R420 to support 3 types of memory interface?

[991060]

ATI'S TEAM headed by Dave Orton team has no intention of slowing down after its successful coup with the R300.
You will be run over if you stop for even a second in the graphics business and even R420 is still not totally cast in stone.

Someone at ATI is thinking of the next speed update of this chip that should come roughly six months after.

ATI is acting far sightedly. It will have support for all memory marchitectures in R420 as the chip will support DDR1 and that could mean cheaper cards, it will support DDR 2 that will be the memory of choice for this chip and then there's GDDR3 memory, which could be used in future speed upgrades of the R4x0 family.

We understand that the initial R420 will use DDR 2 memory which is finally stable and available, and can deliver over 1000MHz.

The R300, R350 and consequently R360 also had support for DDR 1 and DDR 2 as R350 had its DDR II version following introduction of R350, Radeon 9800 PRO 128 MB DDR 1 few months earlier.

The R420 will have 256 bit memory interface that can always be cut up to 128 bits if you need that.

Will NV40 work with GDDR3 and DDR 1. We think it should better have those capabilities. µ

http://www.theinquirer.net/?article=13748

It sounds a bit oddly, despite the source:the inquirer,never appears to be a reliable story teller, isn't the information itself hard to believe? We don't even have a GPU on the market supporting 2 types of memory. My question is that is such a design possible from a technical point of view? If it is, how's the cost to implement such a design compared with making 3 revisions on the silicon, with each supports 1 type of memory?

 

[Natoma]

It sounds a bit oddly, despite the source:the inquirer,never appears to be a reliable story teller, isn't the information itself hard to believe? We don't even have a GPU on the market supporting 2 types of memory. My question is that is such a design possible from a technical point of view?

The R350 does support two types of memory, as stated in the quote. You can buy a Radeon 9800 Pro 128MB, and that has DDR1 memory. I have a Radeon 9800 Pro 256MB, which has DDR2 memory. Same core.

If it is, how's the cost to implement such a design compared with making 3 revisions on the silicon, with each supports 1 type of memory?

I would think that implementing a different core for each type of memory would be far more cost prohibitive than making a single core capable of accepting different memory types.

 

[991060]

The R350 does support two types of memory, as stated in the quote. You can buy a Radeon 9800 Pro 128MB, and that has DDR1 memory. I have a Radeon 9800 Pro 256MB, which has DDR2 memory. Same core.

oh, my bad, R350 does support 2 types memory, how could I forget that??!! :?

 

[cthellis42]

Many seem to support two right now, so it doesn't seem like supporting three would be a big deal. (And gives vendors the most options during its lifespan.)

 

[Tokelil]

What would be the problem with designing the core to support 3 memory interfaces? It seems to me that timing and optimal cache size would be the two biggest problems. So cache size would only be optimal for one interface, which might make the cache a little bigger than necessary. Or am I way off when not seing it as a big engineering problem? (Compared to the rest of a DirectX 9 chip)

 

[Guden Oden]

As far as I know - which maybe isn't that far, I don't know - all the DDRs are fairly similar technologies. It's not really apples and oranges or DRAM and RDRAM, but more like slightly different flavors of the same thing.

DDR and DDR2 have some timing and voltage differences, but the pinout is the same and the basic protocol is also pretty much the same. GDDR3 on the other hand is from what I understand just about as similar as DDR2 as DDR2 is to DDR, with again voltage differences and the same pinout and basic protocol, but some other kind of signalling scheme which gives better performance at high clock speeds (meaning, past 1GHz data rate, since everything below that is fairly pedestrian these days ).

All this is very general of course since I'm not an engineer, I just sort of try to regurgitate what I've read in the past without getting too much of it wrong.

 

[Pete]

Cripes, I double-checked for the link that was b0rking my (admittedly anemic 800x600) table width only to discover that the troublemaker was Natoma's sig. Natoma, a little help here?

 

[Hellbinder]

They left out the most important part...

R420 Supports "Curiously Fast" LMNOP Ram...

Just like the Bitchi'n Fast 3d 2000

 

[Zaphod]

Cripes, I double-checked for the link that was b0rking my (admittedly anemic 800x600) table width only to discover that the troublemaker was Natoma's sig. Natoma, a little help here?

<offtopic>
What browser? What OS? Mozilla perhaps?

Could be that it is 'misinterpreting' (depending on who you ask) his '-' hyphens as non-breaking hyphens.
</offtopic>

 

[Pete]

Yep, it may have well been Firebird. I notice Mozilla also stretches things, while IE doesn't. I had no idea there were breaking and non-breaking varieties. You learn something new....