AMD: R9xx Speculation

What? Chiphell showed a 256-Bit sample card (8 memory chips) and these very authentic slides say 1GiB and 6Gbps.
Or do you believe AMD is going with Cayman over 300GB/s (512-Bit) BW?:LOL:
Hmm, yes, I forgot it's not possible to have less than 1GB with GDDR5 and a bus of at least 256 bits. So that means Cayman is 256-bit.

EDIT: hmm, that's not right, HD4870 had a 512MB version.
 
It's just a slide. A 5% higher position on that slide does not necessarily mean Cayman XT is only 5% faster than a GTX 480. If Cayman is 15% faster than GF100, that would mean it's about 30% faster than Cypress. Considering they're still on the same process, that would be an acceptable performance boost, in my opinion.



Until we know exact prices and performance we can't tell for sure. The 5850 is ~10-15% faster than the 460-1G. If Barts XT is 10% faster than a 5850, that would amount to a comfortable lead over the 460. A price of 249$ would be ok in that case.

I think they'll wait until christmas shopping season is over before they lower any prices, because production ramp up won't happen over night, and if demand exceeds supply anyway, lowering prices too early would be counter-productive.

Fair points mate.

Still, I have a hunch that these slides tend to show the exact percentage improvements. Most of the time, marketing people tend to overstate differences in graphs. Is it possible that AMD's guys had a change of heart all of sudden? Unless they want to get really nasty and throw a 30% performance lead on the GTX 480, for a 5% pixel difference on the slide.

Still the percentage difference of the Juniper with the GTS 450, seems to be about at the right analogy, so that makes me believe that this will be the case for the Cayman XT vs GTX 480.
 
Last edited by a moderator:
One would have hoped that they learned a little from the 460 referecne cooler.

You mean the one that exhausts the hot air inside the case?

Well, let's hope they learned nothing from that and thank God, it seems they didn't.

Blower type cooler FTW!
 
83317390.jpg


I'd guess this is where the Juniper rebranding rumours originate. If indeed Juniper gets rebranded to 6700 that would pretty much mean there is no Juniper replacement coming. If Turks is Juniper replacement, coming in January, then Juniper will probably keep the 5700 name.

We all know Juniper is making AMD a lot of money but I still can't see why they wouldn't choose to replace it with something smaller and faster. Make even more money?
 
AMD Radeon HD 6850 Specs, Pricing Surfaces
AMD's latest graphics processors is just around the corner, and one of the first of them is the Radeon HD 6800 series. The value version of it is the Radeon HD 6850, its most probable specifications have surfaced. To begin with, HD 6850 is based on AMD's new Barts GPU, built on the 40 nm process. The source mentions that the SKU will have 800 stream cores enabled, from earlier reports we're lead to believe that these stream cores are individually more complex than AMD's traditional 5D (4 simple + 1 complex) approach to unified shaders. There is a 256-bit wide GDDR5 memory interface holding 1 GB of memory, the card uses 5 GT/s memory chips, so the memory should be clocked around 1200 MHz (or 4800 MHz effective), if not more. The core is clocked at 775 MHz. Its FOB (freight on board) price is expected to be US $175. Power is drawn from a single 6-pin PCI-E power connector, the draw is expected to be less than 150W. Partners have the option of using a premium blower-type cooler, or a cost-effective heatsink-type cooler. The latter had been pictured a while back, posted below for reference.
http://www.techpowerup.com/132783/AMD-Radeon-HD-6850-Specs-Pricing-Surfaces.html
 
You mean the one that exhausts the hot air inside the case?

Well, let's hope they learned nothing from that and thank God, it seems they didn't.

Blower type cooler FTW!

Axial fans (at least the bigger ones) are quiter at full load and dont tend to give out high pitched sound. Also if it blows the air into case it mixes it and cools it down. If u have a single PSU fan in your case than its your problem.
 
Axial fans (at least the bigger ones) are quiter at full load and dont tend to give out high pitched sound. Also if it blows the air into case it mixes it and cools it down. If u have a single PSU fan in your case than its your problem.

So, if say the Barts XT has a TDP of 160W, as the 460, will you have the same conditions inside the case, if you use two of each?

Being the owner of a few rigs, I have seen what happens when you use one GTX 460 vs using one 5850 on the same case. And I am talking a Thermaltake Armor with a 25cm sidepanel here and four 12cm fans.

