R360 != 0.13 process ?

[surfhurleydude]

What is surfmark, and where did that R420 number come from?

The R420 and R360 % numbers came from MuFu. Surfmark is a theoretical benchmark I came up with to figure out what FPS range they would all line up to be if they performed with the selected percentages.

 

[Bjorn]

The R360 might "only" be 15-20% faster than the R350, and it will also be a completley viable product.

Depends of course on what the competition does. A refresh of the NV35 (assuming no better FSAA or PS2.0) might of course not scare Ati that much though

 

[Joe DeFuria]

Eh? Wasn't the R350 released in April/May?

Radeon 9700 Pro shipped August 22, 2002. Radeon 9800 Pro shipped April 2, 2003.

That is roughly 7 months. (It's less than 7.5 months).

If the R360 ships Oct 11 or later (which is a decent guess assuming it Launches in September, and ships within 30 days), it will be 6+ months since the R350 shipped.

 

[Joe DeFuria]

The R420 and R360 % numbers came from MuFu.

In what way? (He said the R420 is twice as fast as the R300?)

Surfmark is a theoretical benchmark I came up with to figure out what FPS range they would all line up to be if they performed with the selected percentages.

Thanks, but it wasn't necessary.

How does this impact whether or not R360 would be on 0.15 or 0.13?

This whole discussion is pretty meaningless at this point. No one that I'm aware of, except in this thread, has even speculated that R360 would be 0.13.

 

[surfhurleydude]

Actually it was necessary, as people were talking about the % numbers of the R360.

In what way? (He said the R420 is twice as fast as the R300?)

He said that the R420 was twice as fast as the R300, and the R420 was 30% faster than the R360. I just used my custom benchmark to help fill the numbers in to figure out how much faster the R360 would be over the R350. The R360 is NOT 70% faster than the R300 btw, even though if you just quickly glance at percentages you'd think that.

As far as I can tell, this thread has been leaned more towards the performance of the R360 rather than the process it will be fabbed on. So I think it was a help

 

[Seiko]

Actually it was necessary, as people were talking about the % numbers of the R360.

In what way? (He said the R420 is twice as fast as the R300?)

He said that the R420 was twice as fast as the R300, and the R420 was 30% faster than the R360. I just used my custom benchmark to help fill the numbers in to figure out how much faster the R360 would be over the R350. The R360 is NOT 70% faster than the R300 btw, even though if you just quickly glance at percentages you'd think that.

As far as I can tell, this thread has been leaned more towards the performance of the R360 rather than the process it will be fabbed on. So I think it was a help

I think the bottom line is that the R360 will be 0.15 based, why else would ATI only confirm that the r4xx is 0.13? Working on that assumption its plain to determine that we won't be looking at more than a 10-15% speed increase as their simply isn't too much head room in clock speed available.
As for whether this is a viable product, my personal opinion as mentioned here is 'no' it isn't really how I'd like to see ATI play the game. Launching so many flavours of the now infamous R300/R350 architecture with small incremental speed increases is hardly my idea of good resource management. It didn't help ATI however when Nvidias NV30 was reviewed using somewhat poor IQ and settings as comparisons which no doubt forced ATI to bump the speeds to regain the crown. It was only when the hoo haa finally died down that people actually got to see the 9700 Pro was indeed the superior piece of hardware. I just wonder if the initial reviews of the NV30 clearly denoted that it was a massive let down instead of the wishy washy sitting on the fence ones we saw would have let ATI take the time to move the R350 to the 0.13 process and thus allow a much larger increase to be observed? Actually I doubt they react that quick as the R350 was probably on the boards long before a reivew of teh GF4 was published yet alone the NV30.

Anyway what's done is done but as I mentioned this time around regarding the R350, "Don't expect anything more than a small speed bump and perhaps tweaked hyper z function? 10-15% increase at best! I belive the R360 will be the same all over again

 

[surfhurleydude]

Actually it was necessary, as people were talking about the % numbers of the R360.

In what way? (He said the R420 is twice as fast as the R300?)

He said that the R420 was twice as fast as the R300, and the R420 was 30% faster than the R360. I just used my custom benchmark to help fill the numbers in to figure out how much faster the R360 would be over the R350. The R360 is NOT 70% faster than the R300 btw, even though if you just quickly glance at percentages you'd think that.

As far as I can tell, this thread has been leaned more towards the performance of the R360 rather than the process it will be fabbed on. So I think it was a help

I think the bottom line is that the R360 will be 0.15 based, why else would ATI only confirm that the r4xx is 0.13? Working on that assumption its plain to determine that we won't be looking at more than a 10-15% speed increase as their simply isn't too much head room in clock speed available.
As for whether this is a viable product, my personal opinion as mentioned here is 'no' it isn't really how I'd like to see ATI play the game. Launching so many flavours of the now infamous R300/R350 architecture with small incremental speed increases is hardly my idea of good resource management. It didn't help ATI however when Nvidias NV30 was reviewed using somewhat poor IQ and settings as comparisons which no doubt forced ATI to bump the speeds to regain the crown. It was only when the hoo haa finally died down that people actually got to see the 9700 Pro was indeed the superior piece of hardware. I just wonder if the initial reviews of the NV30 clearly denoted that it was a massive let down instead of the wishy washy sitting on the fence ones we saw would have let ATI take the time to move the R350 to the 0.13 process and thus allow a much larger increase to be observed? Actually I doubt they react that quick as the R350 was probably on the boards long before a reivew of teh GF4 was published yet alone the NV30.

