AYBABTU - A Rage3D Interview with Eric Demers (Discussion Thread)

[Lupine]

Rage3D sits down with AMD's Eric Demers to discuss Radeon technologies and reveal how one earns the right to say AYBABTU!

'Under the IHS' Interview: Click Here

[Sound_Card]

Well done Alex. My exact thoughts on Open CL and Cuda.

[sireric]

Yes, well done Alex.

[Gerard]

*scratches head*

Btw eric you really need to get a new pr pic, that ones from around 2003 or so isnt it?

[sireric]

Originally Posted by Gerard

*scratches head*

Btw eric you really need to get a new pr pic, that ones from around 2003 or so isnt it?

Yeah, I know (I'm better looking now ;-)). I took one with my phone and sent it to Alex as well -- when I saw it on the page (preview), it scarred me. I'm supposed to get some sort of head shot to replace the current one in a couple of weeks. I just don't take good pictures. Some sort of evil curse.

[Lupine]

Originally Posted by sireric

I took one with my phone and sent it to Alex as well -- when I saw it on the page (preview), it scarred me.

It scared me too, so we pulled it with a Quake-style headshot and stuck w/ your 2003 hottie shot.

[AlexV]

Originally Posted by sireric

Yeah, I know (I'm better looking now ;-)). I took one with my phone and sent it to Alex as well -- when I saw it on the page (preview), it scarred me. I'm supposed to get some sort of head shot to replace the current one in a couple of weeks. I just don't take good pictures. Some sort of evil curse.

Actually, it's because you used an iPhone...

[sireric]

Originally Posted by AlexV

Actually, it's because you used an iPhone...

Your ninja-photoshop skills weren't up to snuff perhaps?

[SIrPauly]

Great job and outstanding questions -- enjoyed the inter-connect and Super-AA questions!

[caveman-jim]

Did I miss where 'AYBABTU' was explained?

Also, needs moar facepalm.

And did Eric really say trademark when repeating the word 'Radeon'?

[Lupine]

Originally Posted by caveman-jim

Did I miss where 'AYBABTU' was explained?

Also, needs moar facepalm.

And did Eric really say trademark when repeating the word 'Radeon'?

You just earned moar facepalm.

[AlexV]

Originally Posted by sireric

Your ninja-photoshop skills weren't up to snuff perhaps?

Oh, those....never had em

[caveman-jim]

Originally Posted by Lupine

You just earned moar facepalm.

I get plenty helping you learn how to use computers.

[sireric]

Originally Posted by caveman-jim

Did I miss where 'AYBABTU' was explained?

Also, needs moar facepalm.

And did Eric really say trademark when repeating the word 'Radeon'?

Sorry, it's a nervous (tm) tick I have (tm)...

[caveman-jim]

, its ok - if thats the worst you've gotten from explaining the difference between VLIW and superscalar, then at least you've got a career in PR ahead of you!

[Xion X2]

Nice interview.

I could be wrong in my thinking, but this comment here:

Originally Posted by sireric

The interconnect offers the same features as the PCIe interconnect, plus it allows for a GPU to broadcast memory writes to its own memory and to the other GPU's. The data exchangeable is any data that the driver would desire.

--seemed to suggest that the sideport would allow GPU1/GPU2 to share the frame buffer?

Perhaps I'm just hoping that's what was meant.

[sireric]

Originally Posted by Xion X2

Nice interview.

I could be wrong in my thinking, but this comment here:



--seemed to suggest that the sideport would allow GPU1/GPU2 to share the frame buffer?

Perhaps I'm just hoping that's what was meant.

I don't exactly know what you mean by "sharing", but it's just that when one writes, it can write both to its own local memory, as well as the remote, at the same time.

[Xion X2]

Thanks for the quick reply. I'll see if I can elaborate a little more, although my knowledge in this area's limited.

With the sideport enabled, would both GPUs have the capability to share the 2GB VRAM for rendering purposes (high-res textures and multisampling AA?)

My understanding is that 4870X2 was designed with 2GB VRAM because the GPUs couldn't share a 1GB, therefore effectively splitting it in half (512 for each GPU.) So by equipping the card w/ 2GB of VRAM, each GPU would have access to 1GB.

Does the sideport resolve that issue so that both GPUs have access to share the 2GB for rendering? I'm guessing not and that I'm misunderstanding the exact function of it.

[sireric]

Originally Posted by Xion X2

Thanks for the quick reply. I'll see if I can elaborate a little more, although my knowledge in this area's limited.

With the sideport enabled, would both GPUs have the capability to share the 2GB VRAM for rendering purposes (high-res textures and multisampling AA?)

My understanding is that 4870X2 was designed with 2GB VRAM because the GPUs couldn't share a 1GB, therefore effectively splitting it in half (512 for each GPU.) So by equipping the card w/ 2GB of VRAM, each GPU would have access to 1GB.

Does the sideport resolve that issue so that both GPUs have access to share the 2GB for rendering? I'm guessing not and that I'm misunderstanding the exact function of it.

