Each generation of graphic card is faster than the previous one, but the real question is ‘How much faster?’. KitGuru has been up late at night, drinking with experts and ‘outsiders who claim to be in the know’. We’re ready to stick our flag in the ground on Radeon HD 6970 performance. KitGuru friends, prepare to smile knowingly. KitGuru doubters, sharpen your knives and prepare to stab our backs.
Graphics processing speed can be improved by a huge number of factors. It’s a bit like Formula One cars. Often one team spends ages developing a series of small improvements – each of which gives them a small boost – only to find that the competition made one single change and blasted to the front of the grid (we’re thinking ‘rear diffuser’). The next generation of Radeons is several months away, but it’s looking good for the graphics division of CPU giant AMD. Let’s break the future down into manageable blocks and see what might be possible by Christmas.
What’s all this Tick-Tock stuff?
With processors (CPU and GPU), the most common ‘diffuser advantage’ is simply moving to a new process sooner. The new process will be smaller (die shrink), reduce hot-spots and the effect of aggressor wires etc. Most importantly, in the case of graphics, it will give you more GPUs per wafer.
Back in the disastrous days of the R600/Radeon 2000 series, nVidia was scoring clear wins at almost every price point. The entry level Radeon 2400, however, had one thing going for it. ATI managed to make it smaller and cheaper than the nearest nVidia product. The difference was less than $5, but in creating that small price advantage, ATI won a huge amount of business with the world’s biggest system builders like Lenovo. All of our Far East system builder friends are famous for ‘tick box marketing’ and their inability to ‘say no to a saving’.
Will AMD be able to move to 28nm for the launch of the remaining Radeon 6000 series in Q4? It is unlikely. In the past, both nVidia and ATI have been burned by trying to move a complex technology to a new process. At the same time, the world leader in this kind of manufacturing has a simple strategy called Tick-Tock.
In its own words, Intel creates a new process (i.e. a shrink) which gives a better version of an existing processor. They call this the ‘Tick’. Then Intel unveils a new processor architecture, which is normally much more powerful. They call that the ‘Tock’. The result is a series of solid, predictable steps. It’s a very simple strategy but, most importantly, it works.
Will AMD try to beat the Tick-Tock?
Looking at the complexity of each new generation of graphic cards, we can’t see AMD wanting to take any huge risk in bringing them to market.
So how can you know for sure whether AMD’s Radeon HD 6000 series will launch on 28nm?……….Simple.
If there’s one thing that KitGuru’s learned over the years, it’s that semi-conductor companies LOVE laying claim to any ‘world first’ available. You can almost imagine the CTOs getting armoured up and laying claim to the new lands. There’s been no announcement so far, so we think the die shrink is a safe ‘no’ bet.
“Why has the 5000-series taken so long to re-spin?” we hear KitGuru readers ask. Response? “Lack of DX11 competition”. Without stiff competition for so long, we’d guess that AMD has gone straight past a simple re-spin and will be making significant changes to the architecture.
Think about it this way, even before a new product launches the manufacturer already knows what the weak points are. They will know which parts work and which are busted. Remember, GPU projects don’t run in series, there are strong elements of parallel development. Overlaps. The Radeon HD 6000 series design kicked off a long time before the Radeon 5000 ever launched. Known issues with the Radeon 5000’s basic design will have been addressed, alongside a number of issues that AMD will have found with the 5000 product AFTER then went into production.
From the launch of the Radeon HD 5870 last September, AMD was probably in position to do some kind of re-spin around the CeBIT timeframe (March 2010 – 6 months after launch). However, with little/no DX11 pressure from nVidia, development would have carried on. At the same time, the ATI engineering teams would have been able to analyse the complete Fermi architecture in detail. That analysis lets the Developer Relations team understand what kind of games would work best on Fermi hardware.
If AMD had launched around Computex (June 2010 – 9 months after the initial Radeon HD 5870 launch), then a lot of changes could have been made. Given the actual launch date of Northern Islands (Er… today!), loads of improvements to the fundamental design will have been possible. As a result, you can see that AMD’s latest £135 cards can be overclocked to match performance levels only available around £350 a year ago.
We can expect much more from the Radeon HD 6970.
Radeon HD 6970 Benchmark Prediction
From KitGuru’s first experience with the GTX480 at CeBIT 2010, it seemed clear that nVidia’s biggest win would be in ultra-complex tessellation situations. KitGuru Labs’ full, in-depth testing with the GTX480 has proven that correct. Turning the tessellation sections of the Unigine benchmark to ‘crazy difficult mode’ allows the GTX480 to pull away from it’s Radeon competitors by some margin. Unfortunately for nVidia, none of its TWIMTBP games have been released yet and gamers don’t play benchmarks. Our labs tests show that articifially high levels of tessellation don’t seem to represent any real world examples of gaming. Maybe that will change next year, but not right now. Some games with increased tessellation will be hitting the streets around the Christmas holidays, but mostly at sensible levels.
All of that preamble brings us to the main event. Can KitGuru accurately predict how fast the AMD radeon HD 6970 graphics card will be when it launches just in time for Christmas 2010?
We’ve tried out best.
KitGuru says: Are we correct? Are we way out? With the AMD Radeon HD 6970 sail past 23,000? Would you bet against our prediction? Come on KitGuru followers – you must have an opinion. Also, there is the real question to consider, “Is there any end to this development?”. Each generation of card seems to deliver more performance that we could ever have wished for. Then, within 6 months, we’re looking at the next big thing and all of the previous records lie smashed on the floor. OK, we can’t help it, we’re already drooling and can’t wait to get our undeserving Ultra White Lab Mittens on the Radeon HD 6970 so we can torture test it and render our verdict. Roll on Q4!
Smart/regular readers will know that this is an update of a much earlier piece we did… However, AMD fooled us with its renaming strategy – so we’re re-issuing because we believe the content is still very relevant. Got an opinion? Share it below!