News: R520 third tape-out Successful

[Afterburner]

It appears that R520 taping out has been sucessful but it has taken ATI three tape-outs to achieve its target. Looks like what ATI Taiwan said is true after all. The R520 was reported to run @ 600mhz thereabouts but now its believed that it will run even higher.

Find out more here (http://www.theinquirer.net/?article=24540) :fact:

[Frisch]

So what we are surpose to wait for, is a HOT-BREEZE this autum

[Cooper]

However R520 has only 16 pipelines. But that doesn`t show less performance compared to 24 pipelines in G70. VGA architecture is very flexible, so R520 could beat G70 with less pipelines.
Source (http://www.overclockers.ru/hardnews/19419.shtml)

[krille]

Any English source for that, maybe? Or would you care to expand it yourself?

Thanks

[Cooper]

No english sources. This is original info - comes from reliable sources. Also this was discussed at Computex2005

[lizardmech]

16 pipelines?! It seems like ATI need to have a card thats a fair bit faster than the 7800gtx if they want to make up for being 3 months later than nvidia.

[Frisch]

However R520 has only 16 pipelines. But that doesn`t show less performance compared to 24 pipelines in G70. VGA architecture is very flexible, so R520 could beat G70 with less pipelines.
Source (http://www.overclockers.ru/hardnews/19419.shtml)

I have translated the link provided in the quote, it's ugly, but it doesn't get better. It's from Russian, i can decode the ugly translation, so can you.
Theres two links in the text, where one of them goes to an english source, no problem there, but there's a second link in the text to another russian text which i'll translate later.



The third revision R520 works at the frequency of more than 600 MHz
and has 16 piksel'nykh conveyors Lexagon/12.07.2005 14:33/references
to the material

Last week some sources already reported that in the middle of
the third quarter of 2005 nachnutsya the mass deliveries 0.09 m of
chips R520, and this wakes already the third revision from the moment
of preparing the first operational models. Certainly, the actual
number of revision and its designation can it did not correspond to
the number "three", but the quantity of approaches, which ensured the
readiness of chip for the mass production, exactly equal to three.

By the way, the prime cost of pre-production of each revision is equal
to approximately $5-6 mln.; however, these expenditures can simply it
was dissolved in the budget, spent on the development and the
pre-production of entire family R5xx. furthermore, a good level of the
output of suitable chips can it paid additional expenditures for
redesign sufficiently rapidly. For example, the first revision ensured
output on the order of 20 suitable nuclei from one plate, the prime
cost of production by which was equal to $3000. Easily it calculated,
that each chip of obkhodilsya into $150, while the level of
profitability was located much lower, in region $27-30.

As reports site The Inquirer with the reference to its own sources,
now 0.09 m technical process reached the neobkhodimoy maturity, the
let out chips R520 not only were capable of working at frequency on
the order of 600 MHz and above, but also tyuey otlichayutsya by a good
level of the output of suitable crystals.

Furthermore, for us it succeeded it refined the number of piksel'nykh
conveyors of chip R520. judging by everything, chip wakes it had 16
piksel'nykh conveyors as it was predicted even during March, in the
days of conducting exhibition CeBIT 2005. Only then the information
about the presence of 24 piksel'nykh conveyors in chip G70 introduced
confusion - with many of our associates it seemed that for the victory
above G70 chip R520 must it had more than piksel'nykh conveyors. Here
were added rumors about the presence in chip G70 of partially off 32
piksel'nykh conveyors, that still more "tangled the consequence". Only
recently we was able it ascertained that there are no hidden "on the
black day" of piksel'nykh conveyors in chip G70.

Now came time it was dismantled with a quantity of piksel'nykh
pipelines of chip R520. as we already they said, them only sixteen.
Meanwhile the producers of graphical solutions have already long ago
given he understood that by especially quantitative characteristics
described the level of the operating speed of chips impossibly.
Especially because the up-to-date graphic chips imeyut this
flexible architecture, that can it readjusted the operational
characteristics in accordance with the situation. These sixteen
piksel'nykh pipelines as a result can it worked more effectively
than twenty four pipelines of chip G70.Especially because on side R520
and advantage in the frequency - 600 MHz (or above) against 430/470 MHz
in chip G70. in any case, the facts of reaching by single videokartoy of
result into 10 000 "parrots" 3DMark05 as unsubstantiated recognized
complicatedly, since similar communications was supply with sounied
by several nezavisimymi sources in the days of conducting Computex 2005,
at which the models of videokart on base R520 were demonstrated behind
the closed doors. Cannot it excluded that the present revision of chip
possesses an even higher productivity.

[Cooper]

The second link provides info upon G70 pipelines quantity. If you remember I posted a rumor thread about G70 having 32 pipelines. Well this link describes that it doesn`t. info taken from HKEPC - latest version of RivaTuner shows that G70 has only 24 pipelines

http://www.overclockers.ru/images/news/2005/07/04/g70_01.jpg

P.S. frisch do you know russian, or you just used a translator? If so could you plz pm me about it - might come in handy in such cases like this one. thanks

[Frisch]

oftopic
@cooper
done

[DilTech]

Sounds like ATi's stock is about to drop even lower if the 16 pipeline story is true...

[turtle]

Sounds like ATi's stock is about to drop even lower if the 16 pipeline story is true...

