Xbit on R350: they say 16 TMUs + DDRII

[megadrive0088]

right or wrong, that's what Xbit is saying. heh. I WISH

well actually, they're just mentioning it as RUMOR, but eh, who knows.


http://www.xbitlabs.com/news/story.html?id=1039657263


ATI’s R350 and RV350 VPUs are Already Taped Out?
Posted 12/11/02 at 8:41 pm by Anton

"As I discovered over The Inquirer today ATI’s forthcoming R350 and RV350 processors have been recently taped-out at TSMC’s fabs. According to the media source, the R350 graphics processor is manufactured using 0.15 micron process, just like we revealed a week ago (see this news-story), while the RV350 chip is made using advanced 0.13 micron technology.

R350 graphics processor is based on the R300 architecture, but with certain optimisations that allow higher core-clock speed. According to rumours, the R350 VPU incorporates 8 rendering pipelines with two texture units per pipeline, hence, there will be sixteen TMUs in this chip. Graphics cards powered by the R350 VPU will be equipped with DDR-II memory with 256-bit bus. Actual products powered by the upcoming processor will appear in Spring 2003.

RV350 VPU is said to be the next-generation mainstream offering and will also based on the R300 architecture. There is no information about its technical specifications, though, some expect it to show performance comparable to than that of the RADEON 9700. Given thinner manufacturing process and simplified architecture, the RV350 should be cheaper compared to its elder brethren with equal performance. We can expect the mobile version of the RV350 to be available shortly after the release. The availability timeframes of the part are unknown.

With the tape out now complete we may expect ATI to meet the expectations and launch the R350 based products sometimes in March. I do not believe that the company will unveil its RV350 earlier than in Summer, hence, the company will have a lot of time to optimise its next-generation mainstream VPU. As for the R350, honestly speaking, there is not a lot of time left for ATI to fully finalise the VPU."



I don't see them having two TMUs per rendering pipe now, (maybe!?)
but DDR-II sounds very tastey, and very likely.

 

[jvd]

why the hell not. If we can put a man on mars we can build an 8x2 video card.

 

[hkultala]

why the hell not. If we can put a man on mars we can build an 8x2 video card.

it's not if they can do/design it.
It's how big/expensive it would become.

These same 2-TMU rumors have been circulating for a long time without any "credible source" where these have come from,
some people have just speculated it, and some have misunderstood that speculation as a fact/rumor.

 

[Crusher]

why the hell not. If we can put a man on mars we can build an 8x2 video card.

Your flawed analogy only shows that other people can do other things.

 

[jvd]

Guys it was a joke.... shesh didn't you see the

 

[Kaizer]

I believe the rumour emerged from the speculations that the nv30 would have 2 TMUs which would give it a certain advantage in multitextured scenes and that ATi therefore would have to include 2TMUs in the R350 (which was also then believed to be the next big ATi thing..



Therefore the cause of the rumour might be some nVindian..



I can wait to see the R350, though, as ATi already has proven their capabilities, and can now improve upon the excellent R300 in every way known to mankind (except, perhaps including the RSN-unit)





With Regards

Kjetil

 

[Evildeus]

Any idea of how much more transistor it would require?

And for the price *ouch*

 

[Grall]

Poor Hellbinder... If this rumor is correct, that means he is wrong twice over, since he's already stated R350 will NOT feature DDR2.

Not that I expect them to be right, I have basically no trust left in that website.

*G*

 

[Humus]

The question is, why waste silicon on more TMU's? What we need is higher shader throughput.

 

[Crusher]

Guys it was a joke.... shesh didn't you see the

Apparently you don't read dilbert.

 

[jvd]

haha that was funny and now i get it , but i'm a getfuzzy kinda guy myself. That darn bucky kat reminds me of my sister and I'm snatchel and my dad is the guy.

 

[Grall]

I don't know if you guys noticed this, but a belgian guy has overclocked a Radeon 9700 Pro card to almost 500MHz GPU speed and 390MHz DDR by bumping the voltage and cooling it with 2-3 degree (C) water. The CPU was a 4.15GHz P4. 3DMark score (not that we care about such things ) was around 22500 I think.

It was at 501MHz GPU apparantly, but it started crashing.


I have no idea how many cards this guy tried before he found one that would do these insane numbers, but it still bodes well for ATI I think, despite the extreme cooling measures. 400MHz for R350 doesn't sound too far-fetched to me. Maybe even conservative? Who can say. Speed difference between GF3 and GF4 Ti4600 was positively HUGE, so one can always hope, even though the situation really isn't comparable. (Different chips, and all that...)

*G*

 

[WaltC]

I don't know if you guys noticed this, but a belgian guy has overclocked a Radeon 9700 Pro card to almost 500MHz GPU speed and 390MHz DDR by bumping the voltage and cooling it with 2-3 degree (C) water. The CPU was a 4.15GHz P4. 3DMark score (not that we care about such things ) was around 22500 I think.

It was at 501MHz GPU apparantly, but it started crashing.


