Radeon 9700 NDA Lifted

[CMKRNL]

The Radeon 9700 is still 2 months away(Sept timeframe). That puts the NV30 just 3 months behind, not "almost half a year".


They said it will be released in under 30 days. That's 1 month away, not 2. NV30 on the other hand will launch in oct-nov timeframe but probably wont ramp up and be available in stores until dec-jan (based on the info that nVidia is feeding to the analysts for their financial predictions). Additionally, if you compare the dates for first silicon of the R300 and NV30 and assume that they both require roughly the same number of spins before production, ATI has a 5 month lead. This also matches the projected dec-jan timeframe. Bottom line, ATI has a several month lead over NV30 and by the time it comes out, they'll be ready to counter with a refresh of the R300 in 0.13u.


Still, the NV30, if it uses 0.13u, should have a decent advantage.


If the above is true, I expect that ATI will have a 0.13u part within a month of NV30 being available. This will negate any engine clock advantage that NV30 will have and as noted in one of the reviews, the R300 is already capable of supporting DDR-2.

The one advantage that NV30 will have over R300 is the programmable triangle tessellation unit. It's a fantastic feature but unfortunately one that will not be taken advantage of by game developers in its product lifecycle. (Heck, games are just now starting to take advantage of shaders which have been around since NV20!) Still, it's nice to see this type of innovation because it paves the way for making this a standard feature in future designs.

 

[Ailuros]

I am impressed so far.

Could someone elaborate further on the Smoothvision II algorithm? SA maybe?

 

[Dave Baumann]

Interesting

http://www.viahardware.com/atilaunch_3.shtm

Look at the last comment:

 

[Matt Burris]

Slap me and call me Sally if I'm wrong, but isn't the Parhelia the first card that offers over 20GB/sec of memory bandwidth, or does it mean real world rather than theoretical?

Also, explain that last comment to those of us that has no idea what it means. The 256-bit pipe I can understand, but 256 simultaneous units??

 

[Joe DeFuria]

This is the first time in recent memory that a graphics product has actually exceeded my expectations.

I am still flabbergasted at the engineering feat of ATI producing a 0.15 micron part with 110 million transistors, 8 full floating point pipelines, and getting 300+ Mhz out of it.



I am hugely impressed by the AA scores. Final judgement to be reserved of course, until we can properly judge the compatibility and image quality of their AA and anisotropic implementation. But this is the first time in a long time that I have been really excited about a new part.

And the first part with a $400 MSRP that is, imo, actually worthy of that price tag.

 

[Ailuros]

Matt,

How much bandwidth do you get with 275mhz DDR on a 256bit bus?

 

[LeStoffer]

The one advantage that NV30 will have over R300 is the programmable triangle tessellation unit. It's a fantastic feature but unfortunately one that will not be taken advantage of by game developers in its product lifecycle.

Huh? I missed that info. Where does it come from? I want to read more about it!

Ack, these are interesting times indeed! It's almost so good that you wish you could do a bit of programming to play around with these new tools!

 

[Joe DeFuria]

"256 simultaneous units" say, IMO, multi-chip solutions. Scalability to 256 chips?

Man would that kick ass. That would likely have to be on a multi-machine "render farm" configuraion, of course.

 

[cybamerc]

The R300 was clearly what Carmack was talking about when he mentioned upcoming multi-chip solutions. 256 units in parallel is pretty damn impressive.

 

[Matt Burris]

Ailuros:

Hell if I know, I failed Algebra 3 times, and I'm allergic to calculators.

256 R9700 renderfarms? I guess Morrowind would finally be playable in that kind of setup.

 

[pascal]

This is the first time in recent memory that a graphics product has actually exceeded my expectations.

I am still flabbergasted at the engineering feat of ATI producing a 0.15 micron part with 110 million transistors, 8 full floating point pipelines, and getting 300+ Mhz out of it.



I am hugely impressed by the AA scores. Final judgement to be reserved of course, until we can properly judge the compatibility and image quality of their AA and anisotropic implementation. But this is the first time in a long time that I have been really excited about a new part.

And the first part with a $400 MSRP that is, imo, actually worthy of that price tag.

I am impressed too

If Nvidia use with NV30 the same design methodology they used with GF4Ti then I doubt NV30 will have any clock advantage (maybe memory) because R300 was hand designed, is generating a lot of heat and the .13 micron has only 30% more performance advantage (TSMC info) over .15 micron process.

Also with 5 months difference this card may go sub $300 and my guess NV30 will have hard time competing with price.

