Ati following in nvidia's footsteps ?

[Jawed]

And the 100% practical improvement implied by your 512-bit comment? Do you think that within reach, or getting greedy there?

Just being greedy!

And where does most of that come from? Being able to share better between units without having to shuttle data to/from main memory?

The only data that's going to be shuttled to-from memory is vertex data, texture data (e.g. render to texture) and the frame-buffer. I'm not thinking of "intermediate" data, such as the post-transform vertex cache (buffer) as being stored in main memory.

But internally there are all sorts of data paths, some are shared by multiple units of the same type, others are effectively point to point.

I should have said "various buffers, set-up engines and caches".

I'm just saying all this stuff to see if we can get some more informed opinion...

Jawed

 

[Geo]

Well, whatever it is, I look at the Xenos piece and the kind of conversations that must have gone into it, and am greatly comforted when afterwards Wavey is still making allusions of the kind above.

 

[DemoCoder]

I thought the memory controllers on modern GPUs already do read/write reordering for coherence.

 

[dizietsma]

Well it looks as if Ati's bus, whatever it is, could be a way out of lack of Samsung not being able to release, seemingly, faster memory than 1200Mhz effective at present. The 7800 uses exactly the same speed as the 6800 and X800XT did all those months ago.

Samsung have had up to 1600Mhz listed on their web site for a long time, I even saw a press release saying they would have 2000Mhz GDDR3 by the end of 2005. Well it's July now !

GDDR3 should be about 1.8v but if you look at the Samsung pages with different revisions of the specs it does from 1.9v up to 2.0v and then I think 2.1v before going down to 2.0v ( from memory ). That sort of playing with voltages seems to indicate poor yeilds as far as I can see.

So a nice efficient and clever memory bus to get the most out of 1200 or 1400Mhz memory would be a good thing indeed. It does not sound as sexy as 32 pipes no doubt, but if it delivers the goods then hopefully it will show that number of pipes are not the be all and end all for performance.

Basically, if they release it on AGP then it is a one horse race for me.

On a slightly different note, it was reported recently that nvidia would not be putting SLI on their lower range 7 series cards. Considering the new driver that allows Sli on the 6600 cards which do not have the Sli bridge connector, but with reduced performance, is this a way of nvidia making sure that ordering of performance definitely goes this way

7800plain
7800plain Sli
7800GT
7800GT Sli
7800GTX
7800GTX Sli

without any overlap and hence pinched sales from the next one up?

 

[_xxx_]

The wheels on the bus go round and round. . .

You are alluding to the previously discussed "token-ring" style of bus? Wasn't this more related to job scheduling rather than the actual memory interface? Or are you agreeing with the speculation at the time that this was a way of having extremely efficient memory access almost the whole time?

...one ring to rule them all. With un1f13d memory busszz

 

[rwolf]

If the respin does work ok. The R520 might cost a fair bit less than the G70 to produce. They might price it like the 800XL.

I think the 90nm process is much higher volume with 12inch wafers compared to the 130nm SOI process they were using for the 800XT and 850XT.

 

[ANova]

If the respin does work ok. The R520 might cost a fair bit less than the G70 to produce. They might price it like the 800XL.

That would be a nice change of pace.

 

[HaLDoL]

If the respin does work ok. The R520 might cost a fair bit less than the G70 to produce. They might price it like the 800XL.

It's not logical to price your top performing card like the 800XL.
The costs of the respin itself (and the delay in revenues) must be compensated also.

 

[saf1]

Let's wait to see what R520 looks like and when it arrives. There's no point comparing G70 to R420 and then proclaiming that Nvidia is way out in front. You should be comparing G70 to R520. Then we can tell who did the best job.

If R520 launches end of July with availabilty end of August (ie "late summer"), the two month head start for G70 will be negligable. All the people buying now rather than waiting a couple of months either can't wait because they need to build systems now, or would be buying Nvidia regardless of whether they launch a couple of months earlier or later.

