[_xxx_]

Oops! Just saw that in the R520 photos thread and linked senselessly

[wireframe]

Originally Posted by Chalnoth

That does seem really, really strange, since I doubt it'd be possible to have the two work together without forcing the XT to act like a 256MB board.

Maybe we are just supposed to wait for the 512MB CrossFire 520XT at a later date.

I got a head cramp, but, given the current data, I think I will go with:

520XT 256MB Crossfire = Standard SKU (in other words, you don't buy a non-CrossFire 256MB 520XT)
520XT 512MB = Premium SKU with a possible CrossFire edition in the future.

[Xmas]

Internally the chip will use 512-bit memory transfers. However, the external memory interface will still be 256-bit - expect some nice marketing from ATI regarding that "feature".

It's 256-bit double data rate, so 512-bit internal is no surprise... However, I doubt it will use "512-bit memory transfers", i.e. 512-bit granularity.

Originally Posted by geo

Hmmph. If it didn't already say that it was 256-bit external, I might ask if there was any benefit to doing a "partial" 512-bit bus --in other words, just to some of the memory not all of it. . .

It's not a bus, it's a point-to-point interface to each memory chip. You mean, something like XGI does with XG45, having some memory chips 16 bits wide, and some 32?

Originally Posted by mapel110

hm, is it be possible that ATI has found a way to split the traffic to the 512 MB Ram in two parts? that could explain this 512 bit bus. 256 bit to 256 MB of Ram. But would there an advantage using the 512 MB that way?

hm, a design for two 256bit interfaces would be too big?!

Every modern GPU uses a split memory interface like this, usually with four external channels, each dedicated to one, two or four memory chips. But still everyone calls it a 256-bit interface, even if it's really four independent 64-bit interfaces.

[digitalwanderer]

Originally Posted by geo

I'm still thinking there has to be something Significant in the fact (if it is a fact) that XT is 512MB and XT CrossFire is 256MB.

I'm betting that's just a sticker price shock defense, it'll help to keep Crossfire affordable as I bet it adds a bit of a premium to the cost of manufacturing the card.

I'm still not excited at the idea of getting one, sorry Geo...the "no-AGP" version thing stings me; but I am a bit intellectually excited since we got more info to digest and I gotta admit that's fun.

[Richteralan]

anyone saw these 2 photi?

[Skrying]

Originally Posted by wireframe

Maybe we are just supposed to wait for the 512MB CrossFire 520XT at a later date.

I got a head cramp, but, given the current data, I think I will go with:

520XT 256MB Crossfire = Standard SKU (in other words, you don't buy a non-CrossFire 256MB 520XT)
520XT 512MB = Premium SKU with a possible CrossFire edition in the future.

If you get those, and a 512MB Crossfire comes out later you want to trade that R520XT to me?

[wireframe]

Originally Posted by Chalnoth

But, regardless, I don't expect Crossfire to really catch on unless ATI can bring their motherboard performance up to par. I seem to remember their previous solutions not being so great in performance...

Sorry for splitting your post into two.

Would this really be an issue? From the little I know of CrossFire it has no mainboard requirements (perhaps some superficial/artificial ones?) so wouldn't an Nvidia SLI x16 - or whatever they call the new one with dual x16 capability - work just fine in "Non-SLI" mode? Both cards would have their PCIe x16 and the driver does the rest.

I wouldn't be surprised if CrossFire worked on SLI mainboards (at least the x16 ones) but not the other way around. ATI was late the party and has less mainboard penetration. If they locked themselves into requiring an ATI mainboard they would be hurting. Might be useful for benchmarks, but not as easy to get a hold of in the real world where Nvidia has lots of market penetration with Nforce. Then again, ATI would be looking to use this to leverage its high-end mainboard sales, but then you really need to show some awesome performance to make the industry and customers get out of the SLI "comfort zone".

I'm mainly interested about thoughts on the first part. If CrossFire would be likely to work with a SLI x16 board (or is that x32, would seem more logical).

[digitalwanderer]

Richterlan, you're about two pages late...could you please edit your post to remove the picts as they mess up the page formatting?

[Richteralan]

anyone saw these 2 photi?

