Re: AMD Llano "fusion" discussion
Considering pure cpu performance, its no match for current sandybridge cpu's. A quad core is often beaten by sandybridge dual core i3 2100.
So its going to be hit with a select set of customers who don't want anything more than the igp and currently that set is pretty high in india.
So amd might be all set in achieving some milestones in terms of processor sales.
This is strictly speaking in a desktop context. But on laptops, we have to see how much extra battery life it actually delivers. I will be interested if these can actually beat an i3 2310 + radeon 6470m combo in terms of pure performance and deliver significant battery life.. Dell is currently providing its inspiron line @ 36k all.
Anyway guys here is the review:
Source- tomshardware
Okay guys i have summed up quite a few things about llano which i like to share :
Frankly speaking, in terms of processor technology (fabrication, performance per watt, & pure number crunching performance), amd is almost a generation behind intel. The core i3 2100 beating a phenom 2 955 be in almost all benchmarks is a testament to this fact.
Amd themseves said that its almost impossible to beat intel in a fabrication point of view. So they decided to use their strength which incidently was intel's weakness & planned to capitalize on that. Out comes
llano amd's newest kid which has promised a lot in terms of overall performance with some exceptional power mangement. So what is llano?
*img690.imageshack.us/img690/4844/apuandfusioncontrollerh.jpg
Llano is basically consists of 2-4 x86 cores + Graphics shader cores + Northbridge + FCH( fusion control hub a.k.a southbridge) all packed in one die. Amd calls this FUSION.
Now lets discuss the individual elements:
Cpu side:
*img59.imageshack.us/img59/5940/llanocpudieblocks.jpg
Basically llano's x86 cores are based on amd's previous k10 architecure employed in its phenom 2 and athlon 2 cpu's. Architecturally, they are same but llano cores have a 32nm fabrication process. Not only this but the l3 cache has been completely stripped down and l2 cache is doubled to 4mb. Now each core gets a dedicated 1mb l2 cache.
Amd claims that the cores give 6% more performance than the earlier ones owing to larger cache and an efficient
hardware prefetcher. The prefetcher actually loads intructions into the processor cache for computation. AMD enhanced the prefetcher intelligence with Instruction Pointer (IP)-based prefetching. IP understands the instructions accessing the memory and finds a specific pattern and load the instructions in the cache accordingly. The buffer sizes are also made larger.
All these things give an overall 6% IPC (instruction per clock) performance to llano's computational abilities over its predecessors.
Gpu side:
*img641.imageshack.us/img641/8383/llanofusiona8kh29622513.jpg
The gpu part is based on discrete class radeon 5570 (redwood) following the vliw5 architecture. Its codenamed SUMO and though its similar to redwood, has fewer displayports and uvd3 instead of uvd2. But amd has a small trick up its sleeve which they call
dual graphics.
This enables the on-die sumo gpu to for a bond with another discrete level radeon gpu and work in tandem. Its simply a crossfire but the fact is there need not be similarities in the gpu's architecture. Disimilar gpu's can also work together. For eg. a 6620g (apu) + 6630m (discrete gpu) will form 6690G2 (dual graphics). The llano apu' can also handle opencl codes which the traditional sandybridge cpu's cannot process on their own and needs a discrete gpu. But in a real world performance pov, the dual mode is actually having a performance drop and will be rectified in future driver releases. Another worth mentioning is that dual graphics mode work better in dx10 and dx11 mode only.
Apu:
There's something that differs llano from sandybridge cpu's and prevents them to be called a s FUSION parts. They are internal communication methods employed within llano.
There are five main components of Llano that have to communicate with each other: the CPU complex, the GPU complex, the northbridge, the traditional I/O block, and the DDR memory I/O block.
*img684.imageshack.us/img684/3264/llanonorthbridgeplumbin.jpg
Now we've seen cpu- northbridge interconnect before and its no different here. What's the difference here then?
That is two all new buses that are present between gpu - northbridge and cpu. The first one is called as
RADEON BUS which allows the gpu to have full access and bandwidth to system memory. It kind of provides a priority access for high bandwidth operations.
Second one is the most important and is responsible for this chip to be termed fusion. Its the
FCL ( fusion compute link) which allows the gpu to access the cpu cache minimizing bottlenecks. Previously, this communication was made possible using pci-e links. This also lends a big hand towards power efficiency.
Turbo core also finds its implementation here but slightly differs here than previous iterations because of the extra gpu here. Amd incorporated a hw module here called the APM (advanced power management). This consists of a p-state and a p-state manager.When its finds that there's more headroom, the state changes to p-boost and the clock speeds are incremented. But the gpu's clocks cannot be incremented but can only be decremented depending on the chips tdp levels.
Amd steady video
Here's an interesting result of amd's APP ( accelerated parallel processing) initiative and can be achieved using llano. Let's say you'sre shooting a marriage ceremony with a handycam. Now the human arm is prone to shakes and is never steady. This results in a shaky video. Now amd's steady video fixes all that and makes the resulting video completely steady for a better viewing experience. The hardware has to track video vectors and shakes, it has to compensate for the frequency of camera movement, and it has to add the missing pixels when the camera shakes off of the screen. The result is worlds better than poor source material. Now that is an innovation and amd's step towards gpgpu computing.
Performance:
Summing up the performance, the llano chip provides almost 56% more performance than a sandybridge i5 2520m with hd 3000 graphics in gpu intensive games. But with cpu intensive titles like dragon age origins, llano and sandy are neck and neck and we can owe this to llano's much weaker cpu.
In pure number crunching performance, again intel's sandybridge chips crush llano in almost all synthetic tests. Same can be said on productivity tests like winrar , 7zip etc.
Tom's hardware content creation tests also gave the same results with the entire adobe suite, blender and cinebench. Although llano was a bit better in the cinebench test which uses the gpu to show smooth viewport fps.
Even in media encoding tasks, sandybridge dual cores are well ahead. Remember that these are purely cpu intensive and intel even beats llano without using quicksync.
Though llano employs app ( accelerated parallel processing) to speed up computation in video encoding tasks by using its internal shader cores, its no match for intel's cpu based encoding and decoding. Turning quicksync on, turns up the heat and simply blows everything out of the water.
When it comes to content viewing, llano score more over sandybridge's igp when watching an hd movie. We can owe that to llano's superior gpu core. Though intel handle noise and skin tones quite nicely, it was poor in 2:2 Film Resolution cadence support, substandard contrast enhancement, and terrible chroma up-sampling and scaling performance.
So enabling video filters are better in llano than on sandybridge's igp. But a sandybridge + discrete gpu will do everything easily in the expense of battery performance.
Coming down to power, llano is a winner. A llano apu based laptop gave almost an hour extra backup than a sandybridge based laptop having intel hd graphics only. Now this is a big plus in the laptop community.
Ultimately it comes down to usage patterns to pick one. LLano shows strong gpu performance and sandybridge shows strong cpu performance. Ofcourse we know sandybridge when paired with a discrete gpu turns things around but sacrifices on power and battery. So llano can attributed with a decently specced system capable of giving a balanced performance with some exceptional battery performance. The only thing its lacking now is the cpu side performance but expect bulldozer based trinity cores to fix them.
In the end, fusion is the future of computing , a breakthrough technology and the start of a whole new era in the world of microprocessors.
In other words, its just the beginning.