I'm not sure you "should" be comparing the G70 to the R520... I do not think anyone has a clear understanding of the G70's history in relationship to the product road map - NV48, 50, etc. I'm willing to bet a soda that it is more inline with the R4xx series than the R5xx. I always hold the right to be incorrect though

 

[AlexV]

I will rape a herd of donkeys if ati`s high-end, rather troublesome by the looks of it, souped up mega GPU is priced like an XL.No way that is going to happen, unless God has finally taken heed to the pain of hardware enthusiasts and has added an eleventh commandment-Thou shalt not overprice like hell thy products, for that is evil

 

[trinibwoy]

I will rape a herd of donkeys

Sig!!!

 

[ANova]

I will rape a herd of donkeys if ati`s high-end, rather troublesome by the looks of it, souped up mega GPU is priced like an XL.No way that is going to happen, unless God has finally taken heed to the pain of hardware enthusiasts and has added an eleventh commandment-Thou shalt not overprice like hell thy products, for that is evil

Some people are thinking the R520 may be priced at $500-550, rather then $600 like the 7800 GTX.

 

[Geo]

Well, given they need 512mb of the fastest memory for the framebuffer to enable the 16x aa their new bus makes possible, I'm thinking not.

Oops. I've said too much.




[I AM kidding]

 

[rwolf]

I will rape a herd of donkeys if ati`s high-end, rather troublesome by the looks of it, souped up mega GPU is priced like an XL.No way that is going to happen, unless God has finally taken heed to the pain of hardware enthusiasts and has added an eleventh commandment-Thou shalt not overprice like hell thy products, for that is evil

Run donkeys.

 

[Bjorn]

Some people are thinking the R520 may be priced at $500-550, rather then $600 like the 7800 GTX.

That's a far cry from the price of the XL though. And i seriously doubt that the price of the 7800 GTX will be 600$ when the R520 arrives. Though i guess that that depends on the price of the R520.

 

[AlexV]

The GTX is more like 550$, from what I`ve seen.Heck, it`s 514 euros where I live, and this country is anything but cheap.500-600 seems to be the spot for 256mb high-end cards, but i don`t think the XL fits in the picture, unless someone makes an uber-expensive XL.

IMO, the only imaginable chance of it being in the 200-300 range is for it to suck so bad it forces ATI to use dumping prices-how likely is that now?

 

[Nappe1]

If that were the case they would just release it on the fastest available. Sometimes discussions about a memory bus isn't necessarily just about the memory interface.

Did anyone else happen to notice Wavey's italicized bits? That's usually one of his clever little ways of dropping a biiiiig clue....

[offtopic]
funny thing... I had Parhelia's "number specs" several months before release on my signature at muropaketti.com (biggest finnish HW site, around 40,000 users.) and nobody didn't even noticed. And I used to post a lot back then.

I originally thought putting same here, but thought that being too risky.[/offtopic]

My comment? well, IF R520 still shows up in late summer and is as fast as G70, but priced a bit lower, then we'll see some serious real driver tweaks from both sides. (3 years ago sam situation would have mean good price competition, but that's gone.)

IF R520 is inferior to G70, then we'll see some quick respins from that core to secure some of the markets (and possible driver optimizations again.) Maybe that would accelerate ATI's rather slow R&D (lately... since R300) quite bit too.

 

[digitalwanderer]

What if the R520 blows the doors off the G70? :|

 

[Bouncing Zabaglione Bros.]

What if the R520 blows the doors off the G70? :|

Nvidia has an advantage in that they are shipping now with a really good looking part. ATI may have to drop prices in order to get that marketshare back, and they should be able to afford to do so if their yields are good, as they will be getting a lot more chips per wafer than Nvidia.

However, if R520 is really great, ATI may be able to keep their prices high by showing that R520 is the best thing out there by miles.

 

[AlexV]

How likely is that?What are the odds of either NV or ATI blowing the other out of the water?Even with the NV3X it wasn`t a blow the doors off situation in terms of immediately apparent performance.