Fusion speculation. Will it be based on Phenom?

[MTd2]

AMD's Fusion chip will be based on Phenom processor
AMD's Fusion will be shrunk from the Phenom core and optimized to make the chip more power-efficient while increasing graphics capabilities

The Fusion chip, which will combine a graphics processing unit and CPU on one chip, will be a redesign of the company's current Phenom processor, said Patrick Moorhead, vice president of advanced marketing at AMD, in an interview. However, the Fusion chip will witness significant design changes from the current iteration of Phenom, Moorhead said.

The Fusion chip, code-named Swift, will be shrunk from the Phenom core and be optimized for use in a notebook, Moorhead said. The optimization will focus on making the chip more power-efficient while increasing graphics capabilities, Moorhead said.

The graphics processing unit on the Fusion chip will include multiple "mini-cores" that breaks down code from a program, like 3-D games, to process data faster, said John Taylor, an AMD spokesman. Fusion's graphics processor will be based on a graphics card AMD plans to release in the near future, Taylor said, declining to give details.

http://www.infoworld.com/article/08/01/25/AMD-Fusion-chip-will-be-based-on-Phenom_1.html

 

[ShaidarHaran]

Fusion is (was) supposed to be based off of Bulldozer (supposedly) a new architecture. Swift is more of a response to Nehalem w/graphics than it is "Fusion" in the original sense.

 

[3dilettante]

AMD has already said that the first Fusion chip will use a CPU based on current designs when it redid its roadmaps and pushed Bulldozer off into 2010 or beyond.
I guess a gen 3 STARS core is a redone Phenom, perhaps given a once-over by the Griffin design team.

By the way, it might be a good idea to only show selected passages from a news source. Pasting an entire news article is not considered good form.

 

[Sound_Card]

R700?

The graphics processing unit on the Fusion chip will include multiple "mini-cores" that breaks down code from a program, like 3-D games, to process data faster, said John Taylor, an AMD spokesman. Fusion's graphics processor will be based on a graphics card AMD plans to release in the near future, Taylor said, declining to give details.

 

[MTd2]

There is a short analysis here: http://arstechnica.com/news.ars/pos...dozer-taps-phenom-for-cpugpu-fusion-edit.html

I guess AMD will go the easy way here. I guess they are inverting the situations with Intel here. Phenom, complex, fusion simple. Core 2, simple ("fake 4 cores", Larrabbee, complex.

 

[INKster]

There is a short analysis here: http://arstechnica.com/news.ars/pos...dozer-taps-phenom-for-cpugpu-fusion-edit.html

I guess AMD will go the easy way here. I guess they are inverting the situations with Intel here. Phenom, complex, fusion simple. Core 2, simple ("fake 4 cores", Larrabbee, complex.

By the time "Larrabee" and "Fusion" come to market, it's likely that the Core 2 will be the least of AMD's problems.
The "Nehalem" µArch will be as much, if not more complex than "Larrabee", and there's a direct "Fusion"-like derivative of it on the roadmap...

 

[Kaotik]

http://www.tgdaily.com/content/view/38703/135/

TSMC to manufacture first Fusion CPUs, codenamed Shrike, at 40nm process.
The Fusion CPUs will feature dualcore Phenom based CPU part and RV8xx based GPU part

 

[ShaidarHaran]

Very interesting development. Excellent news for the graphics performance, no doubt, too bad about the CPU perf though, but better than the immediately-previous info.

 

[Tchock]

Very interesting development. Excellent news for the graphics performance, no doubt, too bad about the CPU perf though, but better than the immediately-previous info.

If Deneb scores 10%-15% upwards of Agena (in particularly optimistic moments? ) it should be good enough until Bulldozer. That's basically Yorkfield clock/clock perf, I don't think anyone would really ask for more performance on the CPU first than the GPU at that stage...

 

[Albuquerque]

If Deneb scores 10%-15% upwards of Agena (in particularly optimistic moments? ) it should be good enough until Bulldozer. That's basically Yorkfield clock/clock perf, I don't think anyone would really ask for more performance on the CPU first than the GPU at that stage...

We've seen Deneb scores in another thread over in the "Processor and Chipset" forum of this site, and it's not getting the performance you're suggesting at this point in time. Perhaps when it's final silicon, but only if it finds another ~10% performance bump between now and it's release.

 

[Tchock]

We've seen Deneb scores in another thread over in the "Processor and Chipset" forum of this site, and it's not getting the performance you're suggesting at this point in time. Perhaps when it's final silicon, but only if it finds another ~10% performance bump between now and it's release.

2 issues.

1. This C0 spin is pre-Agena B3, which means that the TLB manual workaround is still done in software instead of hardware.

2. Any SP1 release of Vista enforces the software fix- unless you do some changing in MSR, it will stay enabled.

The WinRAR scores evidently show the workaround in action, with a terribly heavy penalty.

I suggest we wait for the C1/C2 steppings as they would be free of this and be more of an indication of what we're getting. Yes I know, we're all sick of the waiting game. But sometimes they just won't leak.

 

[Albuquerque]

Even the SuperPI results weren't where they need to be (honestly, anywhere close) to validate that they will be within reach of the yorkfields clock-for-clock.

I don't mind waiting at all, but if these newer processors coming out can only hope to match the Yorkies, then they're still going to be well behind when Nehalem debuts the end of this year. Hell, at the rate AMD is going, Bloomfield may beat Deneb to market.

 

[ShaidarHaran]

AMD will *never* achieve SPI parity with Intel, unless they pursue an aggressive L2 cache scheme. Their recent addition of an L3 pretty much guarantees this.

 

[Thorburn]

Even the SuperPI results weren't where they need to be (honestly, anywhere close) to validate that they will be within reach of the yorkfields clock-for-clock.

Pentium M used to beat Athlon 64 at SuperPI, its not the be all and end all.

Bloomfield may beat Deneb to market.

I certainly wouldn't bet against it.

 

[I.S.T.]

Weren't Pentium Ms at the same clock speeds faster than the Athlon 64's of the day?

 

[3dilettante]

In many programs that didn't stress the weaker FP resources, Pentium M was better per-clock.

There were a few voices in the pre-Core2 days that pointed that out as a sign K8 would not remain dominant once the limits to the clocks and FP pipes were lifted.