Anandtech Reviews R600!

[Cheex]

The 8800GTS 320MB offers almost as much performance as the HD 2900 XT but at a price much lower. I think that an OC'd 320MB could equal its performance at low (800x600) to medium (1680x1050/1600x1200) resolutions for a lot less money.

 

[apoppin]

Originally posted by: Extelleron

Originally posted by: Regs
The whole review I got the impression that Derek was trying to say "Well at least it's not a complete flop".

Though brand loyalty is much stronger for a GPU part than it is for a CPU part. You can label brand loyalty "fanboi" all you want but people are going to buy what they most feel comfortable with.

If they want to ignore the signals that this card shows obvious drawbacks and needs a revision to show it's worth for the near future, then I almost feel sorry for them for wasting the 400 dollars for games that wont be out until later this year. DX10, I agree with many others, is for the 65nm parts coming from both Nvidia and hopefully ATi later this year.

However if you want better performance today from a x800 or x1900, and continue to be loyal to ATi, then I see no reason why not to buy the X2900. Just don't ask me to recommend it to you.

It really depends on the drivers for me. If drivers can improve performance by at least 15-20%, which I think they will be able to, then I think the HD 2900XT is an excellent value at $400. However, if things really don't change, then for the most part the 2900XT is a decent deal. If you intend to keep your card for a year or more, however, I would recommend the 2900XT over the 8800GTS.

i wouldn't recommend anything ... too dangerous here


anyway ... i DO think they will improve performance by about the *same* as the 8800 series has improved in the last 6 months ... and that the prices are gonna drop to about GTS 320 pricing

with a game bundle ... 'features' ... and a good price ... why not?

if you are really a performance junkie with a BIG budget, then the GTX ultra is for you

it is not for me

 

[apoppin]

Originally posted by: Regs

Originally posted by: apoppin

Originally posted by: MadBoris

I'm starting to thing some of the great specs are being crippled by some bottleneck internally. 512 bit memory bus, 320 shaders is great and all but...something is holding this thing back and it isn't just drivers.

is it a *feeling* ?

Part of the bottle neck was explained by derek in the conclusion:

And here's what AMD did wrong:

First, they refuse to call a spade a spade: this part was absolutely delayed, and it works better to admit this rather than making excuses. Forcing MSAA resolve to run on the shader hardware is less than desirable and degrades both pixel throughput and shader horsepower as opposed to implementing dedicated resolve hardware in the render back ends. Not being able to follow through with high end hardware will hurt in more than just in lost margins. The thirst for wattage that the R600 displays is not what we'd like to see from an architecture that is supposed to be about efficiency. Finally, attempting to extract a high instruction level parallelism using a VLIW design when something much simpler could exploit the huge amount of thread level parallelism inherent in graphics was not the right move.

Assuming what Derek said was accurate, it's a clear case of Ati having too many "good ideas" at once. A design too complicated and ahead of its time. The card could of spent another year or two in development.

that 'VLIW design' is evidently 'intentional' ... something that appears to continue throughout AMD's future designs

... so pretty hard to call it a "flaw" .. or even a 'bottleneck' ... just yet
the reviewer doesn't call it a 'bottleneck'

and that "thirst for wattage" is easily explained by being forced to run at a higher clock --by nvidia's g80

NOR does it *imply* the far fetched conclusion you are drawing from his comments

 

[fern420]

arggggg!!! those BF2 benchmarks are a little depressing. maybe i should of just opted for two x1950's instead, hehehe. at least the 1920x1200 resolutions are basically identical with a gtx and those are the only ones that matter to me.

 

[terentenet]

Extelleron, both camps can put out new drivers, both improve performance. Nvidia, after 6 months, just sorted out the BIG bugs in it's drivers. From here on, each Nvidia driver should fix minor bugs and improve performance as well.
ATI just started. They will have some hard months to iron out the problems with their drivers, especially in Vista. New card, new OS = big problems.

 

[ShadowOfMyself]

Originally posted by: apoppin

Originally posted by: Regs

Originally posted by: apoppin

Originally posted by: MadBoris

I'm starting to thing some of the great specs are being crippled by some bottleneck internally. 512 bit memory bus, 320 shaders is great and all but...something is holding this thing back and it isn't just drivers.

is it a *feeling* ?

Part of the bottle neck was explained by derek in the conclusion:

And here's what AMD did wrong:

First, they refuse to call a spade a spade: this part was absolutely delayed, and it works better to admit this rather than making excuses. Forcing MSAA resolve to run on the shader hardware is less than desirable and degrades both pixel throughput and shader horsepower as opposed to implementing dedicated resolve hardware in the render back ends. Not being able to follow through with high end hardware will hurt in more than just in lost margins. The thirst for wattage that the R600 displays is not what we'd like to see from an architecture that is supposed to be about efficiency. Finally, attempting to extract a high instruction level parallelism using a VLIW design when something much simpler could exploit the huge amount of thread level parallelism inherent in graphics was not the right move.

