speed).

 

[Dave Baumann]

Prescott has a whiff of Willamette behind it - give it time to scale up in speed and have a couple of enhancements (better process understanding, possibly a few minor architecture tweaks) and probably in about 6-12 months it'll come into own. As Joe say, it does appear to be a relative achievement that they've managed to add as many stages in the pipeline and drop such a small amount comparatively - to the end user, though, they only heard of the "good" things (SSE3, better branch prediction, larger cache) so hearing that it is a shame to see it land a little slower in most cases.

However, "flop" can have quite a wide scope, and when considering that I would tend to also take into account who its coming from and what it does to the company.

Intel is pretty huge and has a commanding presence with the OEM's - if Intel looses significant marketshare to AMD then Prescott (and possibly their current processor direction) may be measured as a flop down the line, but my guess is that while there may be some market share loss in the interim I doubt it will be too significant. It'll be interesting to see how quickly they can get to gips with Prescotts heat and power requirements and scale the clocks up significantly.

XGI on the other hand really was coming from nowhere, and I really wasn't expecting too much because of the complexity of modern 3D chips and how good NVIDIA and ATI are at it that the moment. I would have hoped they could do more, but I'm not too surprised where it is. However, I think its still a reasonable achievement to come in from cold and produce a working (most of the time) DX9 part.

NV3x is a highly capable architecture and massively competant in many way, however I think it looses in two ways. Clearly the public perception is somewhat wayward from the reality once it was released - the PR had to go this was because it was late and ATI had a very competant part available, the PR had to keep people from purchasing ATI by trading on NVIDIA's history in releasing parts - with the competetive performance on the outcome this may have harmed the longer term perception. Secondly, it was probably more that ATI surprised than NVIDIA dropping the ball very much (beyond the delays) - architecturally I see a lot of similarities with what they have done in previous generations to NV3x: accelerate current games now, provide developer functionality for the future.

 

[pascal]

Maybe prescott has dual core on die.
See this thread: http://www.abxzone.com/forums/showthread.php?s=154c6537b12232a6bec716593841731d&threadid=65059

 

[Guden Oden]

Prescott is not, I repeat, NOT a dual-core chip. You can see that just from looking at a picture of the die for cryin out loud.

With a bit of logical thinking, why would Intel manufacture and sell dual-core chips with just a single core enabled? That's NUTS. They'd be wasting the entire second core for no reason.

So no, that idea is totally, completely wrong.

 

[AlphaWolf]

With a bit of logical thinking, why would Intel manufacture and sell dual-core chips with just a single core enabled? That's NUTS. They'd be wasting the entire second core for no reason.

I agree that it's not a dual core, but the concept is not without precident. Crippled chips are often sold in another form to improve effective yield or even just to fill a market segment.

 

[Luminescent]

In the thread about prescott, one of the forum members posted the following, which I was about to post myself:

Going by Hans DeVries' account, it's actually 64-bit in the works, not dual-core.

http://chip-architect.com/news/2003_03_26_Prescott_clues_for_Yamhill.html

A dual core CPU should have dual FPU, dual ALU, dual pipelines, and a single mulitiplexing MMU (basically two physical CPUs sharing a single MMU). Nothing on the Prescott core photos seem to indicate the FPU is duplicated for example.

Chip-Architect is a really good website. Read his bits on Athlon 64 and you'll know he knows what he's talking about!

-kONGO

Prescott may contain two integer cores, not complete processor cores; perhaps it is in preparation for the addition of 64-bit extensions at some point (maybe enabled in another revision of the core or another one altogether). Remember, until northwood, hyperthreading was not enabled for the P4 core.

 

[Nappe1]

1.) MicroChannelArchitechture (=MCA Bus)
2.) Pentium Pro.

 

[pascal]

Pentium Pro was not a big flop. It is the first release of the P6 core which include the PII and PIII

My guess with the Prescott chip size, transistors and power dissipation you could get 4 way SMP with 4 P3-S at 2.4GHz