- Sep 16, 2010
- 6,654
- 5
- 76
UPDATE:
VR Zone updated their article: http://vr-zone.com/articles/updated--rumour-amd-radeon-hd-6700-specification-chart-leaked/9936.html
"UPDATE: Over at Chiphell, the source of most leaks, Napoleon, has leaked the full version of the same table. Two very similar tables coming from different sources certainly lends much more credibility to this chart than previous ones. The full chart reveals further details - power consumption. While notably higher than HD 5700, the HD 6770 is also lower than HD 5850, a GPU it will most likely beat comfortably. The HD 6770 features a TDP of 146W and the HD 6750 a very reasonable 116W. Idle power usage is 23W and 20W respectively. This means the HD 6770, while under 150W, will probably release with 2-pin PCI-e connectors, while the HD 6750 can easily do with just one."
Additionally, at least one S|A'er with a history of inside knowledge says that these specs are real.
http://vr-zone.com/articles/-rumour-amd-radeon-hd-6700-specification-chart-leaked/9936.html
In case AMD makes vr-zone take down the article or at least the pilfered slide, I'm copying parts of the article here and bolding/underlining some of the highlights (someone else can screenshot or copy the .jpg if they want):
"Codenamed Barts XT, the HD 6770 features 320 VLIW-4 shader clusters, i.e. 1280 SP. It features the same ROP count (32) and memory bus (256-bit) as the HD 5870, though TMU count is dropped to 64 (from 80). The core clock is a whopping 900 MHz, making it the fastest stock speed of any discrete GPU. The memory clock and thus bandwidth is a shade lower than the HD 5870, at 134.4 GB/s. Barts Pro, branded HD 6750, features 280 clusters, or 1120 SP and 56 TMU and a much lower clock speed of 725 MHz, though the memory speed is only 50 MHz lower at 1 GHz. Incidentally, these clock speeds are identical to the HD 5850.
The chart also reveals details about the SIMD structure of the Barts core. The Evergreen series SIMD featured 16 VLIW-5 clusters of 5 shaders each, for a total of 80 SP. Each SIMD also featured 4 TMU. The top part, Cypress XT / HD 5870, thus comprised of 20 SIMD, which is 320 clusters = 1600 SP and 80 TMU. The first major change, that has already been widely speculated, is the structure of each shader cluster. It now features only 4 shaders. However, each cluster is now more efficient, providing equal or greater performance for a smaller die area. Each SIMD now oddly consists of 20 such VLIW-4 clusters. (8, 16, 32... are usually expected) The total shaders per SIMD, thus remains the same - 80 - though they are arranged differently. Barts XT features 16 SIMD, or 320 clusters with 4 shaders each. I.e. a total of 1280 SP. The TMU per SIMD ratio remains constant at 4. Hence, Barts XT features 64 TMU. If these specifications are true, the end result is a HD 6770 card which soundly defeats a HD 5850, as recent rumours suggest, and gets very close to a HD 5870."
These look way more legit than the previous "leak" that had numbers that made no sense and a black smudge on them. But it's still not convincing enough for me to believe it's necessarily true. The whole "Each SIMD now oddly consists of 20 such VLIW-4 cluster" thing makes me wonder....
VR Zone updated their article: http://vr-zone.com/articles/updated--rumour-amd-radeon-hd-6700-specification-chart-leaked/9936.html
"UPDATE: Over at Chiphell, the source of most leaks, Napoleon, has leaked the full version of the same table. Two very similar tables coming from different sources certainly lends much more credibility to this chart than previous ones. The full chart reveals further details - power consumption. While notably higher than HD 5700, the HD 6770 is also lower than HD 5850, a GPU it will most likely beat comfortably. The HD 6770 features a TDP of 146W and the HD 6750 a very reasonable 116W. Idle power usage is 23W and 20W respectively. This means the HD 6770, while under 150W, will probably release with 2-pin PCI-e connectors, while the HD 6750 can easily do with just one."
Additionally, at least one S|A'er with a history of inside knowledge says that these specs are real.
http://vr-zone.com/articles/-rumour-amd-radeon-hd-6700-specification-chart-leaked/9936.html
In case AMD makes vr-zone take down the article or at least the pilfered slide, I'm copying parts of the article here and bolding/underlining some of the highlights (someone else can screenshot or copy the .jpg if they want):
"Codenamed Barts XT, the HD 6770 features 320 VLIW-4 shader clusters, i.e. 1280 SP. It features the same ROP count (32) and memory bus (256-bit) as the HD 5870, though TMU count is dropped to 64 (from 80). The core clock is a whopping 900 MHz, making it the fastest stock speed of any discrete GPU. The memory clock and thus bandwidth is a shade lower than the HD 5870, at 134.4 GB/s. Barts Pro, branded HD 6750, features 280 clusters, or 1120 SP and 56 TMU and a much lower clock speed of 725 MHz, though the memory speed is only 50 MHz lower at 1 GHz. Incidentally, these clock speeds are identical to the HD 5850.
The chart also reveals details about the SIMD structure of the Barts core. The Evergreen series SIMD featured 16 VLIW-5 clusters of 5 shaders each, for a total of 80 SP. Each SIMD also featured 4 TMU. The top part, Cypress XT / HD 5870, thus comprised of 20 SIMD, which is 320 clusters = 1600 SP and 80 TMU. The first major change, that has already been widely speculated, is the structure of each shader cluster. It now features only 4 shaders. However, each cluster is now more efficient, providing equal or greater performance for a smaller die area. Each SIMD now oddly consists of 20 such VLIW-4 clusters. (8, 16, 32... are usually expected) The total shaders per SIMD, thus remains the same - 80 - though they are arranged differently. Barts XT features 16 SIMD, or 320 clusters with 4 shaders each. I.e. a total of 1280 SP. The TMU per SIMD ratio remains constant at 4. Hence, Barts XT features 64 TMU. If these specifications are true, the end result is a HD 6770 card which soundly defeats a HD 5850, as recent rumours suggest, and gets very close to a HD 5870."
These look way more legit than the previous "leak" that had numbers that made no sense and a black smudge on them. But it's still not convincing enough for me to believe it's necessarily true. The whole "Each SIMD now oddly consists of 20 such VLIW-4 cluster" thing makes me wonder....
Last edited:
Facebook
Twitter