Assuming what Derek said was accurate, it's a clear case of Ati having too many "good ideas" at once. A design too complicated and ahead of its time. The card could of spent another year or two in development.

that 'VLIW design' is evidently 'intentional' ... something that appears to continue throughout AMD's future designs

... so pretty hard to call it a "flaw" .. or even a 'bottleneck' ... just yet
the reviewer doesn't call it a 'bottleneck'

and that "thirst for wattage" is easily explained by being forced to run at a higher clock --by nvidia's g80

NOR does it *imply* the far fetched conclusion you are drawing from his comments

And this is why it was a good move for ATi to move away from angle-independent AF on R300, and wait until a card that had the power to pull it off without performance loss

Nvidia didnt, they lost

Now Ati made the same mistake, they thought their card could handle it, and it cant, and now the AA performance is pretty much dreadful

Note to engineers - Do not add features your card cant even run decently

 

[Stoneburner]

so, architecturally what is the reason the R600 is not performing as well as one would expect?

 

[apoppin]

Now Ati made the same mistake, they thought their card could handle it, and it cant, and now the AA performance is pretty much dreadful

no it isn't

where do you get "aa perf is dreadful"?

 

[Wreckage]

Originally posted by: Stoneburner
so, architecturally what is the reason the R600 is not performing as well as one would expect?

Well if the R600 came out last year and the G80 was not out yet, it probably would not have been considered too bad of a card.

 

[apoppin]

Originally posted by: Wreckage

Originally posted by: Stoneburner
so, architecturally what is the reason the R600 is not performing as well as one would expect?

Well if the R600 came out last year and the G80 was not out yet, it probably would not have been considered too bad of a card.

it still isn't considered too bad of a card by credible reviewers

nvidiots already have their minds made up it is crap

 

[schneiderguy]

Originally posted by: apoppin

Now Ati made the same mistake, they thought their card could handle it, and it cant, and now the AA performance is pretty much dreadful

no it isn't

where do you get "aa perf is dreadful"?

this is just going from the Anandtech review since I havent read the others yet besides the Hardocp one which is useless.

falls behind the x1950xtx in BF2 w/ 4x AA, it was ahead without AA
falls behind the 8800gts's in Oblivion w/ 4x AA, it was ahead without AA
prey it does okay
R6 and STALKER weren't tested with AA

so in 2/3 benchmarks it performs poorly with AA on compared to other cards

even the article says AA performance is worse than it should be

Forcing MSAA resolve to run on the shader hardware is less than desirable and degrades both pixel throughput and shader horsepower as opposed to implementing dedicated resolve hardware in the render back ends.

So while the AA performance isnt "dreaful" it isnt great either, especially considering how much memory bandwidth the 2900 has.

 

[Stoneburner]

but there seems to be some gap between the performance expected considering its specifications and its actual performance. Is there one thing that's holding it back? Is it the failure of the anti aliasing hardware? Or is it the decision to including only 16 texture units in?

 

[Nightmare225]

Its that:

Forcing MSAA resolve to run on the shader hardware is less than desirable and degrades both pixel throughput and shader horsepower as opposed to implementing dedicated resolve hardware in the render back ends.

 

[Matt2]

Originally posted by: Stoneburner
but there seems to be some gap between the performance expected considering its specifications and its actual performance. Is there one thing that's holding it back? Is it the failure of the anti aliasing hardware? Or is it the decision to including only 16 texture units in?

I lean towards something about the architecture that is holding it back. Not just drivers.

If you read the DriverHeaven review, they hit 865/1026 on their card and it only gained 6 fps in FEAR and 1 fps in Prey.

I'm not sure why no one has mentioned this before, but it doesnt look good.

 

[apoppin]

So while the AA performance isnt "dreaful" it isnt great either, especially considering how much memory bandwidth the 2900 has.

read the rest of the reviews

i am seeing variability in performance due to drivers

Yes, Matt2 it's called *immature drivers*




they make a HUGE difference in performance

try loading the release drivers that came with the GTX ... right now ... then test performance ... i bet the HD 'wins' - much more - over the GTX and certainly the GTS
:Q
if nvidia cna improve drivers, so can AMD ... and i think AMD will execute faster and better.

can you say "new" HW ?

 

[Nightmare225]

Originally posted by: apoppin

try loading the release drivers that came with the GTX ... right now ... then test performance ... i bet the HD wins over the GTX and certainly the GTS

Judging by what Extelleron (spelling?) posted in another topic, nowhere close. Don't get your hopes up

 

[Matt2]

Originally posted by: apoppin
read the rest of the reviews

i am seeing variability in performance due to drivers

Yes, Matt2 it's called *immature drivers*




they make a HUGE difference in performance

try loading the release drivers that came with the GTX ... right now ... then test performance ... i bet the HD 'wins' - much more - over the GTX and certainly the GTS
:Q
if nvidia cna improve drivers, so can AMD ... and i think AMD will execute faster and better.

can you say "new" HW ?

I dont see how drivers are causing the card to have ZERO performance increase when overclocking the GPU by 115mhz and the memory by ~400mhz.