Finally when ATI launch a .13 micron R300 it will get a nice 30% boost 8)

 

[Ailuros]

MattB,

On final release Parhelia has with 275mhz DDR 17.6GB/s bandwidth. If ATI uses higher than 315mhz DDR on R300 then the above claim isn't unrealistic or is it?

 

[Doomtrooper]

I can only say from the riddles from my 'contacts' ...from the Radeon Maxx on Ebay, from the John Carmack comments on multichip cards coming in the new year that....a MAXX card is coming

BTW ATI, a very deserved congratulations on a excellent product...looks like my 8500 will be up for sale very soon

Maybe this will finally end the poor gaming Image and always the Bridesmaid, never the bride ATI has had over the years.

 

[fanATIVdiot]

"256 simultaneous units" say, IMO, multi-chip solutions. Scalability to 256 chips?

Man would that kick ass. That would likely have to be on a multi-machine "render farm" configuraion, of course.

I would have happily answered this about 15 minutes ago had I known that B3D forums do not work with fully compliant W3C browsers (Opera).. <wink>...

in regards to the term "256 simultaneous units", it indeed does refer to up to 256 R300 VPUs operating in parralle.


From ATI's R9700 Whitepaper:


White Paper - RADEON 9700 Architecture and Key Features


Introduction:

The RADEON 9700 is the most advanced graphics processor ever created. With 107 million transistors, it is a completely new architecture designed around the concepts of high bandwidth, parallelism, efficiency, precision, and programmability. The performance of this new architecture is staggering, more than doubling anything on the market today in every category:


From ]-[ardOCP's 9700 (p)reveiw:

"Parallelism is a bit more ambiguous though and a few more conversations with engineers has uncovered the fact that the R300 VPU is capable of being utilized in a multi-VPU configuration with up to 256 R300 VPUs being utilized on a single "card". I know the first thing going through many of your heads is the bogus pictures we saw of the "8500 MAXX" product in the last couple weeks. When directly asked, ATi has once again stated that there are no plans to build a VidCard with two or more VPUs at the current time although they could certainly do so. ATi did mention that there were certain government and military application that could take advantage of such an architecture and they wanted to be able to be viable in that market."





---------------------------------------------------------------

Matt,

How much bandwidth do you get with 275mhz DDR on a 256bit bus?

256/8*2*275=17600

256bit bus
8bits per byte
times 2 = DDR
275mhz


Edit: 1st post followed by 1st edit, then a 2nd and lastly a 3rd to edit the Edit that was supposed to be the Edit =)

 

[Matt Burris]

MattB,

On final release Parhelia has with 275mhz DDR 17.6GB/s bandwidth. If ATI uses higher than 315mhz DDR on R300 then the above claim isn't unrealistic or is it?

Ahh, I was over at Matrox's Parhelia Technology site to confirm if it is 20GB/sec bandwidth, and it was confirmed there. Guess I should've known better, companies like to market before they tell the truth.

 

[Ascended Saiyan]

This is the first time in recent memory that a graphics product has actually exceeded my expectations.

I am still flabbergasted at the engineering feat of ATI producing a 0.15 micron part with 110 million transistors, 8 full floating point pipelines, and getting 300+ Mhz out of it.



I am hugely impressed by the AA scores. Final judgement to be reserved of course, until we can properly judge the compatibility and image quality of their AA and anisotropic implementation. But this is the first time in a long time that I have been really excited about a new part.

And the first part with a $400 MSRP that is, imo, actually worthy of that price tag.

I agree but are you flabbergasted because it exceeded your expectations or because it exceeded the expectations of the rumours that've been floating around which I was impressed with.

 

[MfA]

Well they must have given up on MAXX then, because a 256 frame delay is not acceptable by any stretch of the imagination (Or they were just being facetious.)

 

[Ailuros]

MfA,

I'm hoping for a comment on the SV II algorithm from the data so far. Anything?

 

[MfA]

Seems they are just doing multisampling and using their pixel cache to perform framebuffer "compression" (compression between quotations because it likely only removes the redundancy of having samples with identical color caused by multisampling). Should be great for bandwith, but the sheer size of the framebuffer would still prevent you from going far beyond 4X (apart from that it seems likely that if you want more than 4 samples per pixel it will take more clock cycles, they probably optimized for that case). The random sampling positions are nice, but I doubt at 4X you can do much better than what 3dfx used.

The extreme performance advantage is probably mostly a result of the extra bandwith, and perhaps also they dont share the disadvantage of the GF4 where multisampling impacts the performance of the early Z rejection.

 

[Ailuros]

Thanks a lot MfA (no I didn't figure heh ).