I'm surprized I haven't seen this posted on Rage3D yet, but AMD just officially announced the Carrizo APU architecture, and it's looking good so far.
Here's the link:
http://www.anandtech.com/show/8742/a...us-for-h1-2015
Some of the main points include:
-Will be released 1H 2015
-28nm fabrication process
-2-4 Excavator CPU cores
-GCN 1.2 GPU cores
-Integrated south bridge
-Full HSA 1.0 support (Kaveri didn't have this)
-AMD secure ARM coprocessor
-Optimized for improved power efficiency
-Carrizo and Carrizo-L are both using the same new FM4 socket
-Carrizo-L is actually the successor of Beema, using Puma+ CPU cores and GCN 1.2 GPU cores
There isn't any mention whether its going to use DDR3 or DDR4, so we'll have to wait and see on that. I'd like to point out that even if Carrizo stays with DDR3, there will still be a big improvement in effective memory bandwidth since GCN 1.2 has color delta compression. Considering the Radeon 285 and 280 comparison, it suggests that Carrizo could have a 30-35% performance improvement from memory efficiency alone. There isn't any mention of how many GCN cores Carrizo will have, but even if they stay with 512, there will still be a substantial performance improvement.
It's a shame they're staying with a 28nm fabrication process, but considering pretty much everyone except Intel and Apple are stuck at 28nm, its not very surprising. Since the 28nm node is so mature at this point, I bet AMD can get better power efficiency, higher clocks, and improved yields to reduce costs at least.
Looking forward to seeing the Carrizo SKU's detailed early next year!
Here's the link:
http://www.anandtech.com/show/8742/a...us-for-h1-2015
Some of the main points include:
-Will be released 1H 2015
-28nm fabrication process
-2-4 Excavator CPU cores
-GCN 1.2 GPU cores
-Integrated south bridge
-Full HSA 1.0 support (Kaveri didn't have this)
-AMD secure ARM coprocessor
-Optimized for improved power efficiency
-Carrizo and Carrizo-L are both using the same new FM4 socket
-Carrizo-L is actually the successor of Beema, using Puma+ CPU cores and GCN 1.2 GPU cores
There isn't any mention whether its going to use DDR3 or DDR4, so we'll have to wait and see on that. I'd like to point out that even if Carrizo stays with DDR3, there will still be a big improvement in effective memory bandwidth since GCN 1.2 has color delta compression. Considering the Radeon 285 and 280 comparison, it suggests that Carrizo could have a 30-35% performance improvement from memory efficiency alone. There isn't any mention of how many GCN cores Carrizo will have, but even if they stay with 512, there will still be a substantial performance improvement.
It's a shame they're staying with a 28nm fabrication process, but considering pretty much everyone except Intel and Apple are stuck at 28nm, its not very surprising. Since the 28nm node is so mature at this point, I bet AMD can get better power efficiency, higher clocks, and improved yields to reduce costs at least.
Looking forward to seeing the Carrizo SKU's detailed early next year!
Comment