Especially since it was overclocked with the OC utility provided to reviewers by AMD.

 

[apoppin]

Originally posted by: Nightmare225

Originally posted by: apoppin

try loading the release drivers that came with the GTX ... right now ... then test performance ... i bet the HD wins over the GTX and certainly the GTS

Judging by what Extelleron (spelling?) posted in another topic, nowhere close. Don't get your hopes up

well, from what i read ... nvidia has improved performance in the 8800 by quite a bit

anyone know the overall percentage? ... i briefly looked at some of the new drivers and saw double digit increases in some games over the previous release ... especially the first two months when the nvidia drivers were a *mess*

if those same games that got 10%+ increases would obviously be running 10% slower with the old drivers ... and that is less than the difference with the 'xt' in some cases

i *expect* to see at least 10% improvement in HD-XT scores - across-the-board - and probably closer to 20% in many cases where the benchmarks are *odd*

if not, i will write off the HD XT as a "turd"

and buy an nvidia card

I dont see how drivers are causing the card to have ZERO performance increase when overclocking the GPU by 115mhz and the memory by ~400mhz.

Especially since it was overclocked with the OC utility provided to reviewers by AMD.

then do some research and you will be enlightened

drivers ... control ... well, *everything*

 

[Regs]

Far fetch maybe. I can't stop to help out AMD a little since I am a loyal AMD customer.

But if you want the blunt truth the they did too much, too late, with too little.

 

[PhatoseAlpha]

Even if there were huge improvements, would it be reasonable to expect the same from the HD2X00 lines? IIRC, the 8800 was a new architecture for nVidia, so you'd expect fairly immature drivers. The HD2X00 though, that's a second generation of the X1800ect architecture, so you'd really expect the drivers to bit a bit more mature from the get go. I'd really expect better stability out the door, but less room for optimization, since at least some of that would've been done last generation.

 

[Nightmare225]

Originally posted by: PhatoseAlpha
Even if there were huge improvements, would it be reasonable to expect the same from the HD2X00 lines? IIRC, the 8800 was a new architecture for nVidia, so you'd expect fairly immature drivers. The HD2X00 though, that's a second generation of the X1800ect architecture, so you'd really expect the drivers to bit a bit more mature from the get go. I'd really expect better stability out the door, but less room for optimization, since at least some of that would've been done last generation.

Nah, R600 was completely new, at least for the PCs. It shared more with the R500 in the 360 than the R580 used on pcs...

 

[GundamSonicZeroX]

Originally posted by: DaveSimmons
It looks like the hack to get AA working hurts performance bad.

Numbers for Unreal 3 engine (Rainbow 6, no AA) and Oblivion without AA were both good. But given the power draw (19 watts above a GTX) and poor AA speed I think I'd wait for the 2950 refresh.

Luckily I'm waiting until the end of the year for my next card anyway, so I'll hopefully have 2950 and 8900 to pick from. nvidia Vista drivers might even work by then

Maybe Vista will work by then.

 

[Bill Brasky]

Originally posted by: Nightmare225
Its that:

Forcing MSAA resolve to run on the shader hardware is less than desirable and degrades both pixel throughput and shader horsepower as opposed to implementing dedicated resolve hardware in the render back ends.

IMO, this was the most insightful point from Anand's review, and unfortunately this bottleneck isn't something that can be corrected with better drivers. I'm sure efficiency can be improved, but this won't mask the fact that the shaders are also calculating AA. I would bet AMD is waiting to release the XTX until this AA problem is fixed and the full potential of this architecture can be realized.

 

[Regs]

Originally posted by: GundamSonicZeroX
and 8900 to pick from. nvidia Vista drivers might even work by then

Maybe Vista will work by then. [/quote]

Better hope n7 doesn't see that remark!

 

[apoppin]

Originally posted by: Regs
Far fetch maybe. I can't stop to help out AMD a little since I am a loyal AMD customer.

But if you want the blunt truth the they did too much, too late, with too little.

*AGREED*

100% agreed

and i NEVER liked AMD

ever

i respected them ... and loved their products to keep intel 'honest' ... but i hated their inane marketing and ridiculous boasts

and now they *killed* the only company i DID like ... it is GONE ...
... anmd to make it much WORSE [in my book] - they LIED about r600 and spun a fantastic story about a "family launch" and "no problems with the HW"

i called BS on them in January

remember how *everyone* hated me?
[where's the beef?]

and kept ragging on them like a rabid bulldog with lockjaw :lips:

then it 'changed' ... Ruiz musta got the 'message' that we were sich of the deception and he came as 'clean' as i have ever seen a CEO outside a courtoom ...

and i began to really see what AMD intended, the *reason* for the merger and the only way they had to fight intel ... and they leveled

then Anand came out with his AMD article and i began to anticipate - and appreciate - this weird little AT-AMD 'hybrid GPU'

if you notice, my predictions were 'spot on' in many cases regarding HD2900xt

so i have VERY *mixed emotions* when i consider AMD


/end of long-winded personal story of enlightenment regarding AMD