R520 Definitely 16 Pipes, Confirmed at AnandTech

[SickBeast]

Originally posted by: KristopherKubicki
I removed the comment about Ultra-High resolutions and the fill rate. I tried to push too many ideas into one clever sentence and now I've just confused everyone and even myself a little. I also just got my wisdom teeth pulled today and I am hopped up on so much Codine that everything is a little too fuzzy to begin with anyway.

Kristopher

What exactly did you mean when you referred to the frame buffer?

Eat some soup and yogurt, it's good for you.

 

[Soccerman06]

OMG the X1800XT is going to be the same price as the 7800GTX after a month, so stop whining. The 7800GTX was $600 or more at launch and dropped roughly 17%. Now that the X1800XT/XL have 512mbs might keep it at roughly $525 (give or take) but there are always bundles that will make the extra money worth it.

Remember you cant say that the R520 is a OCed X850XT because its not. The memory is larger and faster (1.26ns vs 1.6ns), the core is completely different, the pipes have been reworked (completely?), theres SM3.0, and all the other HD and AA/AF stuff.

Also, the reviews havent been released yet, so dont bash until you have seen theyre released.

 

[ZobarStyl]

Given ATi's penchant for low-availability launches of their upper tier cards, I doubt you'll see the price drops that the 7800GTX had. We're likely to see day-of-launch sales of X1800XT (nVidia basically has forced them to do this) but I doubt they'll stay in stock long, meaning the price won't drop so quickly. I think this is why people are worried about the price...last gen it wasn't uncommon to see the XTPE for sale at 100 dollars over retail from reputable vendors for quite some time after launch. Let's hope ATi worked out their yields and can keep this gen on the shelves.

 

[Cookie Monster]

You know the price difference from 1.26ns from 1.6? Its ALOT.
ATi retaped there core 3 times. You expect them to drop the prices?, unless you wanna see ATi losing even more money, how about no.

Like i said all things considered (Performance equal or maybe little bit faster than a 7800GTX), Heat, power, noise etc.

The 7800GTX at 480 is way more attractive than a X1800 XT at $599 or even more due to limited quantity of the higher ATi parts. Yea and thats bundles included.

Nvidia only released it as 599 because there was NO compeition and it was a chance to make lots of profit out of the situation.

Or would you buy the X1800 XT? its either your loaded with money, or just an ATi fanboy that is all.

Its not bashing either. Hey, im interested at all this internal 512bit bus ring and HDR +AA, but at a consumers point of view, economically, i dont see how a X1800XT will be better off than a cheaper 7800GTX.

 

[ronnn]

Originally posted by: Gamingphreek

Originally posted by: ronnn

Originally posted by: Gamingphreek

3. Those clockspeeds while cannot amount in some cases to the G70's Pixel Pipes and what not, they will still compete. Remember the Pixel Pipes right now for ATI, as shown, are just barely holding them back in terms of Fill Rate. They actually beat the G70 in terms of Memory Bandwidth. The downside to this is, as we have seen, low yields, and more than likely higher power usage. More pipes will always (AFAIK) be more efficient than ramping up clockspeeds.

-Kevin

Several sources including Tom's have state the yields are fine now. The places pushing the low yields are the same places that kept repeating that silly 32 pipe rumour and they are covering their tracks. As for power, I haven't read anything anywhere on that - but I thought that generally a smaller process meant lower power consumption? I agree that more pipes should be a big advantage.

I wouldn't be so bold as to say yields are fine. 700mhz memory is pushing the limits, not to mention a brand new 90nm technology running at 600mhz. I would say yeilds are probably acceptable, but not fine. THen again, i know no more than you so that is merely speculation on my part.

While the smaller process should yield lower power, AFAIK, it takes significantly more power to ramp up the clockspeed (inherently ramping up voltage) than to simply put more pipes in. They give the same end result, but i would be more inclined to go with more pipes than a higher clock frequency.

-Kevin

For the record.... i wouldn't trust toms hardware, they are easily bought out for biased reviews.