[Richteralan]

Originally Posted by digitalwanderer

Richterlan, you're about two pages late...could you please edit your post to remove the picts as they mess up the page formatting?

oops...

[wireframe]

My brain is working overtime, which means probably not at all, but did anyone else notice the scrambled square on the "front" image near the PCIe connector? What would warrant them doing that?

Here is the cropped section:

[aaronspink]

Originally Posted by wireframe

Ok, but I don't get why that would raise eyebrows. I presume you are talking about Nvidia and their "two on a slant" design. Are you saying that it would imply that the VPU is so wide that this layout makes better sense?

The implication that Dave isn't saying but is probably thinking is...

The NV40 and G70 implement a memory interface consisting of 4 64b interfaces. This is reflected in the dram layout of the 2 dram arrangement. Part of the reason to keep the chips close is to share the address/cmd bus between the 2 dram chips

The board for the 520 doesn't not appear to have any such organization and from the layout one could make the inference that the memory is organized as 8 32b interfaces. This would allow a higher efficiency utilization of the memory with the overhead of requiring 2x the DRAM controllers.

Aaron Spink
speaking for myself inc.

[Geo]

I'm thinking we are going to be, much more than anytime since early NV30, "pick your benchmark" arm-waving discussions. . .

[wireframe]

Originally Posted by aaronspink

The implication that Dave isn't saying but is probably thinking is...

The NV40 and G70 implement a memory interface consisting of 4 64b interfaces. This is reflected in the dram layout of the 2 dram arrangement. Part of the reason to keep the chips close is to share the address/cmd bus between the 2 dram chips

The board for the 520 doesn't not appear to have any such organization and from the layout one could make the inference that the memory is organized as 8 32b interfaces. This would allow a higher efficiency utilization of the memory with the overhead of requiring 2x the DRAM controllers.

Ok, so then we just assume there is some benefit to the physical one-on-a-slant versus the two-on-a-slant that Nvidia simply cannot exploit due to the pairing of modules. Because you could still put two on a slant with the situation you propose, and which makes perfect sense with the internal "double bandwidth".

[mapel110]

Originally Posted by Xmas

Every modern GPU uses a split memory interface like this, usually with four external channels, each dedicated to one, two or four memory chips. But still everyone calls it a 256-bit interface, even if it's really four independent 64-bit interfaces.

Yes, I can imagine that.
A 512 bit Memory Interface would be too complex, but what about two 256 bit Interfaces?

Gigabyte has two 6800 GT GPUs on one PCB, so there have to be two 256bit memory interfaces. So it should be possible, to get two 256bit Interfaces to work with one Chip.
http://www.digit-life.com/articles2/video/nv45-4.html
thats what I meant.

[3dcgi]

Originally Posted by Skrying

What will the 512-bit internal memory bus mean for the R520? Added bandwidth or is it some other factor? I'm not to keen on the two different buses thing, so could some clear this up for me?

Xmas made this point, but I figure it's worth repeating. Since the 256 bit external memory bus is DDR the internal bus is 512 bit SDR. I'll go out on a limb and say every graphics card with a 256 bit memory interface is really 512 bits internally.

Edit: Xmas is also correct that the memory bus is typically split up and not one huge bus.

[Skrying]

Originally Posted by 3dcgi

Xmas made this point, but I figure it's worth repeating. Since the 256 bit external memory bus is DDR the internal bus is 512 bit SDR. I'll go out on a limb and say every graphics card with a 256 bit memory interface is really 512 bits internally.

Edit: Xmas is also correct that the memory bus is typically split up and not one huge bus.

Then why would they bring it up now? That makes little since to me. Being DDR on the external wouldnt actually double the bandwidth bus would it? I've never heard of that, and I would expect if that was true then we'd have already heard a lot more as far marketing goes with that.

[mapel110]

ahh, forget about my theory. If there are two real 256 bit Interfaces you would be allowed to call it real 512 bit interface. but thats not the case obviously.

[Chalnoth]

Originally Posted by wireframe

Would this really be an issue? From the little I know of CrossFire it has no mainboard requirements (perhaps some superficial/artificial ones?) so wouldn't an Nvidia SLI x16 - or whatever they call the new one with dual x16 capability - work just fine in "Non-SLI" mode? Both cards would have their PCIe x16 and the driver does the rest.

