Speaking of which, where is that thing? I thought we'd have seen it by now.
AMD: R8xx Speculation
- Thread starter Shtal
- Start date
Speaking of which, where is that thing? I thought we'd have seen it by now.
Silent_Buddha
Legend
I'm guessing supply constraints have put it low on the burner. Additionally I think they are still working to polish eyefinity over crossfire, which I would imagine would be almost required if you were to try gaming on 6x monitors.
Regards,
SB
I had been thinking April timeframe for the refresh, my birthday is around then.
I had been thinking April timeframe for the refresh, my birthday is around then.
They took you off the list!
Last time they were looking for Santa he was having a sleepover at your place!
I had been thinking April timeframe for the refresh, my birthday is around then.
So what are the expectations for a refresh? Something akin to the RV790 in having a cap ring and larger die size along with ~10% higher clock and faster GDDR5?
FrameBuffer
Banned
It's seems that most people are suggesting that a 5890 refresh would be in order, however who is to say that many of the changes that occured in the 4890 (cap ring/decoupling etc) aren't already in place ? It would seem to me that ATI would have put in place any changes from the RV790 forward to the RV870. Also unless I'm behind the times (and I may just be), I have yet to see any die shots of Cypress so who is to say what's hidden (a repost from R3D):
lol
PSU-failure
Newcomer
There is one very low res die shot in AMD's papers and there's definately not more than the actual 20 SIMD
The only thing they could be hiding is a separate clock domain for these, but it would have been used since the first series, and a wider bus, but that would not be useful unless they have a higher ALU clock... (not counting the fact that Cypress seems to suffer from a relatively slow triangle setup)
Do you mean this one?There is one very low res die shot in AMD's papers and there's definately not more than the actual 20 SIMD
that is RV770 blended with some CPU die-shot... (first-gen Fusion
)
//ed.: details: http://forum.beyond3d.com/showthread.php?t=49120t&page=167
There is one very low res die shot in AMD's papers and there's definately not more than the actual 20 SIMD
The only thing they could be hiding is a separate clock domain for these, but it would have been used since the first series, and a wider bus, but that would not be useful unless they have a higher ALU clock... (not counting the fact that Cypress seems to suffer from a relatively slow triangle setup)
Correct!
That's why I can only see two routes for AMD's refresh:
- Higher clocks (as was the case with RV790) + faster GDDR5
- Another, bigger die with more of everything ...
FrameBuffer
Banned
Oh well.. so much for a 4890 Part Deux. Huddy seemed pretty confident enough to say that ATI will have the fastest solution for a majority of the year. Given that they'll (ATI) have a 3 month lead on Fermi for '10, that pretty much gives us until about the last week of August or 1st week of September for ATI to deliver a much improved part then ride out the rest of the year. 6 Months and 1 week (or 27 weeks) would be a "majority" of the year, though I suspect mid august would be the latest target to make the back to school sales.
If I'm not mistakend, Huddy only said ATI would have the fastest card, so he could just as well have been referring to the 5970.
FrameBuffer
Banned
ahh I guess your right:
http://www.bit-tech.net/bits/interviews/2010/01/06/interview-amd-on-game-development-and-dx11/5
though he does go on to say:
I would think that would mean that they would have something to combat any fermi X2 product.. maybe with the announcment that GF is ahead of shcedule with their 28nm, AND feels a bit more comfortable (though I bet TSMC might not be). It looks like it will be a fun fun year indeed.
Oh, I definitely suspect any changes required to reach higher frequencies are already there. rv790 somehow looks like an accident to me, which didn't make that much sense from an engeneering point of view (why produce 2 different chips which are so similar - I'd guess either AMD couldn't realize the clocks they initially wanted with rv770, or they never really planned to go higher until they realized it would be so close to nvidia's high end chip).
