ATI Radeon HD 4000 Series discussion

[adamsleath]

not surprised

[Levish]

I'd pay $500 for TOTAL erradication of more important things that affects gameplay, like stuttering. Oh, and also for a driver that works 100% without any kind of error. You guys only focus on FPS

the closest you'll realistically ever come to that is a console

[Helmore]

I'm not sure whether this has already been posted or not, but it seems that NordicHardware is doubting their own earlier supplied information (in a positive way though).
They are also pointing to may as the time for the first cards to surface, probably not retail though.

[clayton]

Damn. AMD got it right with their video (thanks to ATI that is). They gotta get their mind straight in the CPU war. I wanna be their consumer!

[perkam]

Looks like ATI will finally be able to get back up on its feet from the AMD takeover and the R500 fiasco (where it lost both the midrange and highend to Nvidia all in one swoop).

2 Years ago, if you told someone that ATI or Nvidia would be coming out a 480-shader card with 1GB of GDDR5/4 for $299, they'd laugh their heads off, give you the $300, and recommend you seek therapy.

Gone are the days of $1000 super rare uber-vid cards

[LyP0]

Gone are the days of $1000 super rare uber-vid cards

sssshtt...... not so loud, they'll hear you.

but indeed, the last 1000+ card was the 7800gtx 512 if i remember correctly.
the 8800ultra was like 300 bucks cheaper. and now the 9800gx2 is 200 cheaper than the 8800 ultra.
so if we wait we can have a 10800gx2 for like 300

[Zytek_Fan]

Dropping prices can do good to help out PC gaming (of course)

[junkun13]

looking good

[v_rr]

Prices are falling down again. But the discrete cards market share is to low (30&#37. The other 70% are IGP.
From that 30% many are low-end and mid-low-end so we can clearly see that most of the grafic cards are from that level.

Anyway the AMD 780G is a revolution in the IGP market (70% of the market).

[Hornet331]

its interesting to see they discreet market share climbed in one year 10% (from 21% to 31%).
It seems that people finally realize that igps suck.

[Mechromancer]

Only 16 ROPS just like my crappy 2900XT? Performance in today's (non-UT3 engine) games will probably still be mediocre. The R700 will probably PWN in all UT3 based games and newer game engines, though. Nvidia won't be able to keep up in shader intensive game engines this time around I bet. Unfortunately for ATI, there aren't too many of those game engines currently. I think the R700 needs to drastically increase the texture fillrate, compared to the R600, to be competitive.

[v_rr]

its interesting to see they discreet market share climbed in one year 10% (from 21% to 31%).
It seems that people finally realize that igps suck.

It´s mostly in notebooks, because they are growing much more faster then desktops.
Also many desktop come with in tegrated, the problem is that performance is low-end or something.

[Scimitar]

Only 16 ROPS just like my crappy 2900XT? Performance in today's (non-UT3 engine) games will probably still be mediocre. The R700 will probably PWN in all UT3 based games and newer game engines, though. Nvidia won't be able to keep up in shader intensive game engines this time around I bet. Unfortunately for ATI, there aren't too many of those game engines currently. I think the R700 needs to drastically increase the texture fillrate, compared to the R600, to be competitive.

It depends on what a certain game needs to run well. If it needs a high pixel fill rate, then only having 16 raster operation pipelines (ROPs) will probably be a limitation. If the game requires a high texture fetch rate, then the new RV770 core with 32 TMUs (double that of R600) should provide excellent performance. The R600 and RV670 were already good cards at pure shader limited apps like 3dmark06 and Oblivion. Adding 50% more shaders and increasing the clock speed will only make things better.

The 8800GTX was such a good card because it was so well balanced, equally good at many things. I am hopeful for the RV770, but having the same number of ROPs as the X1800XT does not inspire confidence for applications that require a high pixel fill rate.

Another downfall of the same # of ROPs is the AA /AF performance as well as the post processing capibilities. This was identified as a weakness in the R600 architecture. Unless ATI has figured out a way to do more with 16 ROPs than they did before, this will probably be a weakness again.

[LordEC911]

It depends on what a certain game needs to run well. If it needs a high pixel fill rate, then only having 16 raster operation pipelines (ROPs) will probably be a limitation. If the game requires a high texture fetch rate, then the new RV770 core with 32 TMUs (double that of R600) should provide excellent performance. The R600 and RV670 were already good cards at pure shader limited apps like 3dmark06 and Oblivion. Adding 50% more shaders and increasing the clock speed will only make things better.

The 8800GTX was such a good card because it was so well balanced, equally good at many things. I am hopeful for the RV770, but having the same number of ROPs as the X1800XT does not inspire confidence for applications that require a high pixel fill rate.

Another downfall of the same # of ROPs is the AA /AF performance as well as the post processing capibilities. This was identified as a weakness in the R600 architecture. Unless ATI has figured out a way to do more with 16 ROPs than they did before, this will probably be a weakness again.

Wow...