Er, that's now what I remember from ATI's initial release docs. Ah, here we go:
http://www.ati.com/technology/crossfire/faq.html

5. What motherboard is required for a CrossFire system?

A. Radeon Xpress 200 CrossFire Edition and Intel i955X based dual-slot motherboards are supported platforms.

..considering that the vast majority of boards with two x16 slots are nForce SLI boards, well, this doesn't seem like a great move for ATI.

Ah, this is there, too, by the way

4. When will CrossFire graphics cards be available?

A. Radeon X850 and Radeon X800 CrossFire Edition cards will be available mid-August.
...
6. When will CrossFire Ready motherboards be available?

A. Radeon Xpress 200 CrossFire edition motherboards will be available from our partners beginning in June.

(while there are Radeon Xpress 200 motherboards available, a quick search through Newegg showed none with two x16 slots)

[Arty]

Originally Posted by Chalnoth

Ah, this is there, too, by the way

(while there are Radeon Xpress 200 motherboards available, a quick search through Newegg showed none with two x16 slots)

Beating a dead horse ? no-no back on topic ...

[3dcgi]

Originally Posted by Skrying

Then why would they bring it up now? That makes little since to me. Being DDR on the external wouldnt actually double the bandwidth bus would it? I've never heard of that, and I would expect if that was true then we'd have already heard a lot more as far marketing goes with that.

I don't know why it would be brought up now. The reporting web site might just be confused.

You have heard a lot of marketing about DDR.
512 bits @ 500MHz == 256 bits DDR @ 500MHz (1GHz effective).
The 1 GHz effective is the part marketing always picks up on.

[Chalnoth]

Originally Posted by serenity

Beating a dead horse ? no-no back on topic ...

Well, considering that no Crossfire solutions are yet available, I'd say the topic is still quite relevant. Just not much to discuss about it.

[wireframe]

Originally Posted by Chalnoth

Radeon Xpress 200 CrossFire Edition and Intel i955X based dual-slot motherboards are supported platforms.

Ok, so the inclusion of Intel i955X would suggest that there is really nothing special needed. It's just about driver supporting (recognizing) the chipset (this is what meant by artificial restriction). My point was that ATI has not locked themselves in to requiring some special function on the mainboard. It would seem that it is up to ATI to decide if CrossFire works on Nvidia's SLI x16. Because this board offers full x16 on both connectors it would seem obvious that you just set your dip-switches/BIOS to single board (non-SLI) and let ATI's drivers do the rest. However, the point is that enabling this is up to ATI, not Nvidia.

PS. I suppose it would be fair to say that Nvidia's "SLI" restrictions are also "artificial". Especially given how the discrete bridge is supposed to do all the work. That looks more like a way to get away from mainboard restrictions than impose them.

[Chalnoth]

That may be, I suppose, but not supporting the nForce SLI motherboards seems to be cutting ATI out of a huge portion of the market.

[wireframe]

Originally Posted by Chalnoth

That may be, I suppose, but not supporting the nForce SLI motherboards seems to be cutting ATI out of a huge portion of the market.

Well, sure, but the problem would be that you could use CrossFire on an SLI mainboard but not use SLI cards on a "CrossFire" mainboard. So, ATI has to play its cards, so to speak to get Nvidia to buckle (which I am not sure they will), but if they don't they still need to sell cards (and the performance associated with them). Using your video card and mainboard businesses synergistically is one thing (Nvidia has done this well), but I really don't think ATI can afford to let both go down hill just because nobody is using their mainboards (if that is the case).

I certainly see no roses in ATI's garden regarding this aspect, but your point was that if CrossFire mainboards fail the whole shebang goes with it. I am saying they could exploit the SLI user base if they wanted to. They are just hungry to force customers into buying their chipsets, but if push comes to shove they will very probably be able to support CrossFire on SLI mainboards and might have to. They may have to relegate their mainboard business to non-CrossFire where it might still make sense, but this wouldn't necessarily imply quitting CrossFire altogether.

But, who knows, maybe CrossFire mainboards will be all the rage and people will be saying "Nforce - who dat?" soon.