What sort of ram is xenon likely to use?

[Brimstone]

If the 'leaked' diagram showing 22.4GB/s bandwidth is accurate, I can't see Xenon's main ram being anything other than 700mhz 128-bit GDDR3 or 4. In light of Brimstone's comments about GDDR4, I'd guess GDDR3 would be better.

Apologies if this is stating the obvious. Or already known to be untrue.

The 22.4 GB/s number is most likely the HyperTransport 2.0 spec. Not the RAM used.

Sunnvyale, Calif., February 9, 2004 --

HyperTransportâ„¢ Technology Consortium today announced a major new release of the HyperTransport Technology I/O Link Specification. The HyperTransport Release 2.0 Specification introduces three more powerful bus speeds and mapping to PCI Express, an emerging I/O interconnect architecture. HyperTransport's speed capability extends from the 1.6 Giga Transfers/second (GT/s) of Release 1.1 Specification to 2.0, 2.4, and 2.8 GT/s using dual-data rate clocks at 1.0, 1.2, and 1.4 Gigahertz, delivering a maximum aggregate bandwidth of 22.4 Gigabytes/second. The electrical protocols supporting the new clock rates are backward compatible with all previous versions of the HyperTransport electrical specifications.

http://www.hypertransport.org/consortium/cons_pressrelease.cfm?RecordID=2

 

[AzBat]

It's a great question. May I suggest you make a poll in this thread.



GDDR-4 seems the most likely choice with its speed and density. I haven't read any detailed info on GDDR-4 but is obviously the next evolutionary step of GDDR-3. ATI designed the memory controller for the Xbox 2 and they're most familar with this class of memory.

I suppose the wild card would be some sort of Quad Band Memory version of GDDR-X.


My guess would be 512mb of GDDR-4.

My question is how about the role SiS might have with the memory selection? What class of memory are SiS more familiar with? I honestly don't know.

Tommy McClain

 

[function]

The 22.4 GB/s number is most likely the HyperTransport 2.0 spec. Not the RAM used.

Thanks.

22.4 GB/s bandwidth at 1.4 Ghz does imply 128-bit DDR 700 though, I reckon. I was just thinking that it's the kind of memory you'd likely be seeing in a mid range PC graphics card by late 2005 / early 2006.

I'm sure I read in one of the graphics card reviews here at B3D that connecting 256-bit memory to smaller chips like the GF6600 and X700 was currently likely to be problematic. Maybe this could be a factor in chosing 128-bit memory over 256-bit for the Zenon (if it were to be the case) ... ?

 

[Brimstone]

The 22.4 GB/s number is most likely the HyperTransport 2.0 spec. Not the RAM used.

Thanks.

22.4 GB/s bandwidth at 1.4 Ghz does imply 128-bit DDR 700 though, I reckon. I was just thinking that it's the kind of memory you'd likely be seeing in a mid range PC graphics card by late 2005 / early 2006.

I'm sure I read in one of the graphics card reviews here at B3D that connecting 256-bit memory to smaller chips like the GF6600 and X700 was currently likely to be problematic. Maybe this could be a factor in chosing 128-bit memory over 256-bit for the Zenon (if it were to be the case) ... ?

I'd think it would be better for them to put in RLDRAM-II over DDR-II. The downside to RLDRAM-II would be density and cost. The advantage is almost SRAM like latency and 100% bus utilization. With so many cores having a memory pool fast enough to be considered a level 3 cache would be advantageous.

 

[Guden Oden]

There's almost zero possibility IMO that MS will use some sort of esotheric memory like RLDRAM in nextbox. There just won't be any mass-production benefits from such a move. "Nobody" else will be making memory chips like that, so they can't source such chips from multiple sources should they need to, and there wouldn't be competition in the market to keep cost down. Also, RLDRAM die size is larger than straight DDR memory, I'm not sure how much, but perhaps considerably so. That would also raise costs.

 

[Brimstone]

There's almost zero possibility IMO that MS will use some sort of esotheric memory like RLDRAM in nextbox. There just won't be any mass-production benefits from such a move. "Nobody" else will be making memory chips like that, so they can't source such chips from multiple sources should they need to, and there wouldn't be competition in the market to keep cost down. Also, RLDRAM die size is larger than straight DDR memory, I'm not sure how much, but perhaps considerably so. That would also raise costs.

RLDRAM is multi-source. Micron and Infineon both produce it. Also RLDRAM will shop up in telocommunication equipment and just might end up in HDTV's. It will be at a premium but, not so much more than XDR and the performance clobbers XDR in certain aspects.

I think the chance are slim, but hardly far fetched.