Well, they can already split certain things already -- For example, the display buffer can be 1/2 in one, and 1/2 in the other. The parts unique to each are kept only in their own memory, without duplication. In a way, that's "sharing" as you envision it, I believe.

But to "truly" share memory, each ASIC would need full read and write access to the "other" memory, as well as having a significant amount of the bandwidth made available (say equal amounts remote as local). If they only had read/write, but much less bw, then it would be a numa type system, and it's not clear how useful that would be, performance-wise.

In our case with sideport, it's a write only memory, which much less bandwidth than local GDDR memory. Consequently, some things can be accelerated (like generating the same render target on both, using the broadcast write), but, in general, you need to have copies of everything needed in each memory -- You need to be able to read it, and have enough BW reading it to not slow you down too much.

Consequently, for now, AFR still ends up being the best way to get both rendering and geometry acceleration. The added BW and write capabilities of the sideport won't generally be used in AFR, since each is working on their own thing. But we are looking at non-AFR ways in the future, to accelerate rendering in multi-gpu, without duplicating the data.

[Lux]

Many thanks for the interview etc - it is quite remarkable that somebody in your position takes so much time over the years to communicate with fanbase. Thank you for that!

My question would be - any info on if/when ATI plans to release new amazing Ruby demo, to showcase the really amazing hardware 4000-series is?

[GanjaStar]

Rage3D: Hello Eric, it's a pleasure to finally meet you!

Eric: Well, we are only meeting on the phone…

i knew it was gonna be an awesome article right at the start

[sireric]

Originally Posted by Lux

Many thanks for the interview etc - it is quite remarkable that somebody in your position takes so much time over the years to communicate with fanbase. Thank you for that!

My question would be - any info on if/when ATI plans to release new amazing Ruby demo, to showcase the really amazing hardware 4000-series is?

Not 100% sure on demo status -- The froblins was released, I believe. As for some of the other stuff, like otoy related ones, I'm not sure if those will be released or not.

[Xion X2]

Originally Posted by Lux

Many thanks for the interview etc - it is quite remarkable that somebody in your position takes so much time over the years to communicate with fanbase. Thank you for that!

Indeed we do appreciate it, Eric. Thanks for your last clarification.

My thanks to the driver team as well as they're doing some excellent work thus far. Please let them know it's appreciated by us.. especially those of us with high-end setups (Crossfire.) I've seen some significant performance increases in these last two driver sets in DX10 games like Crysis and Lost Planet. Crossfire scaling in itself has been incredible in most games I've tested. Aside from performance, each driver has been stable, with very few bugs, if any.

In regards to AFR, I'm wondering if ATI is keeping tabs on Lucid's new multi-GPU tech, the Hydra. Not sure if you can speak to that or not, but it sounds as if it has a lot of potential.

[sireric]

Originally Posted by Xion X2

Indeed we do appreciate it, Eric. Thanks for your last clarification.

My thanks to the driver team as well as they're doing some excellent work thus far. Please let them know it's appreciated by us.. especially those of us with high-end setups (Crossfire.) I've seen some significant performance increases in these last two driver sets in DX10 games like Crysis and Lost Planet. Crossfire scaling in itself has been incredible in most games I've tested.

Aside from performance, each driver has been stable, with very few bugs, if any.

Those guys do read the forums and they send their thanks, I'm sure. They are a superb bunch of people and do work very hard. It's somewhat of a thankless job (people only complain), and very tricky and demanding. They've made huge improvements on stability and performance in the last year, and there's much more to come...

[Xion X2]

Originally Posted by sireric

It's somewhat of a thankless job (people only complain), and very tricky and demanding.

Yeah, I can imagine that. It's a good thing they have you as a conduit to communicate at least a little appreciation from some of us.

Their effort hasn't gone unnoticed, and for those of us who pour a significant amount of money and time/effort into this stuff, they are our best friend(s). I always look forward to their next piece of work.

[sireric]

Originally Posted by Xion X2

In regards to AFR, I'm wondering if ATI is keeping tabs on Lucid's new multi-GPU tech, the Hydra. Not sure if you can speak to that or not, but it sounds as if it has a lot of potential.

Yes, we do keep tabs on that kind of technology. There are different possibilities for scaling -- We are investigating some of those for the future, of course.

[Kombatant]

Eric, long time no talk.. and yes btw, he is much better looking these days than the 2003 era

[AlexV]

Originally Posted by sireric

But to "truly" share memory, each ASIC would need full read and write access to the "other" memory, as well as having a significant amount of the bandwidth made available (say equal amounts remote as local). If they only had read/write, but much less bw, then it would be a numa type system, and it's not clear how useful that would be, performance-wise.

Ah, merde, I just figured out a question I could've/should've asked and didn't. Oh, well...next time.

[jam2k]

Great interview, very interesting stuff.

[caveman-jim]

Originally Posted by AlexV

Ah, merde, I just figured out a question I could've/should've asked and didn't. Oh, well...next time.

Originally Posted by Lupine

You just earned moar facepalm.

All your facepalm belong to alex!