R500 will Rock

[rwolf]

Remember that ATI licensed Fast14 Technology from Intrinsity.

http://www.intrinsity.com/home.htm

Well Intrinisity has update their site with 90nm and 65nm performance projections.

Higher Performance
CMOS process improvements no longer yield dramatic performance increases. Fast14 Technology has achieved multi-GHz performance in processes ranging from 180nm to 130nm. Performance of 400 MHz to 4 GHz is achievable with Fast14 Technology at 90nm and 65nm without requiring exotic processing techniques.

Looks like the math engine is fast and probably more important is that consumes far less power. This might mean that when ATI goes 90nm they may have lots of extra transistors to play with without wicked overheating.

Edit: Url

 

[991060]

I'm an EE idiot, who can tell me, if ATI applies fast14 in the design of R500, will the manufacture of the wafer(TSMC in this situation), automatically realize all the promised advantages of fast14?

edit: replaced some misleading words

 

[chavvdarrr]

Is it even possible that any fast14 technology could be used in R500, ATi licensed it seveal months ago, right?

 

[pc999]

I think it was been licensend about 1 year or so.
I realy belive that this thing is (also ? ) for XB2,cheaper, fast,cold and with a GPU a lot more general processing than ever before ...
And P.G. from MicroprocessorReport had said that it is very possible (or he belives on that possibility ...).
I cant wait

BTW it is easy to put this in PC GPUs

 

[RussSchultz]

I'm an EE idiot, who can tell me, if ATI applies fast14 in the design of R500, will the manufacture of the wafer(TSMC in this situation), automatically realize all the promised advantages of fast14?

I'm not sure what you're asking.

Will all of TSMCs wafers (even for other designs) suddenly have the advantages of fast14? No. Fast14 is (likely, as I haven't really looked at it) a layout technique and possible an optimized standard cell library. Looking at their webpage, there's some clock tree optimizations and other design methodology--but its only going to be applicable to people who license fast14 and use their design methodology. TSMC gets no 'backdraft' out of it.

Will TSMC have to do some special work for ATI to utilize fast14? Dunno. Maybe. Part of the mojo might be extra mask steps, or a tweaked process or something.

 

[Sage]

there is no R500


duh

 

[Kaotik]

there is no R500


duh

Huh?
I was under impression R500 = Xenon/XBox2 VPU?

 

[Sage]

Huh?
I was under impression R500 = Xenon/XBox2 VPU?

i was under the impression that Xenon had a name other than R500.

 

[KimB]

I'd like to say that just looking at heat concerns, and comparing them to CPUs, I seriously doubt you're going to get a 200+ million transistor part in the multi-GHz range (remember that CPUs are mostly cache, while GPU's are mostly logic, and therefore are typically much hotter). The test parts are most likely much smaller, and thus heat wasn't as much of a concern.

They may gain some efficiency from licensing this technology, but it's more likely to give gains on the order of 10%-20% or so.

 

[Leto]

If this Gigahertz 200+ million transistor mojo works it would be like a 2x leap in performance at least :? Sounds a little far fetched

 

[KimB]

Well, it's not like we haven't seen doubling of performance before, but I just doubt that this buys ATI that much over what else is available out there for transistor layouts.

 

[pc999]

I remember that ATI expected to win 4X the performance.

 

[Ailuros]

I probably live in an entirely different dimension, but I haven´t heard anything about GigaHz frequencies for Xenon. Rather like something more than half of it.

 

[sexualrug]

sort of off topic, but, any concensus for an approximate time frame of r5x0 release for the desktop? Will it be a PCI Express only part?

 

[MasterBaiter]

sort of off topic, but, any concensus for an approximate time frame of r5x0 release for the desktop? Will it be a PCI Express only part?

It's probably safe to say that it will be PCI Express only. My guess will be that we will see it in summer 2005 (June/July?). A more accurate appraisal most likely will be found here though: http://www.mattelgames.com/magic8/flash_index.asp

 

[hovz]

id say actual availability, late 2005

 

[jvd]

sort of off topic, but, any concensus for an approximate time frame of r5x0 release for the desktop? Will it be a PCI Express only part?

It's probably safe to say that it will be PCI Express only. My guess will be that we will see it in summer 2005 (June/July?). A more accurate appraisal most likely will be found here though: http://www.mattelgames.com/magic8/flash_index.asp

It may be native pci-e but they will have a bridge chip to make it agp. There will still be a large market of people who will have ahtlon 64 3.5s or whatever that don't need a chip or mobo change but may want a video change.

 

[sexualrug]

I just wish they would include agp slots on mainboards for a while more, or if there were some type of socket convertor so one can install a pci express card into an AGP slot.

I'll tell you, I am not thrilled about purchasing a card that I cannot use in my next system upgrade. I like to upgrade my system a bit at a time

 

[rwolf]

I'm an EE idiot, who can tell me, if ATI applies fast14 in the design of R500, will the manufacture of the wafer(TSMC in this situation), automatically realize all the promised advantages of fast14?

edit: replaced some misleading words

http://www.intrinsity.com/press/images/IntrinsityLicensesFast14toATI.pdf

Intrinsity already has a 2.0GHz Math CPU it has a 1MB cache and consumes 15W vs 50W for a desktop processor and was fabbed on a .13 micron process at TMSC with 70Million transistors.

http://www.intrinsity.com/technology/docs/public/Intrinsity_Fast14_Technology.pdf

This technology runs cooler because it uses because it uses dynamic logic instead of CMOS technology, yet it can integrate on a chip with CMOS.

The potential is interesting, however we'll have to wait to see if it bares fruit.

 

[KimB]

I just wish they would include agp slots on mainboards for a while more, or if there were some type of socket convertor so one can install a pci express card into an AGP slot.

I'll tell you, I am not thrilled about purchasing a card that I cannot use in my next system upgrade. I like to upgrade my system a bit at a time

Platform changes don't happen very often, but pretty much always cause pains like these. Just save up for once. It won't happen again for a few years.