AMD: R7xx Speculation

[mao5]

R700 lost battle to GTX 280?
We learn from pcinlife forum that ATI next generation flagship R700 lost battle to GTX 280 in closed doors tests.

Taiwan graphics cards manufacture obviously got both 4870x2 and GTX 280 graphics card samples for these comparisons. But as we told you before, we still don't know the specification of the test hardware and software configurations.

We hope ATI Catalyst Develop Team could do miracle to save Radeon HD 4870x2 in next gorgeous chips battle between AMD and NVIDIA.

http://www.pczilla.net/en/post/12.html

 

[Lukfi]

Oh, such a tragedy [/sarcasm]

 

[ShaidarHaran]

R700 lost battle to GTX 280?
We learn from pcinlife forum that ATI next generation flagship R700 lost battle to GTX 280 in closed doors tests.

Taiwan graphics cards manufacture obviously got both 4870x2 and GTX 280 graphics card samples for these comparisons. But as we told you before, we still don't know the specification of the test hardware and software configurations.

We hope ATI Catalyst Develop Team could do miracle to save Radeon HD 4870x2 in next gorgeous chips battle between AMD and NVIDIA.

http://www.pczilla.net/en/post/12.html

Rumors of 4870x2 delays for driver work abound. I think this is far too early to be even remotely relevant.

 

[Jawed]

I fail to see what's the problem really with increasing the TU batch size?

I'm saying that TU and ALU batch size is the same and also saying that I'm doubtful batch size will increase beyond 64.

Yes, but AFAIK point sampling in pixel shaders (or fetch4) is quite rare compared to bilinear texturing, so it still doesn't make much sense to me.

I can't offer much here I'm afraid. Fetch4 has been relegated to D3D10.1 so it'll only turn up as a tweak for improved shadow filtering performance or faster deferred rendering MSAA resolve, that kind of thing.

I think adding "only" more TF units would make sense - maybe there could be some restrictions where the results would need to go (either all 8 filtered texels from a "quad-unit" to the same shader array, or 4+4 must go to a different array or something like that) but it should be doable IMHO without screwing up the basic r6xx design principles too much.

TF in R600 is baseline fp16 format (int8 textures are converted to fp16 for filtering) - so the filtering units are big. So you're suggesting a doubling of an already very big unit.

And I still don't know how the TU architecture would support this. I suspect it's tied down. The ATI guys have talked in the past about being able to scale lots of components individually (e.g. width of ALUs, number of RBEs) but there's not been any hint of varying TA:TF ratio.

If I happen to trip over a relevant patent document you'll be sure to hear about it...

Jawed

 

[w0mbat]

http://www.nordichardware.com/forum/-vp25902.html#25902

 

[ZerazaX]

From that link:

asic name: rv770
revision: a12
board no.: 102-B50102-00
engine clk: 625mhz
memory manufacturer: qimonda
memory type: gddr3
memory config: 16Mx32
memory no.: HYB18H512321BF-10
memory clk: 990mhz
vddc (core voltage): 1.13v
mvddc/q (memory voltage): 2.03v
tdp: 114w

performance rv670xt +80%

+80% over 3870 would be amazing

 

[ShaidarHaran]

From that link:


+80% over 3870 would be amazing

I noted that as well. Doesn't jive with the most-recently-rumored clocks however.

 

[ShaidarHaran]

Hmm, Arun seems to have removed his post. Did you get a note from the AMD fairy wagging his finger?

 

[trinibwoy]

If AMD was able to increase the ALU count by 150% for a 30% increase in die size I will be very impressed. However, if it's not accompnied by a corresponding increase in texturing power then we really have to wonder if they have some maniacal fetish with unbalanced texture bound designs.

 

[trinibwoy]

Hmm, Arun seems to have removed his post. Did you get a note from the AMD fairy wagging his finger?

Or it could be that his math was slightly wrong

800/320 * 625/775 ~= 2.0

 

[Arun]

Or it could be that his math was slightly wrong
800/320 * 625/775 ~= 2.0

Ding ding ding, I just removed it because I realized it was incredibly embarassing

 

[ShaidarHaran]

If AMD was able to increase the ALU count by 150% for a 30% increase in die size I will be very impressed. However, if it's not accompnied by a corresponding increase in texturing power then we really have to wonder if they have some maniacal fetish with unbalanced texture bound designs.

Indeed. At this point an increase in shader compute power should be the lowest item on ATi's list of issues to address, at least with the multiple fillrate/texturing bottlenecks currently in place.

Or it could be that his math was slightly wrong

800/320 * 625/775 ~= 2.0

LOL, works for me.

 

[ShaidarHaran]

Ding ding ding, I just removed it because I realized it was incredibly embarassing

I used to get in trouble for doing all my math in my head and not writing out any steps or using a calculator. I just use a calculator now

 

[AnarchX]

Seems time is here to make your own AMD documents:

http://www.hardware-infos.com/news.php?news=2075

 

[Jawed]

If AMD was able to increase the ALU count by 150% for a 30% increase in die size I will be very impressed. However, if it's not accompnied by a corresponding increase in texturing power then we really have to wonder if they have some maniacal fetish with unbalanced texture bound designs.

10:1 ALU:TEX would be silly at this juncture

Jawed

 

[Arun]

I used to get in trouble for doing all my math in my head and not writing out any steps or using a calculator. I just use a calculator now

That's not the best part; I actually used a calculator... But yeah, if it's 800 SPs, I'd also expect 32-40 TMUs. We'll see.

 

[ShaidarHaran]

10:1 ALU:TEX would be silly at this juncture

Jawed

But not unlike ATi at all

 

[ShaidarHaran]

That's not the best part; I actually used a calculator...

LOL, ouch. Fat fingers, eh?

But yeah, if it's 800 SPs, I'd also expect 32-40 TMUs. We'll see.

I'm still holding out hope that there will be 32 TMUs in the 480SP configuration. Let's be honest here, ATi needs it. I hope it's in the die-size budget...

 

[w0mbat]

Seems time is here to make your own AMD documents:

http://www.hardware-infos.com/news.php?news=2075

Seems fake to me, but it could be AMD faked to finde some holes.

 

[Slyne]

Last time I checked, cell phones were at least in-part subsidized by (usually) co-marketing agreements between service providers and phone manufacturers whereby the BOM of the phone has no direct correlation to the price which consumers pay.

The same situation does not exist in the world of graphics cards.

Likewise, the price of a component that is nearing its end of shelf-life is not necessarily related to its hardware, but for different reasons.

In the example I was replying to, both RV670 and GDDR4 are approaching it really fast, and the amount of stock in Powercolor's warehouse may matter more than initial hardware costs.