Rumors from Chiphell have suggested that not only is AMD planning a new HD8000 family for the consumer market, they may even announce it to the world at Computex Taipei next month.
The HD8000 family already available is composed of two markets – OEMs and mobile chips. The current desktop HD8000 graphics cards are used by OEMs and are re-branded HD7000 GPUs. The OEM market always does this and Nvidia’s GT300 series was just a rehashed GT200 lineup with higher clocks for OEM machines. The mobile HD8000 lineup is composed of mid-range and high-end HD7000 chips, but does replace the lower ranges with GCN products, as the low-end HD7000 mobile range was actually based on the HD6000 family.
Its all very confusing but it looks like AMD may be gearing up to sort it all out. A user on the Chiphell forums unearthed a brand new, never-seen-before specs list and it does appear to match up to a new GPU family and not a refresh of HD7000. For comparison, I’ve added in the Radeon HD7970 and HD7950 as well as the new Nvidia Geforce GTX780.
| AMD Radeon HD8000 |
HD8970 | HD8950 | HD8870 | HD8850 | HD7970 GHz Edition |
HD7950 | GTX780 |
| Estimated RRP | $599 | $399 | $299 | $229 | $449 | $350 | $649 |
| Codename | Curacao XT | Hainan XT | Hainan Pro | Hainan LE | Tahiti XT | Tahiti Pro | Kepler |
| Process | 28nm | 28nm | 28nm | 28nm | 28nm | 28nm | 28nm |
| Transistors (billion) | 5.5B | 3.5B | 3.5B | 3.5B | 4.31B | 4.31B | 7.1B |
| Shader cores | 2304 | 1792 | 1536 | 1280 | 2048 | 1792 | |
| Core clock | 1.1GHz | 1.2GHz | 1.1GHz | 1.0GHz | 1.0GHz | 800MHz | 863MHz |
| Boost clock | – | – | – | – | 1.05GHz | 900MHz | |
| Bus width | 384-bit | 256-bit | 256-bit | 256-bit | 384-bit | 384-bit | 384-bit |
| Memory size | 6GB | 4GB | 2GB | 2GB | 3GB | 3GB | 3GB |
| Memory clock | 7.0GHz | 7.0GHz | 6.0GHz | 6.0GHz | 6.0GHz | 5.0GHz | 6.0GHz |
| Peak performance | 5.07TFLOPS | 4.3TFLOPS | 3.38TFLOPS | 2.56TFLOPS | 4.3TFLOPS | 2.8TFLOPS | 3.97TFLOPS |
Although it may not be entirely authentic, some things are believable. The rumoured HD8970, for example, would have 36 Compute units composed of 64 stream processors each. This is only four more than the HD7970 and its easily within the reach of AMD to include the extra hardware. That would send its performance above the GTX780 and close to the Titan as well, which has a theoretical maximum of 4.5TFLOPS of computing throughput.
However, it is highly unlikely that AMD would give up the one thing that makes their cards highly superior – dropping the bus width on the HD8950 down to 256-bit would actually drop memory performance and allow Nvidia to catch up in the bandwidth stakes. It is a delicate balancing act, though. If AMD can foresee more games prioritising memory amounts instead of bandwidth numbers (which it is in a position to do as they power all three next-generation consoles) then the move may make some sense, as well as push down the manufacturing cost.
Compared to the current HD7900 series lineup, it would be a decent upgrade and the rumoured HD8970 could, in theory, match the GTX 780’s performance and give the GTX Titan cause for concern. But its strange for the company to not have mentioned this before. Perhaps they’re gearing up for an official announcement at Computex, along with a launch in Q3 2013.
Source: WCCF Tech, Chiphell
More Hardware news:
G.Skill announces new range of headphones for 2013
MSI gaming laptops SA pricing: GX70 3BE and GX60 3AE compared