R520 - talk, rumours, speculation; the lot!

ATI says their own GPUs are 50-70% efficient.

As I keep saying (looks like I will have to change my sig) I'm expecting a new scheduler to produce a radical improvement in shader efficiency (separately improving the efficiency of both ALU instructions and TMU instructions).

Jawed

 

Surely it depends on what your game is doing? If it's spending 90 percent of it's time doing shaders, and you can make shaders 30-50 percent more efficient, that's going to have a massive effect on your overall game performance.

Even if Nvidia is 90 percent efficient (which I guess depends on how you measure efficiency), that extra 10 percent could make a significant difference if it impacts what your game is spending a lot of it's time doing. This is especially the case in today's complex GPU hardware where one set of inefficiencies can have a knock-on effect on other parts of the chip waiting on that ineffcient part.

 

If it's based on G70 it's a "refresh" to what exactly? Just because the 7800 carries a different codename, it doesn't mean that it doesn't belong to the NV4x family.

 

It would not only mean double the performance under specific situations, but another triple performance on top of that for R580. So either some people are expecting too much here or the latter is going to have up to 5x times more performance than X800.

 

I thought it was widely accepted round here that G70 is NV47 by another name.

In theory NV4x is running out of digits, so we should expect NV50 soon

Jawed

 

Yep.

 

AMD FX55 @2700Mhz Vs AMD64 Winchester 3200+ @2600Mhz

LEADTEK 7800 GTX Vs Sapphire Radeon X850XT 520/540 VgpuMod @630/621

http://img99.imageshack.us/img99/9370/3dmark0za.jpg

http://img99.imageshack.us/img99/9114/farcry5ds.jpg

http://img99.imageshack.us/img99/6669/doom37gq.jpg

http://img99.imageshack.us/img99/1272/halflife28gv.jpg

http://www.gpumania.com/foro/viewtopic.php?t=3707&postdays=0&postorder=asc&start=30

 

The performance ballpark between the two used CPUs is too big and the resolutions used too low to escape most of the CPU limitations.

 

It is what I have found.

 

I don’t see how a quad based GPU can be 90% efficient. What happens when you have one pixel with a 100 instruction pixel shader running and three pixels without shaders. I would suspect the entire quad is in limbo until the long shader finishes running.

 

No, it'd just start working on a different set of pixels. But working with 3-pixel batches would be horrible for any architecture.

 

Dany - Shader 3.0 and Branching

Here is a wonderful thread from Jan 2004:

http://www.beyond3d.com/forum//showthread.php?t=8645

sireric neatly summarises pretty much everything you need to know here:

http://www.beyond3d.com/forum//showpost.php?p=228697&postcount=19

And the post following contains this rather sensational tidbit:

In a recent leaked pdf, ATi suggested not using dynamic branching before R5XX, does that mean all of the problems you mentioned will be resolved in R5XX? Since I don't see there's any fundamental solution for the fighting between long shaders and limited resources, if possible, I'd like to hear how you guys at ATi find a way around.

What a great thread - isn't hindsight (not complete, yet!) a wonderful thing? Damn, this delve into the mists of time practically deserves it's own thread!

Jawed

 

All the Eric says IMO and I cannot see any other experienced HW engineer disagreeing with it in that thread is that dynamic branching isn't an absolute eulogy or absolute panacea on SIMD processors.

Let me fetch some help:

Those that have read the ShaderX2 books know where this one comes from.

 

How's one part of a quad pipe supposed to work on something else entirely while this single pixel gets shaded properly? Maybe you mixed quads with GPUs.

 

From a TD Newcrest report...

The problem:

So, something that was common to RV530 and R520, but no RV515 and Xenos caused the issue. I once said it related to the memory bus, I think may have been correct...

Ongoing...

So, along with R580 we now have RV560, RV540 and RV505 on the roadmap - start speculating the configs of those!

 

on nooooo....please dont.... this topic will never die.
are you trying to make this longest topic in history of B3D?

i am lurking in here, cause i really have nothing to add (technically) and all i get is being more and more confused every time i read it... and this thing is growing like crazy.
what was biggest thread ever on B3D?


EDIT :: i noticed "R520 infomania" with 1700+.... thnx god that one is locked

 

A refresh with 4 Gpu (3 Value)? Isn't that too much?

 

Hmm, Xenos is 128-bit and RV515 and RV530 are rumoured to be 128-bit memory interfaces - so that aint it then.

Jawed

 

a) Preparing for a new personal hate-wave from fellow users for the teasers? LOL

b) If you'd tell me whether ROP amount will remain idle in the mainstream segment it would help.

c) Dat report speculates about "32 pipelines" ROFL.

d) Last time I speculated for WGF2.0 GPU releases for late 2006/early 2007, you gave me the impression that they might appear earlier than that. Was I really barking that much at the wrong tree?

 

I'm gonna guess that at least one of the new RVs is an old core recicled, again. RV505 = R200 derivative with like 1.75 extreme pipes, twelve TMUs1 and a 32-bit memory interface. Or something.