Why? Didn't the article just say that when tested behind closed doors the 16 pipeline R520 from the first tape out was more efficient (ie better) than G70 with it's 24pipes, and that one card scored over 10,000 in 3dmark05? It also goes on to say that the third tapeout probably allowed even higher productivity (and clockspeed?).

Sounds good to me. I just wonder how ATi is going to find a match for the 7800gt if it is indeed $400 (and gtx $500) and just a lower-clocked gtx...especially if the GT overclocks well. This 2+ month advantage on nvidia's part is really going to help in the end, because it'll allow them to drop prices quicker because they'll already have made their money off of their high-end and will be ready to release the refresh before long.

Even if R520 is a superior product, to me at least it's going to depend on by how much as to wither or not it's going to be worth the extra ching over the 7800 series. Let's hope it is. :D

[Turok]

They shoudl make a first.

Their r520 should stay at $500 and at the same time beat the 7800gtx by 30%, that will stomp nVidia for good since now ATI will have SM3.0, will offer CrossFire, and will be a cheaper and xtreamely powerful dual card colution.

The 16 pipeline being faster than 24 pipeline sounds a bit simmilar to what P4 Prescot's pipeline is compared to A64's pipelines. Prescots had a lot of unnecessary pipelines that just made it more expensive and harder to cool down

[cadaveca]

nv3x anyone?

r3xx?


g70, r520?


sound familiar?

[DilTech]

Why? Didn't the article just say that when tested behind closed doors the 16 pipeline R520 from the first tape out was more efficient (ie better) than G70 with it's 24pipes, and that one card scored over 10,000 in 3dmark05? It also goes on to say that the third tapeout probably allowed even higher productivity (and clockspeed?).

Sounds good to me. I just wonder how ATi is going to find a match for the 7800gt if it is indeed $400 (and gtx $500) and just a lower-clocked gtx...especially if the GT overclocks well. This 2+ month advantage on nvidia's part is really going to help in the end, because it'll allow them to drop prices quicker because they'll already have made their money off of their high-end and will be ready to release the refresh before long.

Even if R520 is a superior product, to me at least it's going to depend on by how much as to wither or not it's going to be worth the extra ching over the 7800 series. Let's hope it is. :D


The 10k 3dmark05 score was said to be made with a fully working r520(as reported by a few news sites). At that point, the R520 was suppose to weigh in at 32 pipelines, and they were having problems getting all the pipelines working. This would mean, you aren't going to see a 10k score out of the R520, atleast not if the 16 pipe rumor is true. This new tapeout is the one to bring the pipes down to 16, as it originally was infact a 32-pipelined monster.

Either way, this looks like ATi's version of the NV3x mistake... Only time will tell, but alot of sites have been calling it that for months now.

[Shadowmage]

If the 16 pipe rumor IS true, then it was designed as 16 pipes (although it's 16x2) the entire way. IF ATI goes with 16 pipes, then these won't be conventional pipelines, for obvious reasons.

[Orion24]

10,000 3DMark2005 & 16 pipelines only? I need to see it to believe it. The 10,000 probably refer to 32 or 24 working pipelines, like DilTech says.

[DilTech]

Shadow, did you forget they just re-spun the chip?... It differs from the original version, just like how the NV43 A1-A3 had 8 pipes, but the NV43 A4 had only 4. It's possible to remove pipes on a re-spin.

Also, did you see it say 16x2 anywhere?.. I agree that that's probably the case, but I didn't see posted on ANY site that the card will be 16x2... I did see claims that the 10k score came from a 32 piped R520 though, ironically enough it was on EVERY site that posted that tidbit of info....

[Shadowmage]

Seriously, it's pointless to release a standard 16 pipe chip. Therefore, the news is fake.

The only way it COULD be real is that it performs better than the G70.

[DilTech]

I'm not doubting this at all shadow, up until like 9 months ago, ATi was readying the R500 for this round. The r500 was bought by microsoft, making ATi work on a completely different gpu. 9 months is nowhere near enough time to finish up that complex of a chip, especially on a newer process, and get everything working properly. Thus, they lowered the pipeline count... Of course, I'm sure they raised something else up in it's absense, like maybe ALU's, but it's still 16 pipes if this is true.

However, how could "the only way it could be real is that it performs better than the g70"... That's a fanboy statement if I've ever heard one. Are you saying ATi isn't capable of messing up? Need I remind you of pre-9700pro? There was one specific case where ATi released a new card that beat the NV card, and NVidia simply released a new driver which put their OLD card ahead of ati's new one.... I wish I could remember which card it was, but I'm sure someone here will know what I'm talking about.

I'm gonna repeat this again.. ATi might just be biting off more than they can chew.

[Entity_Razer]

Shadow has a point. 16 conventional pipes would have to be clocked VERY high to beat a 24 conventional pipe architecture. Unless ATI already is using the X archtexture (X was it not?) That would give at least 50% performance gain at certain times.


I think SM3.0 and some other goodies are added in the R520. And I don't think they removed pipes. That would kill ati

[cadaveca]

ahahahahaahahaha.


it has ALWAYS been 16 pipes. some dude misunderstands "32-way pixel pipeline" to mean 32 pipes, instead of "16 unified shader pipes" now everyone thinks it's 32. my god.