That doesn't rule out a refresh part though (using the same chip) in my books. First, while the HD5870 is quite close to the 225W limit, it's not that close as some other cards. It imho might be possible with chip binning (or maybe slightly better chips thanks to improved TSMC manufacturing too) to increase frequency some more, though I wouldn't expect more than 950Mhz. Plus, probably more important, gddr5 ram at 6gbps instead of 5gbps should be available now, which probably would make quite some difference too (assuming the memory controller can handle that frequency). Certainly it wouldn't be an earth-shattering improvement, but something with clocks like 925/1400 should offer over 10% performance improvement. Now, if nvidia has done their homework that shouldn't be enough to be as fast as fermi, but then again the HD4890 shouldn't have been almost as fast as the GTX 285 neither, rivaling GTX 275 (and GTX 280), certainly not from nvidia's point of view
. If it doesn't have a chance to be competitive with the slower fermi part, maybe we might not see a HD 5890 though would be my guess.Or, if nvidia is willing to go for 300W TDP with a single gpu card (big if), AMD might be too, in which case I'd guess 1Ghz core clock should be easily doable.
Last edited by a moderator:
One would think that with the massive problems the fabs have had with 40nm a part 6-8 months later will have acess to much more mature processes. I would think ati would get higher clocks from a cypress at that point.
Ati released at a time when i'm sure they new nvidia would not release a part for some months. I don't think they needed anything faster than the 5870 and so they didn't really release anything except for the 5970 which is based on the 5850.
So perhaps what we will see is a faster 5870. Not drasticly but mabye close to what over clockers are getting. That would take over the 5870's price point. The 5870 will drop down in price and so would the 5850. Perhaps with 6 months of improvements we can see the 5870 running cool enough and low enough in power that we get a dual card based on it.
Who knows how that will compare against nvidia , but i highly doubt that ati is jsut sitting aorund doing nothing the last 6 months since they released the new cards
Ati released at a time when i'm sure they new nvidia would not release a part for some months. I don't think they needed anything faster than the 5870 and so they didn't really release anything except for the 5970 which is based on the 5850.
So perhaps what we will see is a faster 5870. Not drasticly but mabye close to what over clockers are getting. That would take over the 5870's price point. The 5870 will drop down in price and so would the 5850. Perhaps with 6 months of improvements we can see the 5870 running cool enough and low enough in power that we get a dual card based on it.
Who knows how that will compare against nvidia , but i highly doubt that ati is jsut sitting aorund doing nothing the last 6 months since they released the new cards
It seems that we can't expect that particular cat to be dragged out of the bag again, you know the story. Once bagged, twice shy. I don't see much progress from decoupled shaders as I doubt the architecture is particularly shader limited, I don't see them investing too much into a refresh unless they intend to use the architecture going forward once the r9xx comes out SO...
Hmm at best I could expect a respin refresh for Cypress for lower thermals which will allow for higher clocks, and maybe even a few sales to desktop replacement style laptops. Have they ever respun a good design in the past to push higher clocks?
If they did respin they could fix a few things and slap on a wider bus most likely and let their chip bathe in the glow of even more bandwidth. It could also let them get away with a faster 5850 style chip as they could keep it relatively bandwidth limited compared to 5870 or whatever the number. In addition to this any improvement in TDP for the single cards is translated in double for the Hemlock cards as they are strictly limited by heat and have two instead of one.
For their high end card im sure they have some wicked margins and improving the performance along the lines of a wider bus and faster chip would cost them but in addition to this allow them to move their entire range up a notch in performance and yield higher margins on their chips once the competition releases. I hope they think that way too, because I want MOAR performance.
r420->r480 (x800->x850)
That wasn't much of a difference unless Wikipoodia is wrong or misleading. Circa 20mhz? I guess we shouldn't expect too much, if thats also what you're trying to say as well.
Yeah, it wasn't a huge change by any means. I'd expect HD 5870 -> HD 5890 to have a larger clock differential than that, of course. I'd guess 950-1000MHz core clock.
Similar threads
