NGGP: NextGen Garbage Pile (aka: No one reads the topics or stays on topic) *spawn*

Status
Not open for further replies.
Sea Islands, aka HD8000.

Both Orbis and Durango will be 28nm AMD, so I guess there won't be a huge difference between the architectures. It just doesn't make sense.
 
What's GCN2?

Dave knows. I only can supposse is a GCN architecture evolution in order to improve efficiency ( somebody said improve wavefronts process management ) and add context switching to make feasible fast change between processing graphics and GPGPU mode to avoid ALU cycles stalling.

As of today Orbis seems very off-the-shelf with little custom changes. IMHO Modifying 4 CUs adding an ALU to make them more branching code capable is a cheap way to have better GPGPU capabilities and get your SPUs simulator. The best and more differentiator component in Sony´s console is by far the 4GB of GDRR5, but we still have no clear whether both the ESRAM and DDR3 in Durango have ROPs that would close that gap and even be more interesting for Durango if ESRAM is low latency as in a HSA architecture would allow crazy GPGPU effects. So, let´s wait for what Durango brings to the table to make better comparisons.
 
Last edited by a moderator:
Context switching comes in the GPU iteration after that ("2014" in AMD roadmaps, and really it's a post-Steamroller APU)

What you get in the GPU attached to Jaguar (and the vaporware Steamroller) is single address space, which will probably improve GPGPU a ton on its own.

"GCN2" is a made up shorthand. Without being demeaning, I would call what's in Radeon 8000 and Jaguar APUs GCN 1.1, officially it feels like it's just "GCN".
So, the speculation is Durango has that 2014 GPU. I don't risk much by saying it doesn't.
 
I only can supposse is a GCN architecture evolution in order to (...) add context switching to make feasible fast change between processing graphics and GPGPU mode to avoid ALU cycles stalling.

Graphics Pre-Emption (GPU context switch) is a GCN3 Feature, 20nm Volcanic Islands, or HD9000 if you want. This is totally out of reach for Microsoft.

It would be superior to a single GPU that has to deal with both GPGPU algorithms and graphics rendering, but it would still be inferior to a APU dedicated to GPGPU.

According to the rumors, Durango is still 50% less powerful than Orbis. GCN2 won't be able to compensate for that.
 
Context switching comes in the GPU iteration after that ("2014" in AMD roadmaps, and really it's a post-Steamroller APU)

What you get in the GPU attached to Jaguar (and the vaporware Steamroller) is single address space, which will probably improve GPGPU a ton on its own.

"GCN2" is a made up shorthand. Without being demeaning, I would call what's in Radeon 8000 and Jaguar APUs GCN 1.1, officially it feels like it's just "GCN".

Xenos packed in 2005 tech from a chip launched in 2007 ( R600 ), unified shaders and decoupled texture units -well, this was from the upcoming R520- and ROPs. So, would be feasible MS added post-Steamroller projected tech.
 
According to the rumors, Durango is still 50% less powerful than Orbis. GCN2 won't be able to compensate for that.

According to rumors they're on same performance boat, same cpu, one gpu is 14+4 CU's, the other is 12+ special sauce/unknow CU's
and we're repeating that flops can't be used to do performance comparisons, there're a lot of factors than flops
 
Xenos packed in 2005 tech from a chip launched in 2007 ( R600 ), unified shaders and decoupled texture units -well, this was from the upcoming R520- and ROPs. So, would be feasible MS added post-Steamroller projected tech.

They could, no problem, but remember Xenos caused all manner of problems with regards to the consoles durability. I personally think we will have two incredibly similar machines with "similar" specifications. The only differentiation I can see at the moment (Durango information is limited) is the memory pool setups.
 
Xenos packed in 2005 tech from a chip launched in 2007 ( R600 ), unified shaders and decoupled texture units -well, this was from the upcoming R520- and ROPs. So, would be feasible MS added post-Steamroller projected tech.

Actually it's more like Xenos packed ~2004/2005 technology from a customised version of a shelved processor. If Durango gets something between GCN 1 & 2 with some customisation they probably won't be any further behind than the 360 was.
 
They could, no problem, but remember Xenos caused all manner of problems with regards to the consoles durability.

That wasn't to do with the architecture of the GPU, it was down to the lack of lead solder on a system designed to use it, and what appears to be poorly managed thermals.
 
