[B3D Analysis] R600 has been unleashed upon an unsuspecting enthusiast community

I'm confused what you're trying to say. The original UT has an incredibly low density of texels per output pixel (by modern standards). It should run at hundreds of frames per second, if not thousands.

Jawed

 

i think he means the new UT thats coming out... whats it being called these days?

 

Kombatant speaks his thought.

Intresting points, I look forward to the 8.38 driver.

 

Maybe UT2004? Which was largely texture limited in my experience? Though I'm not sure how good of a benchmark it'd prove to be these days. Even for this type of test.

 

The new engine is called Unreal Engine 3.
And the coming unreal game is "Unreal Tournament 3", formerly "Unreal Tournament 2007"

 

Aha. And erm, yeah, UT is newer than Unreal, erm, whatever from 2000/1 or whenever it was

I'm a Quake 3 guy, maybe that explains it.

Jawed

 

I think Kombatant's viewpoint as an enthusiast means his priorities lie outside of much of the market.

Is the 8-pin connector there for overclocking?
Maybe, it makes sense.

Does a chip's overclocking potential mean it suddenly becomes a better product?
Not necessarily.
The question one has to ask is, if it is such a great overclocker, why wasn't it just released at a higher speed.
The corporate and engineering answer is likely that higher clocks could not be reliably hit within the conservative specs needed for a reliably mass-marketed product.

Overclocking is still a crapshoot, and anecdotal evidence of good clocking is often fodder for wishful thinking.

I reserve judgement on overclocking the card until we see more samples out in the wild.
A good portion of the examples we have were review samples, and I don't think AMD would give review samples that weren't decent overclockers.

 

sounds very optimistic for me. but we'll see. the g80 drivers were also far from being perfect after it was released and it took nearly three months until the drivers worked correctly on most games and on the average there have been performance improvements be 10-15% since then... right now the driver for r600 seems to be not in a very good state.

 

I think his post is more damning to R600 than anything else..
So the card is aimed at enthusiasts that are willing to change their PSUs and tamper with cooling solutions just to reach GTX/Ultra performance? What's the point? And the Conroe analogy didn't work at all, Core2Duo wiped the floor with Athlons without any overclocking to begin with...

And waving the next-gen features line again-tessellator, perf with DX10-, Oh come on ATI haven't you learned anything.

 

You don't put out an under performing (and possibly faulty?) product and then say "well, if you're lucky and you got a good chip, if you put in a bigger PSU, if you change the cooler and have a good case, and you're willing to run it out of the official spec, then you might get it to run on par with competing products". That's no way to sell a product, and it doesn't seem to be the official line from AMD.

Wasn't it ATI not so long ago telling us how they had on-die fuses that would blow to let them know if someone had overheated/overvolted a chip so they could deny a warranty replacement? And now we have an ATI marketing manager commenting on his own time that it's a good thing for people to have a product that instead of beating the competition, has the possibility of beating the competition under all the caveats I mentioned above?

How on earth could ATI paint themselves into this corner? Worse IQ than Nvidia? Massive hit when AA is enabled? Hotter? Noisier? Driver performance problems all over the place? Yes, it's cheaper, but then you'd expect an inferior product to be cheaper. The only thing it has going for it is that if I didn't want to spend much money, I get something okay for the price. Most people buying on a budget won't overclock, and most overclockers won't start with an underperforming part.

AMD had better take the hit and get R650 out the door ASAP, because the promise of that coming soon is the only thing that will stop people who have been waiting on R600 from looking at both R600 and G80 and choosing the G80. AMD is going to be losing both marketshare and mindshare, and that will hurt them for years to come.

I like ATI products, I like their driver support, but why would I buy this product? It simply doesn't live up to expectations.

 

Florin

I think high end computers can be problematic enough already without overclocking, so for me personally this point of view isn't particularly relevant. But nearly all of the reviews I've seen point out that the 2900XT puts out a remarkable amount of heat, so I'll be interested to see if these cards will indeed turn out to be the promising overclockers Kombatant makes em out to be.