No I wouldn't trust Tom's either, but the same goes for vr zone and the inquirer that have been pushing low yields and 32 pipes. So are yields "acceptable" or "low"? If pushing up clock speeds comes with 90mm tech, than faster clock speeds seems reasonable to me. I will wait to see the results myself before I decide which I would go with.

 

[Byte]

looks like ATi might pull an FX...ouch!

 

[mthomas988]

Originally posted by: Byte
looks like ATi might pull an FX...ouch!

Not really...for R520 to deserve such a comparison, it would have to come out with abysmal PS3.0 performance, crappy image quality, a crippled memory bus, and a dual slot cooler that sounds like an electric drill.

 

[biostud]

Originally posted by: Sithtiger
Wait a sec, you go under NDA tomorrow? Do you mean your not under NDA now but you go under an NDA tomorrow? If that's the case then spill out as much as you can now! It's still legal!

you don't think that's just what they did?

I'm pretty sure they don't feed you the juicy stuff until the NDA has started.

 

[RichUK]

This 512bit Ring BUS thingy that ATI has announced.. Is this comparable to the token ring topology, as in a data flow model.

Or is this just a similarity in BUS names ??

 

[gtx4u]

Originally posted by: compgeek89
Well its been out for 3 months..., not 6

if ATI has a whole bunch of x1800xt in stock at compUSA, and Bestbuy and newegg.com at OCTOBER then iz 3 months late, if not then they really are late...

 

[ronnn]

Originally posted by: gtx4u

Originally posted by: compgeek89
Well its been out for 3 months..., not 6

if ATI has a whole bunch of x1800xt in stock at compUSA, and Bestbuy and newegg.com at OCTOBER then iz 3 months late, if not then they really are late...

Being early or late will not effect the performance. Lets just hope it is a very pleasant surprise.

 

[BFG10K]

Ok... umm yes it was. WHile it was based onthe older architecture, it was HEAVILY tweaked; as i said for all intents and purposes it was a new architecture.

No, it wasn't a new architecture. GF1 was new, GF3 was new and GF5 was new. Anything in between that is simply a refresh which includes the 7800.

The NV40 was simply a heavily tweaked version of the NV30 with better specs and better shader design but fundamentally it was still the same card. The 7800 builds on the 6800 but again it still has its foundation built on the NV30.

Likewise for ATi R100, R200 and R300 were new but anything in between were refreshes, which includes R420. However I think it's disingenuous to call the R5xx a refresh because of brand new shader units (SM 3.0, FP32 etc), a brand new memory architecture and possibly other things that we don't know about yet.

It may not be for ATI, however, Nvidia who had gone for 3 years with shader replacement, had finally left it out. That is a huge step forward for Nvidia.

Sure but that doesn't make a new architecture. Also in addition to hardware tweaks you can bet drivers and compiler technology improved as well.

You know as well as anyone here that the FX series were as much a 4 pipe card as they were an 8 pipe card. 4 Pipes in single texturing and 8 in multi-texturing.

No, not 8 in multi-texturing: 4 dual-textured pixels per cycle. There's a big difference to that and having 8 pixel pipelines.

I think we've had this discussion before and it was pointed out to you repeatedly that 8x1 is not the same 4x2. You can't interchange pixel and texel capability just because paper multiplication would suggest it.

And BFG, jesus, you've yelled at Rollo enough times before

This isn't so much as a yell at Rollo as it as against everyone who claims ATi parts are refreshes while nVidia parts are brand new. If you're going to make sweeping claims then you'd better be consistent with both vendors.

 

[nemesismk2]

<double post>

 

[nemesismk2]

I don't think ATI will have a launch like Nvidia did for the 7800 GTX, IMO if ATI can't deliver sufficent stock on the day of release then the launch should be considered a complete failure!

 

[otispunkmeyer]

man....good to see they are using 512mb of ram on the top two cards, should be interesting at high res

First of all, ATI's traditional core design can do "more" per clock cycle (at least on the R420 design) than NVIDIA. Secondly

how do you work that out?

6800U was much slower on the core than the competing X800. yet they performed similar


cant wait for the reviews. theres saying that says theres no replacement for displacement.

well actually there is....its called forced induction.