I have no idea how many cards this guy tried before he found one that would do these insane numbers, but it still bodes well for ATI I think, despite the extreme cooling measures. 400MHz for R350 doesn't sound too far-fetched to me. Maybe even conservative? Who can say. Speed difference between GF3 and GF4 Ti4600 was positively HUGE, so one can always hope, even though the situation really isn't comparable. (Different chips, and all that...)

*G*

With some process tweaks to enhance yields, a little fine-tuning of the R300's pipelines, better cooling and a tad more voltage I could see R350 hitting 400MHz pretty easily, I suppose. I just don't buy the DDRII thing. nVidia *has* to do it because of their 128-bit bus--they have no choice. But since ATI was wise enough to move to the 256-bit bus they have much wider ram choices available. I'd think going to 2.2ns DDR1 or thereabouts would provide plenty of bandwidth for a 400MHz R350. I mean, just with the existing bus with DDR II you'd have darn near 40/gigs/sec bandwidth. Anybody think the R300 core could use it, or most of it?

I could see it for a 500-600MHz .13 micron version of the chip (R400), but until then why mess with all the added heat and cooling, as well as the extra expense, if you don't have to? I also think the dual-TMU rumors are false just as they were with the nv30--if they are shooting for higher clockspeeds a sure way to undercut that would be to add all of that additional circuitry (which is why I think it was left off the R300 and nv30--probably the gains weren't worth the clock-rate hit.)

 

[sabeehali]

What about 500 MHz DDR 1
I think DDR 1 has lower latency than DDR2 at the same speed so in theory 500 MHz DDR 1 would be better (and cheaper acc. to xbitlabs-> Hynix)
Also that would mean ATI will not have to redesign the memory controller until RV350 or R400 with GDDR3.
What do u guys think?

 

[arjan de lumens]

DDR2 has better protocol efficiency than DDR1, which may or may not outweigh the added latency. I would be a bit surprised if 500 MHz DDR1 is cheaper than 500 MHz DDR2, given that design timings would be much tighter in the former and that DDR2 was specifically designed to keep production costs down. And the memory controllers of the R300 already support DDR2, so ATI would not need to redesign anything to support DDR2 (and the changes needed for GDDR3 appear to be very small).

 

[Tagrineth]

AFAIK, memory latency generally isn't a problem for 3D graphics, with the accesses tending to be very predictable, and the pipeline being bloody deep.

 

[Hellbinder]

Poor Hellbinder... If this rumor is correct, that means he is wrong twice over, since he's already stated R350 will NOT feature DDR2.

And I Solidly Stand by that to.. Xbit is flat out WRONG.

Btw.. How does that Translate to Wrong twice over? you mean that Nv30 4x4 thing? *cough*...

Question For Dave B.

I am a little curious about your Doubt in the R350, based on your comments on the main page. You seem to be in the same boat as a lot of people that Fail to Realize that the R300 is Faster Clock for Clock than the Nv30. In current Games. I think thats going to become pretty damn clear in a few weeks time. As it is nearly a 200mhz Advantage only nets them a 30% speed advantage.. and thats based on internal Nvidia tests with older Ati drivers. I have also heard that Nvidia is not going to release the 400/900 version at all. WHy?? its pretty obvious that it would be slower than the 9700pro... (but that is just what i heard, not that its what im claiming)

I also find it really strange that you and Xbit call a 100mhz Core speed decrease a "Slight degrease in clock speed". Since when is 100mhz on a GPU *slight*.

 

[Vince]

The question is, why waste silicon on more TMU's? What we need is higher shader throughput.

Amen to that! It's like people are stuck in this legacy mindset thats left over from around 2 years ago. It's too bad people can't appreciate architectural elegence and beauty as they can bigger nomenclature.

 

[sabeehali]

DDR2 has better protocol efficiency than DDR1, which may or may not outweigh the added latency. I would be a bit surprised if 500 MHz DDR1 is cheaper than 500 MHz DDR2, given that design timings would be much tighter in the former and that DDR2 was specifically designed to keep production costs down. And the memory controllers of the R300 already support DDR2, so ATI would not need to redesign anything to support DDR2 (and the changes needed for GDDR3 appear to be very small).

I was surprised too but read on

http://www.xbitlabs.com/news/story.html?id=1039620241

 

[Evildeus]

]
You seem to be in the same boat as a lot of people that Fail to Realize that the R300 is Faster Clock for Clock than the Nv30. In current Games. I think thats going to become pretty damn clear in a few weeks time. As it is nearly a 200mhz Advantage only nets them a 30% speed advantage.. and thats based on internal Nvidia tests with older Ati drivers. I have also heard that Nvidia is not going to release the 400/900 version at all. WHy?? its pretty obvious that it would be slower than the 9700pro... (but that is just what i heard, not that its what im claiming)

Excuse me but, aren't these assumptions and claims from your part?
- How do you know that the NV30 is not faster than the R300 on a clock basis? Based only on some benchmarks? But if the NV30 is MORE cpu dependent that the R300, then you can't conclude on these benchmarks, you need to see the trend.
- You "heard" that the non ultra is not going to be release, and you assume that it's because it's slower? Well, why not, if it's slower and cheaper, what's the problem to release it?
- Do you put as fact that the R350 will be 425+ MHz?