First games don't "need" a higher pixel fillrate, that is dependent on the resolution, obviously.
I also don't know if I would call G80 balanced...
Also this whole, R600's downfall is the ROPs and broken AA is BS. There seems to be a larger hit from AF, which is why AMD/ATi is doubling the texture units, than there is from AA.
16 ROPs will not hurt the RV770 at all, they simply need to do a little tweaking, like doubling the Z per clock from 2x to 4x, and things should be fine.

[Scimitar]

First games don't "need" a higher pixel fillrate, that is dependent on the resolution, obviously.

ROFL. Yes, they do. At a given resolution and settings, a card with better fill rate will perform better when the API demands that.

I also don't know if I would call G80 balanced...

It certainlly is. Name an area in which it is unbalanced. There are none. I challenge you to explain to everyone how the G80 architecture is not balanced. Compare this to R600 which is good as some things and poor at others.

Also this whole, R600's downfall is the ROPs and broken AA is BS. There seems to be a larger hit from AF, which is why AMD/ATi is doubling the texture units, than there is from AA.
16 ROPs will not hurt the RV770 at all, they simply need to do a little tweaking, like doubling the Z per clock from 2x to 4x, and things should be fine.

ROPS do a lot more than just AA. Read what I wrote.

[LordEC911]

ROFL. Yes, they do. At a given resolution and settings, a card with better fill rate will perform better when the API demands that.

Right... maybe up to a certain point but after that there are much more noticeable bottlenecks than pixel pushing...

It certainlly is. Name an area in which it is unbalanced. There are none. I challenge you to explain to everyone how the G80 architecture is not balanced. Compare this to R600 which is good as some things and poor at others.

Continuous floating point performance?
Non-hardware reliant AA?
There are a few more but that would just lead this topic even more astray.

ROPS do a lot more than just AA. Read what I wrote.

I understand that, where did I say anything different?

The point I was trying to make is that the changes the RV770 is supposedly receiving will make it a formidable opponent.

[pauldovi]

One thing I have learned about AMD / ATI GPU's, never believe the hype.

[ColonelCain]

Gone are the days of $1000 super rare uber-vid cards

Amen. Wait, don't you mean $650 for a single uber-card, that just begs to be shipped with a partner?

I personally like the fact that they are making multi-gpu solutions more affordable. Definatly is a nice change.

[LordEC911]

One thing I have learned about AMD / ATI GPU's, never believe the hype.

Same can be said for practically every company that has ever existed...

[largon]

Continuous floating point performance?

Oh-noes...
Theoretical throughput is lower than peak.
You don't really want to compare G80 to R600 and say SP efficiency sucks on G80...

Non-hardware reliant AA?

There is NO software AA in G80 simply because unlike ATi R6 chip, the ROPs do it 100% for G80. Besides, why should there be software AA? There's only performance to loose when offloading AA to SPs.

[geo]

RV770 is already in production

We learned that the RV770 GPU already went to production and there is a big chance to see R700 generation much earlier than anyone has expected. TSMC is already in pilot production and the chips are, as we reported earlier, developed in 55nm machitecture.

The way it looks now, there is a strong possibility that R700 should show its face at Computex (June 3) while the launch itself might be shortly before or after Computex. We still don’t have the final details.

source: http://www.fudzilla.com/index.php?op...=6721&Itemid=1

[largon]

If it's in production already then the first wafers are done ~2-3months from now. Then couple weeks for assembly, stockpiling and distribution...

[adamsleath]

.........->.......................->................[ati shop]

[LordEC911]

Oh-noes...
Theoretical throughput is lower than peak.
You don't really want to compare G80 to R600 and say SP efficiency sucks on G80...

Right... so 2/3 of the theoretical throughput is a good thing? This isn't a few dozen GFlops difference between theorectical peak and actual performance, this is a few hundred.



There is NO software AA in G80 simply because unlike ATi R6 chip, the ROPs do it 100% for G80. Besides, why should there be software AA? There's only performance to loose when offloading AA to SPs.[/QUOTE]
Ok.... Let's talk again in a few years about this.
Oh, almost forgot about the "only performance to loose" part. Where do you see that?

[largon]

2/3?
Are you referring to the proverbial "missing MUL"? (although then the number would be &#189

It's true that in some contexts nVIDIA says G80 is MADD+MUL. Technically it's true but there's a catch. The "missing MUL" isn't even supposed to handle shader instructions - it isn't even part of the shader core.

Infact, the actual shader ALUs appear to be quite efficiently exploitable as B3D found out:

We can push almost every other instruction through the hardware at close to peak rates, with minor bubbles or inefficiencies here and there, but dual issuing that MUL is proving difficult. It turns out that the MUL isn't part of the SP ALU, rather it's serial to the interpolator/SF hardware and comes after it when executing, leaving it (currently) for attribute interpolation and perspective correction.

http://<font size="1">http://www.bey...ws/1/11</font>

...Unlike the SPs on R6 of which only upto 2/3 are utilized even in synthetic tests.