Nvidia have gone for the 8 Litre V8, Ati have gone for the 3litre Twin Turbo V6

 

[BenSkywalker]

Kristopher

Thanks for the information, one thing-

600MHz core clock, with a 256-bit memory bus connected to 512MB of GDDR3 memory clocked at 700MHz. The 600MHz core clock will give it a lower fill rate than the GeForce 7800 GTX (24-pipes at 430MHz),

The R520 will have a higher pixel and texel fillrate then the 7800GTX. The fill is limited to the ROPs, not the ALUs.

 

[Keysplayr]

Originally posted by: BFG10K

The ROP's usually are idle waiting for one of the 24 piplines to finish shading/processing pixels. There is no 1 to 1 ratio for piplines to ROP's. Those 16 ROP's are not yet near saturation from 24 pipes and is said to be sufficient up to 32 pipes if need be. This is a common misconception.

That doesn't answer the question at all. The framebuffer can be as big as it likes (up to the physical limit of the card of course) and has nothing at all to do with the fillrate, ROPs or pipelines.

I personally can't see the logic behind the original claim.

You don't have to take my word for it ( as if you ever would ) just do a quick re-read of the AT article.

 

[Dribble]

The fact they have 3 different cards all using the same 16 pipe core (so binning very carefully to pick out the fastest) and have gone for the fastest memory money can buy must mean they are really having to try very hard to get something to compete with nvidia. Nvidia must think so too because I suspect having managed to sneek a look at ati's cards they've decided to not even bother with an "ultra".

While I'm sure the fastest X1800 will give the GTX a run for it's money it will also be rare, loud, expensive and will have no o/c head room. Hopefully in the next year ati can get the production process sorted out and then we'll see some 32 pipe versions of this chip which is what ati was trying for in the first place.

 

[BFG10K]

You don't have to take my word for it ( as if you ever would ) just do a quick re-read of the AT article.

The AT article has been changed to remove the comment about the framebuffer. In either case your explanation had absolutely nothing at all to do with said issue.

 

[otispunkmeyer]

Originally posted by: GTaudiophile
Does anyone think ATi is SO STUPID as to release a GPU that is not only 6 months late, but costs as much or more than the competition, knowing that it will barely outperform the competition...if at all?

i dont know

Nvidia did to an extent with NV30, but i highly doubt anyone, even S3 and XGI, will be making a hash job like that ever again

this R520 will probably easily keep pace with the 7800's

 

[otispunkmeyer]

Originally posted by: BFG10K
I'm sorry, the two posts above are nothing more than blatant double standards.

Your example: 4/2, 4/2, 16/6, 24/8 architectures.

Pipes now it is? But your above comment took issue with "tweaked shaders".

If it's pipes how come the R420 "was giving you nothing new" when it went from the 8x1 to 16x1?

I don't think I'm alone in thinking a 16 pipe card in Q4 2005 is somewhat of a letdown.

What about two 4 pipe cards in 2003/2004 when the competition had double that amount?

6800 was, for all intents and purposes, a new architecture.

No it wasn't. It was simply a tweaked version of the previous hardware.

Additionally, the 7800 also was revolutionary for Nvidia because they ditched shader replacement.

So they caught up to ATi. How is that revolutionary?

thought it was built from the ground up to work with 32bit precision, something the NV3x was not?

NV3x was multitexturing hardware? , NV40 wasnt.....i dont understand what you mean, please explain

EDIT... NVM seen you previous post where u explain

 

[Keysplayr]

Originally posted by: BFG10K

Ok... umm yes it was. WHile it was based onthe older architecture, it was HEAVILY tweaked; as i said for all intents and purposes it was a new architecture.

No, it wasn't a new architecture. GF1 was new, GF3 was new and GF5 was new. Anything in between that is simply a refresh which includes the 7800.

The NV40 was simply a heavily tweaked version of the NV30 with better specs and better shader design but fundamentally it was still the same card. The 7800 builds on the 6800 but again it still has its foundation built on the NV30.

