AMD: R7xx Speculation

Status
Not open for further replies.
Yes, no mention of it all. And wasn't 4850's memory supposed to be running at 1100-something MHz? It seems they've been unable to match that. It only runs at 980
 
PC Parter starts mass production of Radeon HD 4850 2008-6-3 19:55:16PC Parter starts mass production of Radeon HD 4850

We are told that PC Partner, the VIP in AIBs, now starts mass production of Radeon HD 4850 GDDR3 Graphics Cards for AMD and ATI. These Graphics Cards are prepared for the RV770 launch in June 15.

As we know, AMD can get enough bulks of Radeon HD 4850 GDDR3 Graphics Cards for the hard launch at June 15 world widely.

We already know, Radeon HD 4850 GDDR3 graphics, with AMD's own Spider quad-core platform and beta version of Catalyst driver, it got P64xx in
3DMark Vantage test.

source
http://www.pczilla.net/en/post/30.htm
l
 
Yes, no mention of it all. And wasn't 4850's memory supposed to be running at 1100-something MHz? It seems they've been unable to match that. It only runs at 980

No need to use 2.2G RAM and risk TDP for a paltry 120Mhz here.

And obviously, partners can OC the rest when non-reference boards come.
 
Wish they had reported the graphics score instead of the overall. My overclocked GTX does 6019 for graphics. Total P score is P6979 with a Q6600 G-0 @ 3.8GHz.
 
Not sure but I saw this:

http://en.expreview.com/2008/04/26/800gtx-scores-5851-in-3dmark-vantage/

9800GTX scored P5851 with GPU score 5636

If the CPU was relatively similar, then thats an impressive P score for the 4850

Depending on how well Vantage scales with that 12% core overclock the stock 4850 may come in closer to the 9800GTX. But that level of performance for $200 is more than welcome.

The way it's shaping up Nvidia may find themselves in a bit of a squeeze. HD4850/4870 could own the $200 - $400 segment and 4870X2 could take the crown at $450 leaving GTX2xx in no-man's land.
 
NV Might be forced to drop the price of the lower end GT2XX part and suffer the hit to their margins until then can get out a part that better targets that price range.
 
NV Might be forced to drop the price of the lower end GT2XX part and suffer the hit to their margins until then can get out a part that better targets that price range.

I wouldn't be sure of that.

Just like camera tricks made Kelly Clarkson look slim, the very right combination of PR can make the 260 sell like hot cakes. And nVidia has always been a power PR player- look at how stoked the people are on the Tegra.
 
I feel stupid saying this (as normal, when using 3DMark as the basis of anything), but according to the Vantage P scores, HD4850 is ~50% faster than HD3850, per core clock. P scores reported so far are a bit vague (including the CPU, sigh) and they're with AF/AA turned off, too.

But, ahem, I'm thinking that along with 480 ALU lanes we're looking at 24 TUs, i.e. 4x SIMDs that are 24 elements wide and 6 quad TUs.

Jawed
 
I feel stupid saying this (as normal, when using 3DMark as the basis of anything), but according to the Vantage P scores, HD4850 is ~50% faster than HD3850, per core clock. P scores reported so far are a bit vague (including the CPU, sigh) and they're with AF/AA turned off, too.

But, ahem, I'm thinking that along with 480 ALU lanes we're looking at 24 TUs, i.e. 4x SIMDs that are 24 elements wide and 6 quad TUs.

Jawed

If RV770 is "only" 24 TMUs then ATi won't have changed their ALU:TEX ratio at all and should not even have bothered. Ok, the increase in texturing ability is welcome no matter what, but they have to see the disparity between texturing ability and math power.
 
Since RV770 is supposed to have a lower efficiency than RV670, I would think 3 SIMDs and 8 TUs should also be possible.
 
I feel stupid saying this (as normal, when using 3DMark as the basis of anything), but according to the Vantage P scores, HD4850 is ~50% faster than HD3850, per core clock. P scores reported so far are a bit vague (including the CPU, sigh) and they're with AF/AA turned off, too.

But, ahem, I'm thinking that along with 480 ALU lanes we're looking at 24 TUs, i.e. 4x SIMDs that are 24 elements wide and 6 quad TUs.

Jawed
Might be true, if 3DMVa is - to some extent- texture bound. But somehow I have a feeling, that this is not the whole truth, we're taking a glimpse at here.
 
I imagine the GPU tests in 3dmark Vantage are mostly math-limited, and the 50% increase in SP count corresponds nicely here, I think.
 
I imagine the GPU tests in 3dmark Vantage are mostly math-limited, and the 50% increase in SP count corresponds nicely here, I think.
The P tests are prolly texture-bound in preference to ALU-bound. ALU-bound shaders being reserved for the X tests.

Jawed
 
The P tests are prolly texture-bound in preference to ALU-bound. ALU-bound shaders being reserved for the X tests.

Jawed

Now if only the extreme setting actually looked better than the performance setting. But that's probably too much to ask :devilish:
 
Status
Not open for further replies.
Back
Top