Anyway what's done is done but as I mentioned this time around regarding the R350, "Don't expect anything more than a small speed bump and perhaps tweaked hyper z function? 10-15% increase at best! I belive the R360 will be the same all over again

1 sentence:

Clock speed isn't everything, memory bandwidth matters too.

 

[Joe DeFuria]

I think the bottom line is that the R360 will be 0.15 based,

Agreed...and again, I'm not sure why anyone else would really question that at this point.

Working on that assumption its plain to determine that we won't be looking at more than a 10-15% speed increase as their simply isn't too much head room in clock speed available.

I'd say 15% is a reasonable guess...although it's just that....a guess. I certainly wouldn't rule out any surprises. For example, Even if the core clock is "only" 15% or so to 450 Mhz, they could conceivably increase the ram to 500Mhz+ (40%+ increase)....and that could lead to significant gains in AA situations.

Launching so many flavours of the now infamous R300/R350 architecture with small incremental speed increases is hardly my idea of good resource management.

Tell that to Intel.

 

[Prometheus]

If we can make any asumptions from the r350 release its seems that ATi might have a problem clocking the memory too high(r300=310,r350=340)or maybe they thought they had enough bandwidth.

 

[Seiko]

1 sentence:

Clock speed isn't everything, memory bandwidth matters too.

To be honest I really didnt think that the R350 was memory bandwidth bottlenecked until you get into insane 1600*1200 screens and 4*FSAA?
I think you'll run out of steam on the fillrate front for the likes of Doom and HL2 long before you run out of memory bandwidth?

I guess in the end you can never have too much but for current fillrate and vertex/pixel shader limited situations the small clock speed bump won't help too much?

Tell that to Intel.

Lol, crumbs dont mention that to ATIs marketing department, I dread to think what naming conventions they'd come up with for all those speed permutations.



Anyway, I think we'll find out more on the R360 at the end of this month

 

[surfhurleydude]

To be honest I really didnt think that the R350 was memory bandwidth bottlenecked until you get into insane 1600*1200 screens and 4*FSAA?
I think you'll run out of steam on the fillrate front for the likes of Doom and HL2 long before you run out of memory bandwidth?

I guess in the end you can never have too much but for current fillrate and vertex/pixel shader limited situations the small clock speed bump won't help too much?

In most modern games, the second you enable 2x AA you will be memory bandwidth bottlenecked, because if you weren't, there would be no performance penalty for enabling it. The primary reason why the Radeon 9800 Pro is only a little bit faster than the 9700 Pro is because the memory was increased a whopping 30 MHz. I.E. Crap. It's not really in too much of a fill rate limited situation right now because it's got twice the number of pipes as the competitor's card, the GeForce FX 5900 Ultra.

 

[KimB]

It's not really in too much of a fill rate limited situation right now because it's got twice the number of pipes as the competitor's card, the GeForce FX 5900 Ultra.

That's a hugely-misleading statement. The texture throughput of both cards is identical. In fixed-function, multitextured scenarios, the FX 5900 will not be much slower clock for clock than the 9800, though it has a higher clock speed. In the end, it does depend on the game, but for fixed function, the 5900 should have very similar effective fillrate to the 9800, not half the fillrate as you seem to want to convey.

For pixel shader functionality, the situation gets much more complex, and is not easily-comparable. Each architecture has its strengths and weaknesses, with the 5900's primary weakness being the penalty induced with register usage. How they compare in advanced shaders will depend on games, and we don't have the advanced shader-capable games to examine just yet.

 

[surfhurleydude]

How is it a misleading statement? Radeon 9800 Pro has 8 pipes per TMU, and GeForce FX 5900 Ultra has 4 pipes per 2 TMU. That is a fact. There's nothing misleading about it, no bending the truth, no nothing.

 

[KimB]

How is it a misleading statement? Radeon 9800 Pro has 8 pipes per TMU, and GeForce FX 5900 Ultra has 4 pipes per 2 TMU. That is a fact. There's nothing misleading about it, no bending the truth, no nothing.

You didn't state the second part: that the 5900 Ultra has 2 TMU's per pipe. That's a rather important part of the statement.

 

[surfhurleydude]

You didn't state the second part: that the 5900 Ultra has 2 TMU's per pipe. That's a rather important part of the statement.

The performance of the NV3x series says otherwise.