That wasn't to do with the architecture of the GPU, it was down to the lack of lead solder on a system designed to use it, and what appears to be poorly managed thermals.

I admit it wasn't to do with the architecture of Xenos but was to do with the heat it generated.
 
It would be superior to a single GPU that has to deal with both GPGPU algorithms and graphics rendering, but it would still be inferior to a APU dedicated to GPGPU.

I don't think that, it would be useful when only doing GPGPU, with low cost and ease of running multiple tasks, it would be useful when gaming (with "GPGPU used for graphics") and in general, would give a mindset not considering GPGPU or graphics as "special things" anymore.

The ideal situation is to have a big ass APU. APU + GPU allows to get more flops (and is more realistic in the PC realm) but the low bandwith and high latency between them is a pain. So, it's superior in a way ; but the big APU with all features is so much more elegant.

Maybe APU + dedicated GPU is a dead end. Why not grow the other way : an APU + CPU :).
This is where you go with Maxwell, and I wonder if AMD will follow that route, eventually selling us "APU on a graphics card" the way nvidia will do. Of course, you might have an APU + APU if you buy all AMD hardware.
 
Actually it's more like Xenos packed ~2004/2005 technology from a customised version of a shelved processor. If Durango gets something between GCN 1 & 2 with some customisation they probably won't be any further behind than the 360 was.

Funny thing is that R400 architecture was way better than R600 one.
 
Xenos packed in 2005 tech from a chip launched in 2007 ( R600 ), unified shaders and decoupled texture units -well, this was from the upcoming R520- and ROPs. So, would be feasible MS added post-Steamroller projected tech.

No.. the Xenos packed R400 tech, which had been developed already but deemed not suitable for the PC market yet (or would have been too late)
They used a cancelled GPU, not a future one.
See this from.. september 2003
http://www.neoseeker.com/news/2771-ati-r400-gpu-cancelled/
 
The ideal situation is to have a big ass APU. APU + GPU allows to get more flops (and is more realistic in the PC realm) but the low bandwith and high latency between them is a pain. So, it's superior in a way ; but the big APU with all features is so much more elegant.

I was talking about the GPGPU capabilities in games. A single GPU will have a very hard time dealing with graphics rendering and GPGPU algorithms at the same time. The GPU will saturate which kills your latency and your performance. A lot of code optimization is required to get either task running on the same GPU.

An APU that only deals with GPGPU algorithms and a GPU that only deals with graphics rendering will deliver a much better performance. In the case of Orbis we have the same thing (if the rumors are correct), but integrated into a SoC: The APU consists of 8 Jaguars and 4 GCN CUs, the GPU consists of 14 GCN CUs.
 
He just said that on "paper" orbis looked better but that was on paper. He reckoned if the game was physic, animation, framerate heavy, orbis would have the advantage, which seems like a sweeping generalisation. He said nothing about the esram or bandwidths but say that Durango may have upper hand with regards to triangle/vertices drawing power.
To be honest I don't think he knows too much about durangos setup.

That sounds reasonable, but why would he think it would have the upper hand with regards to geometry processing? The data move engines and ESRAM wouldn't really help with that, would it, certainly not while Orbis is packing more CUs.

Unless there are as yet unrevealed customisations to Durango's GPU that give it a vertex/tri advantage.
 
No.. the Xenos packed R400 tech, which had been developed already but deemed not suitable for the PC market yet (or would have been too late)
They used a cancelled GPU, not a future one.
See this from.. september 2003
http://www.neoseeker.com/news/2771-ati-r400-gpu-cancelled/

Yes. You are right, but as R400 was in design form, and was cancelled as Directx 10 was late, AMD could have other designs waiting for proper market time. Engineers didn´t stop designing once GCN was in the street.
 
but integrated into a SoC: The APU consists of 8 Jaguars and 4 GCN CUs, the GPU consists of 14 GCN CUs.

Alright, I didn't realize this!, or didn't read it through the noise.
I had no idea why you talked about APU+GPU. This makes better sense.

What could be argued : is a 18 CUs APU with context switch better than 4 CUs + 14 CUs without context switch. I'll feel like saying yes. But that would be a theoretical argument with no relevance to what's in Durango or Orbis.

So, 12 CUs in Durango vs 4 + 14 in Orbis.
 
Status
Not open for further replies.
Back
Top