As for his statement that there is no question that the 2900XT is the better card when compared to the 8800GTS 640.. it kind of sounds like he's trying to convince himself..

 

I like to think newer R6xx GPUs will have smaller SIMDs. In fact RV630/RV610 both do have smaller SIMDs, so they might point to the future. Obviously the cost isn't trivial.

It's interesting that RV630 pays this cost. It's also interesting that Xenos style ALU and TEX layout has turned out to be ATI's choice. I honestly thought R5xx's style was a newer refinement. But now it seems like it was a half-way house, compromised by the scale (and distributed nature) of the architecture and the lack of L2 caching. It was just a lot cheaper to implement it seems.

It's funny, I remember struggling with R5xx's ALU and TEX layout originally because it was different from Xenos's...

Jawed

 

Because it would exceed the power envelope of 225W, necessitating a new PSU for most users...? I reckon the hope was for clocks ~850Mhz on release.

I think AMD's hand was forced by various technical & market factors. Kombatant's treatise just doesn't make sense for any rational firm that is pricing to market. Who expects AMD to follow a similar policy for its Barcelona core? Ie, to release a single default speed grade, selling for a relatively low MSRP, that will be in the lap of the end-user to speed bin...?

 

I think Kombatant's comments about the cooler are unreasonable.
Most reviews of the R600 (and i've seen a bunch of then already) say that it's loud and whiny, definitively not what you would expect from all that copper, heatpipes and blower. Could the slightly different angle of attack from the blades (compared to the equivalent from Nvidia) be responsible for that phenomenon ?

But many reviews also stress that the main culprit is that fan speed management. Some even report the fan speeding up in such simple things as using the standard Vista AERO Glass theme...

 

His analogy with Intel is fatally flawed because although Intel gives you a 2.4Ghz cpu that will overclock to 3.6GHz it also charges you an awful lot for a higher performing chip as well, the X6800. AMD have decided not to do this, if you believe him, but instead are being generous and giving customers the chance to get great value for money by paying less and turning their XT into an XTX at home. AMD must be awash with money at the moment to be so charitable.

His other points may or may not be valid though, even if they are valid you cannot expect nVidia to just sit there twiddling their twizzles.

 

So, what then is the end reason for R600's underperformance?

Is it still considered the texture capabilities, mostly?

That's still what I'm thinking. I hear other things tossed around but they seem to get shot down or be based on rumor. The one consistent thing the reviews have mentioned as a problem is light texture specs. It seems here may be other problems but they seem not as big a contributer.

Next Q, a doosy, does R600 have more shader power than G80? Is it another typical shader heavy ATI design, or can it not even claim that crown?

Has somebody simply ran a bunch of non-texture limited shader benches and compared G80 to R600?

 

The problem is that Nvidia could smash ATI easily with 196 shaders and near 2 Ghz shaders core in 65nm, reaching almost 1 tera flop getting better the MUL dependencies. The only improvement i see in the ATI R650 is a 900 mhz-1 ghz speed boost ( remember that Rv630 will be clocked at 800 mhz in 65nm and being a smaller chip it must be that the architecture is also hot in 65nm ).

 

haven't been paying attention now have we?

http://www.techreport.com/reviews/2007q2/radeon-hd-2900xt/index.x?pg=3

as far as we've seen
It masters Vertex Shading, but loses at everything else.
The only shadermark tests it wins are HDR tests.
it lands between the GTS and GTX on the shadermark scores.

The techreport article also uses an AMD benchmark (posted here two pages upstream I think) where performance difference is shown between serial and parallel instruction performance.

 

What kind of meaningless statement is this?

ATI could easily smash that with 480 stream processors clocked >1ghz.

Or I could have just made another meaningless statement to counter yours?

 

So, it is a good thing that we _need_ 8pin connector for overclocking, yeah. He failed to mention that GTS can be oc'd quite well too, and without "extra juice".

Kombatants comment about fan is quite weird too, because it can and has been objectively tested.