RV560/570 Gemini roadmap

[Dave Baumann]

Why two bridges? Need it for SuperAA bandwidth? But then why not make one twice as wide?

They look about the same size as an SLI connector, don't they? MB manufacturers are already bundling these with motherboards, so why not make use of the same connector?

Edit: Another point could be as a way of chaining even more boards together by staggering the connections on each board.

 

[satein]

They look about the same size as an SLI connector, don't they? MB manufacturers are already bundling these with motherboards, so why not make use of the same connector?

If it needs 2 of the same SLi bridge (hopefully, the bridge connector thing is standard not a patent), may it be possible that ATi solution needs more pin-out to do a dongle-less crossfire? Thus, may it be speculated as wider bus of the bridge and more bandwidth too?

 

[SugarCoat]

They look about the same size as an SLI connector, don't they? MB manufacturers are already bundling these with motherboards, so why not make use of the same connector?

Edit: Another point could be as a way of chaining even more boards together by staggering the connections on each board.

The SLI bridge is patented by Nvidia, at least the ones bundled with Nforce 4 boards if i remember correctly. I dont think ATI wants to pay royalties.

I would assume that two would mean they are using one to upstream data, the other for downstreaming.

 

[Dave Baumann]

The SLI connector itself is just two sockets wired between each other, not sure that that itself can be patented. Its also been bundled with some non-NVIDIA boards as well.

I'd doubt that one would be for upstream and another downstream as most of the time, under ideal operation, the traffic would be in one direction. Besides, if that were the case I'd guess that the stickers on the connectors in the above image wouldn't both be the same way around.

 

[SugarCoat]

I wonder if it would have anything to do with being able to run Crossfire in multi-monitor then.. or perhaps their physics implimentation (doubt it). If it doesnt have a specific purpose, in that it solves some issue, that would be a strange design decision to me. But then again so was the dongle to many of us.

 

[Geo]

Getting multi-mon to work with multi-gpu in a more seamless manner would certainly be a goodnesss. Particularly when you're aiming at a high-end market in the first place.

We know that SuperAA has generally kicked SLI-AA's heinie on performance. . .I still wonder if it just comes down to providing more non-pcie BW for that.

 

[Rys]

Coming to market at the end of July? That's pulling stuff forward two months from some other reports of September, and is at the optimistic end of ATI's February roadmap (which was July/August). Be nice if it works out that way.

The Gemini with RV530 is end of July. He's looking forward to an RV560/RV570 version, of which silicon hasn't arrived yet (end of June though for sampling).

 

[demonic]

excuse the ignorance here, but gemini is 2 gpus in 1 core right?

so if you crossfire them, this is an actual 4x Crossfire, than the psudeo 4x SLI that Nvidia is touting?

 

[no-X]

Gemini = MAXX = 2 GPUs / one graphic card.

 

[Geo]

The Gemini with RV530 is end of July. He's looking forward to an RV560/RV570 version, of which silicon hasn't arrived yet (end of June though for sampling).

Which would then suggest they've slipped a month from the Feb targets. . . so sampling in June (rather than May), production in Aug/Sep.

 

[CJ]

Which would then suggest they've slipped a month from the Feb targets. . . so sampling in June (rather than May), production in Aug/Sep.

Looks like RV560/RV570 aren't the only ones that slipped a month. RV505 was scheduled for May MP according to the February roadmap, but the RV505CE will be going into MP at the end of August and will be known as the X1250 (350/400, 32bit/64bit, 4 ps, 2vs).

 

[Jawed]

80nm?

http://www.dailytech.com/article.aspx?newsid=2889

I wonder if anyone will bother asking ATI why the mainstream parts (RV560/570) haven't been released earlier at the forthcoming financial chat.

Jawed

 

[Sunrise]

http://www.vr-zone.com/?i=3724

New corssfire bridge for 80nm cards

"* Original Pics & Text Removed *"

Looks like it revealed something not supposed to be shown this early.

 

[kemosabe]

More 80nm capacity for RV5XX.

 

[Arty]

First performance graph from ATI courtesy folks from hkepc: (Could this mean chips are nearing their sampling stage?)

They also managed to confuse themselves about Gecube pitting the RV570 as a successor to X1600 and not X1800.

 

[Geo]

Well, dunno about that. They've been talking about 4 segments, with the traditional "midrange" further segmented into hi/lo, for some time. If you look at it that way, today the mid-lo is X1600 territory, and the mid-hi is "Crazy Dave's Bargain Bin". So from that pov, 570 is pointed at retiring the X1800 parts that are hanging around. But I'd think it would be pointed at X1900GT as well, so it's interesting it's not shown instead where X1800XT is on the chart.

 

[no-X]

The last part is very strange. X1600XT > 6800GS ?? (but if I remember, CJ posted similar info some time ago ) I expected, that RV560 will be about 50% faster than RV530 and competitive to 7600GT (something like 8-1-3-x, 128bit, 550-600MHz). This graph shows RV560 only 15% faster than X1600XT... so RV560 clock will be very low or 8-1-3-x rumour isn't true (or rather the graph is stupid )

 

[Jawed]

Wow, RV560 is nothing more than a replacement for RV530, not an 8-1-3-2 part. Seems ATI still isn't interested in competing against 7600GT (and for some reason thinks X1600XT does compete). Oh dear. Unless RV570XL is that part?...

Meanwhile RV570XT looks like it will be a good replacement for X1900GT which is clearly too big, slow and expensive.

That roadmap looks like it dates from January, though, as it only refers to the 7- and 6-series GPUs on the market at the time.

Jawed

 

[Geo]

That roadmap looks like it dates from January, though, as it only refers to the 7- and 6-series GPUs on the market at the time.

Jawed

Shows P: Now on all those 80nm parts tho. So that doesn't fit. I would guess price points have more to do with what they picked to show against (and maybe some cherry picking, you can't rule that out in PR docs).

As far as competing against 7600GT goes. . .as I've been saying, the two companies are asymmetric right now in the mid-range. ATI is hi/lo in a new paradigm vs NV in the center of the older paradigm. Which creates three price/performance points for consumers, actually, if you could find yourself happy with any of the three (some people can, others can't).

 

[no-X]

Jawed: Yes, it's old. Look here