Thursday, December 9th 2010
Radeon HD 6900 Series Products Listed on European Store
A French online store listed AMD Radeon HD 6900 series graphics cards. Both branded by MSI, the Radeon HD 6950 part carries the model number R6950-2PM2D2G5, and packs 1 GB of GDDR5 memory, while the Radeon HD 6970 carries the model number R6970-2PM2D2GD5, and packs twice the amount - 2 GB GDDR5 memory. Besides the memory amount, the other only significant specification is the list price, €398.22 for the HD 6950, and €496.02 for the HD 6970. So far, both SKUs are known to be high-end single GPU cards. AMD's elusive HD 6900 series graphics cards will be launched next week.
Source:
TechConnect Magazine
91 Comments on Radeon HD 6900 Series Products Listed on European Store
double bart = 3.4Billion transistors... seriously!
Of course the Don's didn't like everyone pointing this fact out on the forums...... :shadedshu:D
I do hope you (Gibbo) are right on pricing though, £350 for a HD6970 sounds tasty although I will be waiting for the HD6990 myself ( I want an upgrade not a sidestep).
Oh and hi Raven......
I have no basis for any conclusion at this point...therefore I conclude (based on preconceived bias)...
www.anandtech.com/show/3987/amds-radeon-6870-6850-renewing-competition-in-the-midrange-market/2
1- One of them was the slower memory controler which they borrowed from Redwood and is half the size of the one in Cypress. The high speed memory in Cayman will need a controler as big or bigger than Cypress, since memory is much higher clocked.
2- Lack of double precision math on Barts.
The effect of adding double precision can be seen (or sensed) in Redwood vs Juniper vs Cypress. Each of them is half the next one when it comes to SP/TMU/ROP/MC, but let's look at transistor count:
Redwood = 627 million
Juniper = 1040 million (+66%)
Cypress = 2154 million (+107%)
As you can see the transistor count jump is much bigger in Juniper->Cypress, and the main reason as explained by AMD themselves is that both Redwood and Juniper lack double precision, just like Barts lacks it too as explained in the Anand article.
I know the above it's messy and hard to understand so here's a summary:
- Entire chip has less than 16 SIMDs: VLIW5 >> VLIW4, because it can do 1 more op/second and with this "low" ammount of SPs the setup engine + dispatchers can keep up. aka the chip remains efficient.
- Entire chip has more than 16 SIMD: VLIW4 >> VLIW5, because VLIW5 has demostrated that it cannot scale performance with so many SIMDs, reason for which Cypress is nowhere near 2x as fast as Juniper despite being exactly 2xJuniper. Or how the HD5850 is exactly as fast as HD5870 when both are running at the same clocks.
Is not cayman meant to be all about the new architecture? thus we have no clue what is possible, i mean look at intel's and amd's 65nm cpu's was there not a big gap there and all because of architecture?
I coudl be totally wrong in my thought pattern here as i'm rly dam drunk :roll: :toast:
I'm just trying to say what many others have said really, we don't have a clue what is possible until they come out due to changes in architecture, for all we know it could be 600% faster :roll: yea ok thats a little into the range of impossible but hopefully there was some point in my rambling :laugh:
ok that's enough drunken posting :laugh: