LAST WEEK'S AMD Technical Forum & Exhibition (TFE) in Taipei, now a regular annual event, was an interesting chance to get together with the key company product and marketing personnel, hear more about upcoming products and check the pulse of the all important local Taiwan partners.
Aside from the well covered Radeon HD6800 series - the first 'AMD, not ATI' branded Radeons - and the associated fuzzy numbering issues - yes, as you'll see in our review, the HD6870 is slower, but cheaper, than the HD5870 - there were many other things to take note of and put into the bigger picture about AMD and its position now and into next year.
First of all, the level of energy and confidence among the AMD team was higher than in the past five years at least, as far as memory serves. It wasn't just about the new CPU cores coming out finally after a long hiatus. In the end, that costly and at times painful ATI acquisition did bear fruit. Not only did the GPUs save AMD as a going concern at the time when its CPUs were all but lost, but the first results of the Fusion strategy were out for all to see.
Not only do low-end Ontario and Zacate CPU-GPU - sorry, APU - combinations run well, but so does the higher end quad core Llano desktop part, which is essentially a Fusion part using an improved Quad core 32nm process Phenom with an HD5500-class GPU on-chip. According to unnamed AMD staff there, the latter can overclock quite well too.
AMD also says that, while the Sandy Bridge CPUs and GPUs communicate through L3 cache for low latency, the GPU is inherently different from the CPU in its memory access behaviour, with more importance placed on streaming bandwidth than low latency.
Therefore, AMD believes its approach of linking the CPU and GPU on the Fusion processors through the memory controller, not cache, is more appropriate as it prevents the GPU from thrashing the CPU caches due to the very different access patterns between the two. AMD staff said that Intel most likely didn't know how to design a versatile memory controller that would handle both types of access efficiently, so it ended up using the L3 cache as a buffer of a sort.
When I highlighted that the upcoming mainstream Sandy Bridge CPUs still might have quite a bit faster clock for clock cores compared to current Phenom core performance, AMD told me it will welcome a Sandy Bridge versus Llano face-off in a couple of months, when both are out, and that both the individual component benchmarks as well as overall experience tests should be looked at. You can see the proudly shown Llano wafers here:

As for the low-end, AMD has elevated its 18W Zacate dual core from competing with Intel's Atom chip to fighting even Intel's mobile Core i3 and i5 parts in some cases. AMD is confident the combined CPU-GPU performance will stand the test, especially for thin and light high performance notebooks with multitouch tablet and full 3D functionality. Talking about tablets, the 9W Ontario Fusion part could be the heart of a really lovely Ipad competitor, bringing that same multitouch and real time 3D capability to a 10-inch 1440x900 or higher resolution one pound SSD tablet, for instance.
Like Costumes, Facts Seem to Need Look Into Whoms' Inside....Floating Point refers to One Memory address Areas & Limited in number of decimals can handle, Any Excess goes to Far off memory.
Double Floating Point has TWO Memory Address Points or areas, each unique. First from 32 Byte String or 8 Bits, Then 64 Byte String, appeared two Comaprissons unitss in Double Floating Point. Each Completely Seperate. So 8 times 8 decimals possible ,Not just 8+8. Now from 128 bytes or16 Bits to 256, Make machine seem like Doubled Floating point, Yet long way from 32 bit times 32, from lessr or todays 32+32. 6870 is described as floating Point, Not Double Floating Point.
Thats Difference of about 10 to Ultee'.
So really 5770 with twice floaters, floating about instead of Whopper. Test Grill Marks & See Why Price Is So ?Low.
vondrashek about trillions transactions latter.
Reading another article AMD said that at idle with the new Flex FP scheduler and other enhancements that the power usage would be 2% of the max power utilization. That means that the 9w version will idle at .18w and max at 9w, that is quite a range and in most cases people are surfing online or doing non-intensive work so the 1-2w area is most likely where most people will be, that is cpu and gpu, which could make it competitive with ARM processors.
@Steve T true, atleasts not a tablet like the ones we see with ARM processors, but if you want a fully capable Windows based tablet, it shouldn't be too hard to design it so that we can fit in a 9W APU in there, considering how small Netbooks are working with the same.
This marks an interesting point in AMD's development, they have a dedicated product for mobile market, server market, and soon desktop market, well ok the last one they still do the Phenom II aren't dead yet, but I look forward to Bulldozer as it hopefully will revolutionize the Desktops
You still aren't getting it. ARM based systems like the Apple A4 or the Snapdragon contain IGPs and STILL only draw about 1W at full chat. By the time you've put a display, WiFi, audio etc on top of an Ontario it will be maxing out at 11-12W, something like 4 times higher than the iPad or Galaxy Pad.
An IGP-less option isn't useful (you still need a GPU). A 5W version is still 5 times more power hungry than the ARM systems. Having your 10 hour battery life fall to 2.5 if you try to play a game isn't useful (nether is the heat, which will probably need an active fan).
Bobcat can go all the way down to 5W with integrated graphics and there's also an IGP-less version that can go all the way down to 1W (if I'm not wrong).
the Ontario is far too power hungry for a tablet (at least in the same class as the iPad or Galaxy Pad). They need to AVERAGE 2.5W including the screen, radio and audio systems.
Nebojsa: I think you're wrong when you state that AMD will cede the overall performance crown to Intel this time around. We still don't know some necessary info regarding Bulldozer's overall performance and I'll suggest you take a through read about Buldozzer's Flex FP: http://blogs.amd.com/work/2010/10/25/the-new-flex-fp/
It's too premature to say who will lead who next year. All this stunt of yours makes you look like an intel cheerleader.