AMD: R9xx Speculation

Has this been dismissed yet? :

hd6000.3.jpg

http://www.nordichardware.com/news/71-graphics/41433-amd-radeon-hd-6000-roadmap-leaked.html
Based on SRP for obvious reasons.
 
Well well, where have we seen dates as Quarters before? :rolleyes:
Antilles seems to be already out of schedulle, as their powerpoints said it was December. And Q1 can be anywhere from January to March...
Cayman Q4 2010... Not very specific...
Will we see AMD's "Fermi"?

Lol okay? They often keep release dates under NDA as well

Wasn't Barts release date confirmed til a few days before? Barts was labeled H2 2010

You're grasping at straws right now
 
IMHO, apart from the much more understandable numbers in Civ 5, Damien is presenting,
Do you want to say Anand may measure something not representative of the actual game? :rolleyes:
there's one possible explanation to this I think.

With Civ 5 being quite a CPU intensive application as well, there is the chance it probably wastes some cycles in lower resolutions trying to be clever on Tessellation usage in order to lower the graphics load. As I've pointed out here, there are some techniques suggested to limit tessellation factors at lower resolutions (screen space adaptive tessellation like „Consider the screen space patch edge length as a scaling factor”). Maybe those computations are just omitted at some point or „target screen res”.
That is normally done in the hull shader and it wouldn't make much sense to skip it as it shouldn't be terrible time consuming. By the way, that would fall roughly in the same category as:
I'm thinking about if it is possible that the compute shader used to decompress textures on the fly tries to be clever by using some adaptive scheme depending on the size and speed of the GPU and messes everything up.
 
From Charlie with love:



Do I smell harsh in his tone?

He's right. I was watching the webcast and when the speaker (I think it was Rick Bergman) mentioned that it was a Bulldozer+Cayman rig, I got all excited. Then he showed a freaking HD video.

I mean the point was obviously to show that BD and Cayman are both up and running while not giving anything more away, but still…
 
That would be funskies. Although we know from Chiphell that people do have Caymans so Tom's is at least wrong on that point. AMD has been amazingly good at keeping leaks down of late...much to my disappointment.

Even Charlie seems to be not even semi-sure about what exactly Cayman is. However given this VLIW 4 or 5 debacle which has distracted everyone, I blame Dave. I always blame Dave for these kinds of shenanigans. You should blame Dave too, because its good to blame the most public and high profile AMD representative you have. Blame is a very good way of dealing with disapointment I find.

Why not a 'B' plan ???

Look the Techreport review. A dual Barts can compete with GTX580 (in performance, price, power consumption, etc).
It's not hard. AMD has several dual-GPU card designs.

I suspect the reason why is because dual Barts is probably faster than single Cayman and it would confuse their product lineup to have an 890 series product faster than 950 or 970 series product. Also they probably don't need to. Cayman is likely faster than GTX 580 by a small margin at least.
 
We have got to overhaul B3D's intelligence community.

No appreciable leaks worth diddly? Shameful. Just shameful.
 
We have got to overhaul B3D's intelligence community.

No appreciable leaks worth diddly? Shameful. Just shameful.

We have an insider working for AMD on these forums. How can we have an effective intelligence community when there are people working for the enemy who post here? :rolleyes:
 
Looks like 6970 is closer to 5970 than 5870 and 6950 is about 10-20% faster than 5870. All taking as reference the placement of 6870, since its slotted nicely between the 5850 & 5870 and 6850 is slower than the 5850, both are fairly accurate conclusions.

Just trying to keep Mize entertained.
 
As a side note, did anyone else notice in the 580 Anandtech review how well 6870 was working in Crossfire? Almost X2 the performance of one 6870 in a lot of the cases. Very SLI like. If that scaling can carry over to the 6990 it's going to be a beast.
 
"When it's ready..."


I haven't heard that statement before so, please use that when referencing the release date..

So far nothing about the 580 performance surprises as this card was addressed by Hemlock last year already.
 
neliz said:
So far nothing about the 580 performance surprises as this card was addressed by Hemlock last year already.

That depends on what "addresses" means. A dual-socket board with 2 GB of memory at 256 GB/s, coupled to 670 mm^2 of GPU silicon running at 4.64 TFlop/s should theoretically be destroying the GTX580, yet somehow the GTX580 manages to compete with the 5970 despite having 75% of the memory bandwidth, 78% of the die area, and 34% of the flops.

The 5970 was a halo part which was chronically hard to find and overpriced - not surprising since it was so expensive to produce. It is surprising, though, that anyone would praise a card which makes GTX480 look like an efficient use of silicon.

Luckily, AMD is not relying on 5970 to address the 580....
 
That depends on what "addresses" means. A dual-socket board with 2 GB of memory at 256 GB/s, coupled to 670 mm^2 of GPU silicon running at 4.64 TFlop/s should theoretically be destroying the GTX580, yet somehow the GTX580 manages to compete with the 5970 despite having 75% of the memory bandwidth, 78% of the die area, and 34% of the flops.

The 5970 was a halo part which was chronically hard to find and overpriced - not surprising since it was so expensive to produce. It is surprising, though, that anyone would praise a card which makes GTX480 look like an efficient use of silicon.

Luckily, AMD is not relying on 5970 to address the 580....

Pretty terrible comparison

1) You can't add silicon sizes and say that they should perform equivalently. Two ~330mm^2 dies are easier to produce than one > 500 mm^2 - yields do not scale linearly with size, they change exponentially. Thus one can see 2 x 330mm^2 dies yielding better overall than one 520mm^2 depending on design

2) Flops are another terrible way to compare performance across different architectures from the same company, let alone from different companies

3) Dual GPU bandwidth is hardly calculated that way and again cannot be added together

Of course, you managed to forget the fact that the 5970 also consumes less power than the 480, and in some reviews still less than the 580, despite offering more or roughly the same performance while having to deal with CF issues, while also being released a year earlier

So you managed make a comparison with totally asinine arguments. Congrats
 
Operation Smokescreen failed: 6900 is still coming on time and I'm sorry if I said anything about 1GB.
 
Back
Top