The LAST R600 Rumours & Speculation Thread

[CJ]

It seems that today some more info has surfaced.

R600 will be launched at Cebit. The press will get to see the R600 at the end of February. One week after the Cebit the R600 will be available for everyone to buy. Claims are that the R600 is 5 to 10% faster in DX10 applications than the G80. Note the DX10 claim and no mention of DX9 performance compared to G80.

Price will be $599 and only the 9" PCB version will be out in retail channels. The 12" PCB version will be OEM only.

Scheduled for a launch at the end of April is RV630 as a replacement of the X1950Pro. nV will launch the 8600GTS (G84) one month earlier in March as a replacement of the 7900GS and 7950GT. A few weeks later, mid-April, nV will release the 8600GT with/without HDMI / with HDCP, as a replacement of the 7600GT.

In the lower end AMD will introduce the RV610. First samples have arrived from the factory and they are looking good. RV610 will be the replacement of the X1650 series. Around that time nV will also release 8500GT HDMI (G86 GDDR3) in versions with and without HDCP (G86 DDR2). At the end of April they will also release 8300GS (G86 DDR2) as a replacement of the 7600GS/7300GT.

There you have it.

 

[Geo]

I think a dart board would be at least as accurate for DX10 performance comparisons in March/April, at this point. But maybe that's just me.

Edit: All I mean by that is there is a 3 way transition going on here, any one of which would be a major transition on its own. OS and OS driver model + API + gpu architecture. I think it highly likely that DX10 performance reports are going to be fluid as hell until at least mid year or so as everybody gets a handle on it. . .

 

[Anarchist4000]

And just what the heck are they using to determine performance in DX10? What exactly is the source of that information? And don't tell me it's something that LVL505 site posted again.

We're lucky to get any information at all on R600. And with G80 Vista drivers just recently being leaked, the odds of someone getting a R600, vista drivers for G80, and being able to compare them is out there a bit.

 

[Arun]

And just what the heck are they using to determine performance in DX10? What exactly is the source of that information? And don't tell me it's something that LVL505 site posted again.

CJ is a respected member of the forums, and he has leaked plenty of stuff that turned out to be perfectly accurate in the past - with some exceptions, but that's how these things work.

I also think the performance numbers for DX10 is kind of a random estimate at this point, especially so when nobody knows what aspects of DX10 will be stressed by first-gen games. But the rest of the information "feels" right - that is, I've got zero idea if it's correct or not, but at least much of it makes sense, and the source has some real credit!


Uttar

 

[CJ]

And just what the heck are they using to determine performance in DX10? What exactly is the source of that information? And don't tell me it's something that LVL505 site posted again.

Either you believe me or not. It's all up to you.

 

[IbaneZ]

It seems that today some more info has surfaced.

R600 will be launched at Cebit. The press will get to see the R600 at the end of February. One week after the Cebit the R600 will be available for everyone to buy. Claims are that the R600 is 5 to 10% faster in DX10 applications than the G80. Note the DX10 claim and no mention of DX9 performance compared to G80.
:

5-10% faster than G80 in DX10? If that's true, I'm not that impressed.

G81 vs R600 should be a close call then.

 

[nAo]

laims are that the R600 is 5 to 10% faster in DX10 applications than the G80. Note the DX10 claim and no mention of DX9 performance compared to G80.

Should I start to worry now or should I wait a bit?

 

[Arun]

G81 vs R600 should be a close call then.

Indeed... Given the timeframes involved, you'd expect the GDDR4 G80 (+optical shrink to 80nm?) to be ready then or slightly afterwards. Even a very conservative estimate of 650MHz core clocks with 1.1GHz GDDR4 would imply 15% higher performance, so hopefully AMD is preparing to fight that, rather than a 5 months old part. I'm sure they realize that though, so I wouldn't be too worried for them, personally.


Uttar

 

[Jawed]

Edit: All I mean by that is there is a 3 way transition going on here, any one of which would be a major transition on its own. OS and OS driver model + API + gpu architecture. I think it highly likely that DX10 performance reports are going to be fluid as hell until at least mid year or so as everybody gets a handle on it. . .