Likewise for ATi R100, R200 and R300 were new but anything in between were refreshes, which includes R420. However I think it's disingenuous to call the R5xx a refresh because of brand new shader units (SM 3.0, FP32 etc), a brand new memory architecture and possibly other things that we don't know about yet.

It may not be for ATI, however, Nvidia who had gone for 3 years with shader replacement, had finally left it out. That is a huge step forward for Nvidia.

Sure but that doesn't make a new architecture. Also in addition to hardware tweaks you can bet drivers and compiler technology improved as well.

You know as well as anyone here that the FX series were as much a 4 pipe card as they were an 8 pipe card. 4 Pipes in single texturing and 8 in multi-texturing.

No, not 8 in multi-texturing: 4 dual-textured pixels per cycle. There's a big difference to that and having 8 pixel pipelines.

I think we've had this discussion before and it was pointed out to you repeatedly that 8x1 is not the same 4x2. You can't interchange pixel and texel capability just because paper multiplication would suggest it.

And BFG, jesus, you've yelled at Rollo enough times before

This isn't so much as a yell at Rollo as it as against everyone who claims ATi parts are refreshes while nVidia parts are brand new. If you're going to make sweeping claims then you'd better be consistent with both vendors.

This sounds like it's based entirely on your opinion. Link to a credible source that states that a 7800 is a tweaked 5800.

 

[Keysplayr]

Originally posted by: BFG10K

You don't have to take my word for it ( as if you ever would ) just do a quick re-read of the AT article.

The AT article has been changed to remove the comment about the framebuffer. In either case your explanation had absolutely nothing at all to do with said issue.

It certainly did. At least it was to the post I responded to. Someone was thinking that the number of pipelines was supposed to match the number of ROP's. If you had less ROP's than pipelines, there should be a bottleneck. Hence the 7800GTX should be bottlenecked.
And the R520 has a chance in hell. So, I provided information to state otherwise. If you got something else out of that post than you should have, that's your business.

 

[otispunkmeyer]

Originally posted by: BFG10K

Ok... umm yes it was. WHile it was based onthe older architecture, it was HEAVILY tweaked; as i said for all intents and purposes it was a new architecture.

No, it wasn't a new architecture. GF1 was new, GF3 was new and GF5 was new. Anything in between that is simply a refresh which includes the 7800.

The NV40 was simply a heavily tweaked version of the NV30 with better specs and better shader design but fundamentally it was still the same card. The 7800 builds on the 6800 but again it still has its foundation built on the NV30.

Likewise for ATi R100, R200 and R300 were new but anything in between were refreshes, which includes R420. However I think it's disingenuous to call the R5xx a refresh because of brand new shader units (SM 3.0, FP32 etc), a brand new memory architecture and possibly other things that we don't know about yet.

hang on a sec there man

you say NV40 and G70 are just refreshes of NV30 (heavily tweaked)

then you say that R520 isnt a refresh of R420 becuase of the new SM3 shader units and FP32

but hang on NV30 was SM2, and wasnt at all suited for 32bit, (it could run either 16 which was prefered, and couldnt run 24bit like ATi so had to run 32bit)

NV40 was built to handle and work with 32bit as the prefered operation, it also included SM3 and lots of other things like HDR

so ATI goes from 24bit and SM2 to 32bit and SM3 and its new
NV goes from 16bit/crappy 32bit and SM2 to SM3 and 32bit and its not new? they even changed the way they did AA and AF

ill agree though that G70 is nothing more than NV40, reworked and with a few extra whistles

and surely if any manufacture has a brain, they'll use their past work, and what they've learned as a base for progress, starting a fresh takes too long

so basically somewhere in there designs they have circuitry that is basically a refresh of their very first GPU

 

[BFG10K]

Link to a credible source that states that a 7800 is a tweaked 5800.

The shaders still operate on the core SM 2.0+/FP32 capabilities laid down by the NV30. Apart from FP blending for HDR, tweaking and better specs the architectures are fundamentally the same.

It certainly did. At least it was to the post I responded to.

Nope. Read the very first question he asked which you yourself quoted. I don't know what you were responding to but it certainly had nothing to do with framebuffer sizes.