 

[KimB]

The performance of the NV3x series says otherwise.

The performance of the NV35 is not half the performance per clock as the R350, as your original statement would seem to indicate.

 

[surfhurleydude]

The performance of the NV35 is not half the performance per clock as the R350, as your original statement would seem to indicate.

It would only be half if the R350 was clocked at 450 MHz. Nonetheless, in case you can't tell, the NV3x series has absolutely horrible performance with true Trilinear aniso filtering turned on, which is why they result to optimizations and cheats. Perhaps you aren't aware of many custom benchmarks of less popular games, but in many situations the NV35 actually DOES come out to be a bit over half the performance of the Radeon 9800 Pro.

Perhaps you haven't noticed, but without optimizations, the NV35 is less than half the performance of the Radeon 9800 Pro in the 3dmark2003 Pixel Shader 2.0 test.

Also, if you haven't noticed, without optimizations, the Radeon 9800 Pro is multiple times faster than the NV35 in Shadermark 2.0 tests.

The Radeon 9800 Pro is 50% faster than the NV35 in Code Creatures benchmark when optimizations have been removed from both cards.

The Radeon 9800 Pro is about 90% faster than the NV35 in Unreal II: The Awakening tests with AA and Aniso on. With just Aniso, the Radeon 9800 Pro is about 80% faster.

And the list goes on and on. I hate to be the bearer of bad news, but the pipelines of the Radeon 9800 Pro really do show compared to the NV35 when both are shown under the microscope, it's just that nVidia has kindly hidden its true performance with hacks and cheats.

 

[nonamer]

The performance of the NV35 is not half the performance per clock as the R350, as your original statement would seem to indicate.

It would only be half if the R350 was clocked at 450 MHz. Nonetheless, in case you can't tell, the NV3x series has absolutely horrible performance with true Trilinear aniso filtering turned on, which is why they result to optimizations and cheats. Perhaps you aren't aware of many custom benchmarks of less popular games, but in many situations the NV35 actually DOES come out to be a bit over half the performance of the Radeon 9800 Pro.

Perhaps you haven't noticed, but without optimizations, the NV35 is less than half the performance of the Radeon 9800 Pro in the 3dmark2003 Pixel Shader 2.0 test.

Also, if you haven't noticed, without optimizations, the Radeon 9800 Pro is multiple times faster than the NV35 in Shadermark 2.0 tests.

The Radeon 9800 Pro is 50% faster than the NV35 in Code Creatures benchmark when optimizations have been removed from both cards.

The Radeon 9800 Pro is about 90% faster than the NV35 in Unreal II: The Awakening tests with AA and Aniso on. With just Aniso, the Radeon 9800 Pro is about 80% faster.

And the list goes on and on. I hate to be the bearer of bad news, but the pipelines of the Radeon 9800 Pro really do show compared to the NV35 when both are shown under the microscope, it's just that nVidia has kindly hidden its true performance with hacks and cheats.

 

[KimB]

It would only be half if the R350 was clocked at 450 MHz.

Sure, that's why the two have similar performance, even though the NV35 isn't close to twice the clockspeed of the R350.

Nonetheless, in case you can't tell, the NV3x series has absolutely horrible performance with true Trilinear aniso filtering turned on, which is why they result to optimizations and cheats.

That statement would mean something if the R3xx had the same quality anisotropic as the NV3x.

Perhaps you aren't aware of many custom benchmarks of less popular games, but in many situations the NV35 actually DOES come out to be a bit over half the performance of the Radeon 9800 Pro.

I think that's a self-delusion. Try posting links to back this up.

Perhaps you haven't noticed, but without optimizations, the NV35 is less than half the performance of the Radeon 9800 Pro in the 3dmark2003 Pixel Shader 2.0 test.

3DMark2003 is a useless benchmark. Real games have completely different goals than a benchmark like 3DMark03 has.

Also, if you haven't noticed, without optimizations, the Radeon 9800 Pro is multiple times faster than the NV35 in Shadermark 2.0 tests.

So? The NV35 is hard to program for. That's why nVidia made Cg, and that's why nVidia's got a rather expansive developer relations department. In the real world, nVidia's developer relations will make sure that every developer has the chance to optimize their game for nVidia's graphics cards. Non-optimized scores are therefore meaningless.

The Radeon 9800 Pro is about 90% faster than the NV35 in Unreal II: The Awakening tests with AA and Aniso on. With just Aniso, the Radeon 9800 Pro is about 80% faster.

Link.

 

[noko]

How in the world could Nvidia get away with that or even hide it???

I have a feeling that the R360 is more then just another tweaked R300 series at .15micron. The designs are run in parrallel with different variations. The threat of Nvidia with a killer .13 process was ever present so I would think ATI would have been developing a .13 version along with the .15 version. Since the .13 process was immature ATI did the right thing and ranned with the .15. Now the .13 process is more mature it is time for ATI to make the jump. Just my thoughts, no reall inside scoop.