Hey, hopefully soon we'll have 3DMk07 to guide us on our way.

Jawed

 

[icecold1983]

is the g80 bandwidth or fill rate limited?

 

[Anarchist4000]

I'd agree that timeframe looks reasonable and most of the figures seem right except that 5-10% DX10 performance number. Why would there be DX10 numbers and no DX9? Getting DX9 numbers seems like it would be substantially easier than the DX10 numbers.

Although details were vague enough, and if no high levels of AA/AF were involved, those numbers might make a fair amount of sense. That performance difference could grow considerably when AA/AF are applied and any "benchmarking" that may or may not have occurred likely wouldn't have involved AA since G80, AA, and Vista haven't been getting along very well.

Could we be looking at another 1800->1900 style release. With the 1800 releasing a new memory controller followed by the 1900 with a hefty boost to the processing capabilities along with a smaller process?

If those order numbers for G80 chips are accurate Nvidia is either planning on selling a ton of cards or we're looking at a GX2/GX4 style refresh again. Bandwidth wise GX2 would line up a lot better with the estimates for R600 although the bandwidth doesn't appear to be doing much in DX10.

 

[AlexV]

Why would DX10 bandwidth be different from DX9 bandwitdth?I think that`s not the case-bandwidth does for DX10 what it did for DX9,8, 7 etc. before it.

As far as possible ways of testing DX10 I`m thinking that aside from the examples given in the recently released SDK and home-baked stuff not a lot is around, but those should give ballpark figures for the moment just ok.

 

[CJ]

According to the source, the ATI info came straight from AMD, while the nV info came straight from nVidia. He mentioned that nVidia also had some 'interesting blurps' about R600, which naturally should be taken with a grain of salt. Supposedly they said R600 had heat, power and cooling issues. When asked about G81, they're response was "We are waiting for ATI to launch R600".

 

[Rangers]

Supposedly they said R600 had heat, power and cooling issues.

Sounds like FUD for "Teh Powa".

I looked up Cebit and it's 15-21 March? I thought it was at least like first week March. ATI is so damn slow. It's ridiculous.

I hope at least that NDA's are up end Feb, so that we can read about it, though I kind of doubt it.

 

[Pete]

Hey, hopefully soon we'll have 3DMk07 to guide us on our way.

And here I was surmising that's what CJ's sources were referencing.

 

[jamis]

Claims are that the R600 is 5 to 10% faster in DX10 applications than the G80. Note the DX10 claim and no mention of DX9 performance compared to G80.

Hmm..if that's true G81 should easily beat it. It's looking very much like x1800 all over again, nice but too late, only this time NV's refresh is coming earlier. Wonder if it's going to be the same with the mainstream parts as it were with X1600-series.

 

[zgemboandislic]

All in all, things are definitely Hunky Dory for AMD.

 

[Bouncing Zabaglione Bros.]

Hmm..if that's true G81 should easily beat it. It's looking very much like x1800 all over again, nice but too late, only this time NV's refresh is coming earlier. Wonder if it's going to be the same with the mainstream parts as it were with X1600-series.

At this stage of the game, this could easily be misinformation. It's just as likely to be AMD trying to make Nvidia aim lower with G81.

With all those extra transisitors, a lower process node, all that memory bandwidth and the 512bit memory controller, there's got to be something more in R600 that we don't yet know about.

 

[zgemboandislic]

At this stage of the game, this could easily be misinformation. It's just as likely to be AMD trying to make Nvidia aim lower with G81.

With all those extra transisitors, a lower process node, all that memory bandwidth and the 512bit memory controller, there's got to be something more in R600 that we don't yet know about.

That's what the majority thought about R520. It was also on a lower process node, ring bus etc. In the end, it turned out that it could only match G70.

 

[no-X]

The weak point of R520 were poor initial OpenGL drivers. Today R520 is 1/3 faster than 7800GTX and comparable to GTX-512. The problem wasn't R520 core itself, but slow drivers and loud cooler. I hope ATi will not repeat this mistakes...