R350 first week of March?

[Natoma]

Taking a quick break from work. My friend wanted me to post this for him. Anyways, here's the link:

http://www.theinquirer.net/?article=7381

It says R350 by the first week of March, which from what I've read puts it in the same timeframe as the NV30. They also have a few specs, but they're probably just the rumors that have been floating around.

9800 or 9900
400-425Mhz Core/800Mhz DDR
6GB/sec more memory bandwidth than the R300
128MB default configuration with option for 256MB
Beats NV30 in 8 out of 10 Benchmarks (which ones??)

 

[Joe DeFuria]

Well, that's pretty much what most people are expecting in terms of specs. So no real surprise there.

I have also been expecting a "launch" of the R350 product in late Feb / Early March, with mass availability in April.

I guess we'll just have to wait and see.

 

[Doomtrooper]

R350 should stack up very well....common sense tells what benchmarks it would be superior in with 6+ GB more memory bandwidth.

 

[Sxotty]

Actually it just says it is going to "be introduced" like the FX was in November, well gee whiz that is exciting.

 

[antlers]

R300 went from introduction to availability in 30 days; I imagine R350 will be able to do the same (since it taped out in November)--in fact, I wouldn't be surprised if they were delaying introduction to just before availability so they don't step on R300 sales and have the maximum impact on NV30 sales (kind of like NVidia did when they were introducing the GeForce 2 against the Voodoo 5)

 

[megadrive0088]

wait, R350 should have at least 25 GB of raw bandwidth, that would be 6 GB more than R300, and 9 GB more than GFFX.

 

[Tahir2]

Beats NV30 in 8 out of 10 Benchmarks (which ones??)

Not the Doom 3, Not the Doom 3!

 

[g__day]

This actually seems plausible reasoning - sigh - I just wish it was comming from a much better source than the Inquirer...

 

[MuFu]

It's not exactly groundbreaking news though. :-\

MuFu.

 

[Grall]

Is that a verification they're correct, MuFu?

If R350 launches at 400MHz I have to say I'll be somewhat disappointed. R300 stunned me (and most everybody else it seems) when it hit 325 easy, though if it launches at 425 with a monster ATI-flow cooler that eats the adjacent slot, I'll definitely be disappointed! Eugh!

*G*

 

[Doomtrooper]

Most of the information can be taken from this forum

 

[Doomtrooper]

If it launches at 425 mhz I don't think that is something to be dissapointed about thats a 100 mhz improvement ??
I would be more concerned about memory clock speeds..oh and of course cooling more than likely should be different

 

[KimB]

Beats NV30 in 8 out of 10 Benchmarks (which ones??)

Not the Doom 3, Not the Doom 3!

DOOM3 is perhaps the best-case scenario for the GeForce FX. For example, unless JC has found a way to clear the stencil buffer at the same time as the z-buffer for his algorithm (which I find rather unlikely...but I suppose it is possible), then Hierarchical-Z will be disabled in DOOM3, significantly hurting the Radeon 9700's advantage.

That and DOOM3's rendering will likely be more fillrate-bound than memory bandwidth-bound.

 

[Doomtrooper]

Only for the people that enjoy paying $400 + for video cards with nothing on, like I said the Alpha runs fine with AA on a 9500 NP...no need not to run some FSAA and AF on Doom 3...Alpha proves it.

 

[WaltC]

I don't have a lot of faith in this Fuad character...I mean, if it is true that ATI is waiting on production-grade silicon, whence cometh the benchmarks? Obviously, if they don't even know the clockspeed the yields are going to give them there's no way to even estimate a benchmark. And how is it that someone has a production-level GF FX to test against?

Sorry...but me no buy story. As to the alleged R350 specs, this is no more than has already been speculated many times before.

To provide an example of the sort of level Fuad writes on, I think it was yesterday that he wrote a piece in the Inquirer entitled: "AMD's Hammer Loses MHz Race with P4", or something very similar--as if AMD had designed Hammer around the same sort of MHz scheme Intel is using for the P4. He also wrote a story about ATI back in October in which he claimed that ATI "admitted" its 9700Ps "had a problem" with AGP x8 and that "ATI engineers were slaving over a solution" to this problem, and he used a link that referenced a help file document on ATI's site as his corroboration for the story. Problem was that the link Fuad used on the ATI site said nothing of the kind--what it said was that there was nothing wrong with the 9700P's AGP x8 implementation, but that some motherboard makers hadn't yet implemented AGP x8 correctly. I'll never forget that story--he flat out lied about his subject material in that case. And I ought to know as I'm using a 9700P (I bought in September) in an nf2 motherboard at AGP x8 without the slightest problem.

 

[Crusher]

For example, unless JC has found a way to clear the stencil buffer at the same time as the z-buffer for his algorithm (which I find rather unlikely...but I suppose it is possible), then Hierarchical-Z will be disabled in DOOM3, significantly hurting the Radeon 9700's advantage.

I don't know much about OpenGL programming, but in Direct3D, you can clear the stencil and z buffers in the same function call. I assume OpenGL has the same capability. If you're talking about the volumetric shadows though, the stencil operations are performed last, after everything is already rendered to the Z buffer. The Z buffer isn't changed at all, and I don't see how the stencil operations would affect any Z buffer optimizations of the hardware.

 

[nAo]

The Z buffer isn't changed at all, and I don't see how the stencil operations would affect any Z buffer optimizations of the hardware.

Unfurtunately stencil operation can affect z-buffer optimization even z-writes are switched off.
I know the at least one implementation of zbuffer compression encode the stencil buffer value for a pixel within the compressed z-packet.
In fact the z-compression compress z and stencil values in the same packet. In this particular implementation the stencil values for the group of pixel compressed in the same packet has to be constant, but in other implementations they can be compressed with a DDPCM scheme like the zbuffer values.
Rendering on the stencil buffer can destroy the conditions that make possible to compress (with a fixed compression ratio) a particular group of pixels. When this happens the hw should decompress the packet and write it to memory uncompressed.

ciao,
Marco

 

[Dave Baumann]

DOOM3 is perhaps the best-case scenario for the GeForce FX. For example, unless JC has found a way to clear the stencil buffer at the same time as the z-buffer for his algorithm (which I find rather unlikely...but I suppose it is possible), then Hierarchical-Z will be disabled in DOOM3, significantly hurting the Radeon 9700's advantage

The docs aren't quite exact. They mean that Hierarchical Z is of no use in some cases. Basically, when Z test is set to "equal" or "not equal" or when pure stencil test is done (as I said, there's no stencil Hierarchical Z), then HZ doesn't help.

Not, that HZ is disabled, it just doesn't assist the Stencil operation.

 

[demalion]

"HZ" needs to die as an abbreviation.

Always causes confusion between Hierarchical Z and Hyper Z.

 

[Natoma]

R350 should stack up very well....common sense tells what benchmarks it would be superior in with 6+ GB more memory bandwidth.

Not necessarily so. According to the article, the R350 loses to the NV30 in Quake 3. But hasn't that benchmark shown that it is far more dependent on memory bandwidth than anything else?

I would think that that benchmark would show the extreme benefits of the extra 6GB of bandwidth, but apparently it doesn't. So I dunno...