http://img112.imageshack.us/img112/1097/g80va8.jpg

And if you need the die-pin shot, its here.

:lol: I noticed this divergence awhile ago, but it struck me as being exactly the kind of architecture you might expect if you purged your shader team of everyone who didn't buy the unified approach and put them on the texture unit team. It doesn't seem unified at all. The shader units are unified, but not the texture units.

You have single-channel dedicated addressing units, single-channel dedicated samplers, multi-channel dedicated addressing units, and multi-channel samplers, which are effectively just four-wide single-channel samplers, but, err, they're "dedicated". I'm not really familiar with the complexities of addressing units, but, there doesn't seem much more complexity to addressing a vector of data, so the dedication of four of those addressing units is puzzling as well.

I am sure I do not understand the whys of this particular architecture, so, I will be interested to hear more about it.

As for a comparison, iirc, each of 8800's 8 clusters have 4 TAs/quad-channel samplers and 8 quad-channel filtering units. I don't recall the fp16 speed on them, though the 8800's clock is obviously slower than R600's.

Remember that the 8800 was a surprising texture monster....

Can someone correct me if I'm wrong here?

Texture adress units: G80 = 32 R600 = 32

Texture filtering units: G80 = 64 R600 = 16

Right?


So can someone tell me how many texture samplers G80 has?

HIS webpage down, looks like AMD kicked HIS butt :smile:

what a nice smile, does DX8 support FP16 HDR? http://69.93.88.162/forum/images/smiles/rofl.gif

They won't dare -- valuable AIBs are too scarce those days. :lol:


From R600 Texturing slide:
"Can bilinear filter one 64-bit value per clock"

That should mean one FP16 component per cycle, or dual INT8 bilerps, as in G80.

Still weird its down :wink:

Hehe :D

I don't think it was broken out like that. Presumably each of the addressing units can sample a four-channel texture. So, 128 vs. 80. A GTX would have roughly the sampling power of a 920Mhz R600. Except, of course, that the comparison is false, as R600 has dedicated units and G80 does not. Presumably one quad-channel G80 sampler manages to be less flexible than four single-channel R600 samplers (each of the R600 samplers can come from distinct locations). A comparison by numbers absent discussion of workload would be ... difficult.

Edit: I should probably point out that what seems ideal to me would be a set of address units, and a set of component samplers and filters, which could be dynamically assigned/allocated. The cost/benefit ratio isn't at all clear, though. It also makes me wonder how much of that filter component is a lerp, while I eye G80's SFU in what might potentially be a criminal manner :)

It has 64 texture samplers, if I remember well

They are not necessarily reconfigurable like G80's filtering units. I think they would have made a bigger fuss about it if they could actually do that. G80's filtering hardware is nicely done, IMO.

?????

See my earlier post. Just search for "dedicated" :)

One item that hasn't been addressed is whether each texture unit is assigned to only one of the R600's shader arrays or not. [In that sense, G80 has 'dedicated' texture units, but that's not what I was speaking to....]

to someone can not afford the truth's strike:http://69.93.88.162/forum/images/smiles/rofl.gif

http://www.3dnews.ru/_imgdata/img/2007/03/30/44883.gif
http://www.chiphell.com/attachments/...lLYYLN4ngC.jpg
http://www.chiphell.com/attachments/...Zd4FUZu8iq.jpg
http://www.chiphell.com/attachments/...wPThwHVaVn.jpg

source:
http://www.chiphell.com/viewthread.p...37;3D1&page=25

'

I for one have to agree. It makes following this thread and treading through alot of the BS even harder.

when you provdie a screenshoot of R600XT with fps on it, someone here say: "hey, tester, one pic means meaningless. we want to see avg fps."

ok, when you provide a avg fps using fraps with FP16 HDR on, then the guy said:" What a magnificent dx8 speed."

Does DX8 support FP16 HDR? fine, I will not provide any 2900XT realgame test score anymore. I don't deserve this attitude.

Mao, in order to compare two cards you need to have the same system, the same same settings and the same test. That's the only problem, but it has been persistent throughout this thread. Just because you use the same game (let's say, STALKER), it doesn't mean you can compare results from one source to another.

but what mao5 provides us is by far better than nothing.

Dont get offended by some smarty pants, with comments like those all they are doing is making fool of themselves. ;)

Its good science, variables can change from one system to another, easier to just replace cards, run the same game with the same configuration.
Then there are no question about how the cards perform.

What i can conclude from the numbers you brought mao5 that the card do well.

Better than nothing but useless when compared to other test done in a different way (timedemo/benchmark instead of a single frame with fps number for that instance).
It's still something but it has to be seen for what it is, a screenshot with fps counter and not a timedemo/benchmark (a true benchmark that is!).

And could we please stop with all the strange/abuse of smilies becouse this thread is starting to look like a GAF thread. :smile:

FruitZilla think the hd2900xt default consume more than 240watt power :smile:

Good for him. How about linking to someone who has a shred of credibility....

Some fun needed in this topic too sometimes :wink:

After a cumulative 500+ pages of R600 discussion, I just want relevant information at this point. THAT is fun. Fudo is a waste of time. :wink: