News & Rumors: Xbox One (codename Durango)

[function]

It's time to use smilies

"specifically familiarity with DDR3/DDR4"

DDR4, you say?

I'd given up on the idea of DDR4 going into Xbone after seeing the latency figures of the jdec standards. Maybe I should still be given up on the idea, but with MS sure to be desperately looking for long term cost reduction strategies for Xbone maybe they can find a way...

... or maybe it's for XboneOnePointFive...

 

[function]

EDIT: Oh! Micron is now sampling their 2133 chips at 1.35v (XB1 is using 1.5v parts). The timing of the job offer could be related.
http://www.micron.com/parts/dram/ddr3-sdram/mt41k256m16ly-093

Looks like they're going to be offering 1 GB chips. If they can stretch to 32 bit interfaces on those things (which could make sense for mobile too) then perhaps MS would have an out from that huge amount of mobo area dedicated to housing 16 chips on a single side?

DDR3 is really proving to have some staying power. 1 GB, 1.35, 2133 mHz chips are looking pretty competitive with the early DDR4 speed grades (esp considering CL).

 

[DJ12]

I doubt it has any implications for the xbox one, this has "we are thinking about the next box now" written all over it.

As is usually the case, are the hardware designers to sit on their behinds for a few years or just crack on with the next one.

 

[Dave Baumann]

...this has "we are thinking about the next box now" written all over it.

DDR3/4?

 

[DJ12]

Sure, if you have ddr3/4 knowledge that will transfer to whatever ddr is around in a few years surely?

 

[iroboto]

Correct me if I'm wrong, but in 5+ years HBM stacked die modules should be commercially available in most higher end systems.

 

[Dave Baumann]

Sure, if you have ddr3/4 knowledge that will transfer to whatever ddr is around in a few years surely?

Even GDDR5 is a very different beast to DDR3/4.

 

[dobwal]

"specifically familiarity with DDR3/DDR4"

DDR4, you say?

I'd given up on the idea of DDR4 going into Xbone after seeing the latency figures of the jdec standards. Maybe I should still be given up on the idea, but with MS sure to be desperately looking for long term cost reduction strategies for Xbone maybe they can find a way...

... or maybe it's for XboneOnePointFive...

Is there some technical reason not to go LPDDR3? SK Hynix is talking 1GB chips rated at 1.2v and 2133 going mass production at the end of this year.

 

[function]

Is there some technical reason not to go LPDDR3? SK Hynix is talking 1GB chips rated at 1.2v and 2133 going mass production at the end of this year.

16 1GB chips for the 1B Xboneownone would seem somewhat like overkill.

 

[function]

Sure, if you have ddr3/4 knowledge that will transfer to whatever ddr is around in a few years surely?

esram ... edram ... HBM ... GDDR5 ... MS seem to be pretty specific in terms of the kind of memory generations they're after. Perhaps?

 

[Dave Baumann]

http://majornelson.com/2014/10/20/x...sonalization-smartglass-features-now-preview/

Full breakdown of the November update, including backgrounds.

[Edit] Seems that backgrounds will initially be limited to unlocked achievement art, but a further update will allow any image by way of upload through the Media Player.

The bio "showcase" seems like a nice addition.

 

[dobwal]

16 1GB chips for the 1B Xboneownone would seem somewhat like overkill.

Why would you need 16?

 

[function]

16-bit interface for each chip, so you need 16 of them for the 256-bit bus.

Unless MS are going to be sourcing some of the fabled 32-bit DDR3 chips, which would allow them to reduce to 8 chips, they're going to be stuck with 16 chips.

 

[dobwal]

16-bit interface for each chip, so you need 16 of them for the 256-bit bus.

Unless MS are going to be sourcing some of the fabled 32-bit DDR3 chips, which would allow them to reduce to 8 chips, they're going to be stuck with 16 chips.

I think the SK Hynix chips are 32 bit. Single channel rated at 8.5 GBs.

 

[function]

There appears to be 1GB, 32 bit GDDR3 up to 2.6 gHz (effective), but I can't find any signs of 32-bit DDR3 products ... will keep looking though.

Edit: this perhaps?

http://www.skhynix.com/en/pr_room/news-data-view.jsp?search.seq=2235&search.gubun=0014

Edit 2: How would this be likely to fare cost wise against GDDR5?

 

[dobwal]

There appears to be 1GB, 32 bit GDDR3 up to 2.6 gHz (effective), but I can't find any signs of 32-bit DDR3 products ... will keep looking though.

Edit: this perhaps?

http://www.skhynix.com/en/pr_room/news-data-view.jsp?search.seq=2235&search.gubun=0014

Edit 2: How would this be likely to fare cost wise against GDDR5?

Guess it depends on the uptake in the high end mobile, tabs and ultrabooks markets.

 

[MrFox]

Looking at what DIMM manufacturers are using, maybe smaller chips are lower cost, if only because x16 parts are sold by the millions per week. An x32 die would be ultra-niche because it would max out at 2GB per DIMM so it's useless for the PC business. No volume, no competition, no cost savings?

I think it's funny to look at a PC's 8GB DIMM built with 16 chips, because it's incredibly small compared to the space required on the XB1 PCB. The size problem is not the number of chips. It's probably a combination of the huge number of traces for 256bit wide, the low number of PCB layers, and the precise trace length matching required for DDR3, specially at 2133. Going 32bit doesn't change the number of lanes that have to go to these chips. Wouldn't it be half the chips at more than twice the price, with no gain?

 

[function]

Routing to half the number of chips should be easier as it should, I think, be easier to keep trace length the same in a smaller area (shorter effective perimeter as the chip edge combined lengths should be lower), You would probably also save some complexity on address buses (fewer chips = fewer addressing lanes) and power supply.

This might also allow you to shrink the APU area needed for IO a little, perhaps, and save some additional power spent on off chip communication (no idea how significant that may or may not be).

So you could save a little board space and a little power, potentially. But as you say, it'll all come down to price. No point if it makes the platform more expensive.

 

[Deleted member 11852]

Looking at what DIMM manufacturers are using, maybe smaller chips are lower cost, if only because x16 parts are sold by the millions per week. An x32 die would be ultra-niche because it would max out at 2GB per DIMM so it's useless for the PC business. No volume, no competition, no cost savings?

Out of curiosity, is anybody shipping non-ECC DDR4 DIMMs yet? No doubt Microsoft is looking at DDR for future devices, and perhaps even for a retrofit for Xbox One; it's a low-clocked device so likely has acceptable tolerances enough with a controller redesign that both software and hardware wouldn't know the difference, although I personally wouldn't want to have to test it and qualify it was ok!

In addition there are still aspects of the DDR4 standard to be finalised, like low-voltage options. Traditionally these are more expensive (and probably why Xbox One uses 1.5v DDR3 instead of 1.35v DDR3) but one of the design considerations for DDR4 was more affordable low-voltage variations.

 

[MrFox]

Well, I think next gen would be at the very least HBM 2.

GPUs are already planned to have 1TB/s HBM as soon as 2016, so by 2019 we'd probably have 2TB/s and 64GB of unified memory.

It's hard to imagine this DDR3/4 job offering to be for anything other than a slim revision of XB1.