If I want to use one case fan, it is indeed my problem, but this would keep the noise down as well. A blower type HSF, does not need the additional cooling power of 100 fans inside the case. That's my point.
 
And not fully square like Cypress cuz there aren't that many of them to fill it out. I'd imagine Cayman squarer looking.
If I'm not mistaken, non-square die should also increase the amount of i/o pads possible for the same area (not saying that's the reason here, just another reason why you'd want a non-square die sometimes). FWIW Juniper also wasn't square.
If that 230 mm² number is true (and it looks legit to me?) that's quite ok imho - same size as GF106, with GF104 performance.
So I guess this means still limited to 2 non-DP monitors, what's new here is just DP 1.2 (which can daisy chain monitors - hence you don't need special Eyefinity card for 6 displays if your displays can handle that too).
 
Last edited by a moderator:
If I'm not mistaken, non-square die should also increase the amount of i/o pads possible for the same area

Yeah, a square die minimizes the available perimeter for a fixed area. The more lopsided the aspect ratio, the more perimeter is available.
 
Many things simply don't add up...

How could a 800-SP <900MHz part be sufficiently faster than a 800-SP 850MHz Juniper for them to coexist, without it being at least as fast as a 1440-SP @ 725MHz Cypress in its full glory (960?-SP @ 900MHz)?
 
My guess is $199 for 50 and $249 for 70.

As for Cayman I'd say something like $299 for Pro and either $349 or $399 for XT.
 
Last edited by a moderator:
Many things simply don't add up...

How could a 800-SP <900MHz part be sufficiently faster than a 800-SP 850MHz Juniper for them to coexist, without it being at least as fast as a 1440-SP @ 725MHz Cypress in its full glory (960?-SP @ 900MHz)?
To me the wattage doesn't add up. The 6870 seems to have the same consumption as a 5850, with not so much more performance. Ok, they saved up on die space. But the same (or similar) amount of heat, having to be dissipated from a lower area, wouldn't that create more problems?

Also, we should see lower prices, right?

But that's a rip-off! The same store has GTX460 1GB for 199E. While 6850 might be able to defend itself at this price point, the 6870 pricing is a farce to me!

Two things can save it for me:
1) Significantly better DX11 (ie. tesselation) performace (better than nV, not Cypress!)
2) Enabled double precision
 
To me the wattage doesn't add up. The 6870 seems to have the same consumption as a 5850, with not so much more performance. Ok, they saved up on die space. But the same (or similar) amount of heat, having to be dissipated from a lower area, wouldn't that create more problems?
It is indeed more difficult to cool a smaller die vs. a larger one if it uses the same amount of power, but I'm not sure this really makes enough difference here. You could, for instance, still use the same hsf the fan would just have to spin faster (of course, that's not ideal for noise). Or the chip could be specified to run at a bit higher temperature (though that would cause in itself higher power draw too - might already be factored in in that TDP).
But that's a rip-off! The same store has GTX460 1GB for 199E. While 6850 might be able to defend itself at this price point, the 6870 pricing is a farce to me!
Why? If the HD6850 has same performance as GTX460 1GB it is slightly cheaper then. The HD6870 is expected to be faster than HD5850, and this shop at least only has one HD5850 which manages to beat that preorder price (barely) all the rest are more expensive (sometimes quite a bit). Of course, pricing could always be better, but if the rumored performance holds up this looks competitive to me.

Two things can save it for me:
1) Significantly better DX11 (ie. tesselation) performace (better than nV, not Cypress!)
I think the odds of tesselation performance being better than what GF104 offers are pretty slim. The other question though is why would you want that, just so that it's faster on unigine extreme?
2) Enabled double precision
Hmm yes I think that would be nice to have. Dunno though AMD just isn't focused that much on gpgpu as nvidia.
What do we think the die increase for Cayman is in comparison with Barts? I can't help seeing 50-60% more area as extremely likely.
If Barts is really 2 RPE and Cayman 3 RPE (whatever those actually are...), pretty much no matter how I look at it Cayman would be barely 50% larger than Barts. Now, if Barts is indeed 230mm² that would put Cayman at - nearly exactly the same die size as Cypress...
Seems a bit unlikely but might make sense? If it was initially targeted at 32nm, it might indeed have been 4 RPEs, which would still have put this at around 300mm². AMD might have been unwilling to build ~450mm² chip at 40nm and just redesigned it with 3 RPEs only. Maybe clocks got some boost to compensate at least a little.
 
Back
Top