en.hw-lab.com AMD Radeon HD 6800 series cards (AMD Barts)
Meet the new series of graphic cards — specification and architecture analysis
Architecture
As a result of “Barts” graphic core optimization, AMD managed to achieve performance level of 2 TeraFLOPS together with significant decrease of dies size.
GDDR5 memory is paired up with 256-bit width memory interface.
The architecture of AMD “Barts” (picture above) is very similar to what previous generation cards — AMD “Cypress” (picture below) — have:
Let us compare details of both GPUs in the table below:
| Cypress | Barts | |
|---|---|---|
| Release date | September 2009 | October 2010 |
| Process technology | 40 nm | 40 nm |
| Transistor count | 2.15 billion | 1.7 billion |
| Die size | 334 mm2 | 255 mm2 |
| SIMD | 20 | 14 |
| Stream Processor | 1600 | 1120 |
| Texture Units | 80 | 56 |
| Z/Stencil ROPs | 128 | 128 |
| ROPs | 32 | 32 |
| Memory bus | 256-bit | 256-bit |
There were some major “cut-downs” in AMD “Barts” architecture. Except for one thing: the AMD “Barts” now has two dispatch processors followed by a doubled amount of instruction cache. As a result, with considerable reduction in performance configuration, new GPU is only slightly less powerful than AMD “Cypress”. This is the case when quantity turns into quality:
We have already seen similar architecture optimization in NVIDIA GF104 graphic processor (GeForce GTX 460 graphic cards).
AMD Radeon HD 6870has a full-fledged processor, therefore, the number of execution block has now been changed. AMD Radeon HD 6850 has two SIMDs locked which resulted in reduced number of Stream Processor — 960.
No comments