Friday, June 10, 2011

Retail desktop Llano APU tested!

As per reader's suggestion ,I'm adding numbers for Sandy Bridge (2600K,default) next to Llano's stock results. Sources for 2600K results : link1,link2.
Found some more real gaming (fps) results for Llano ES(not retail) here. Massively faster than HD3000.Check bottom of this blog post for exact numbers.

It seems someone got a hold of retail A8 3850 part and tested it here. Thanks goes to dresdenboy for the link.
Part works at 2.6Ghz and can turbo up to 2.9Ghz (with all cores loaded/GPU idle? correction: this model has no turbo and works at 2.9Ghz). GPU part  works @ 600MHz and features 400SP, so on par with 6550M discrete part.

Now on to results!

User ran a set of  Futuremark tests: 3dmark11, Vantage and 3dmark06. User also managed to OC the part ,both the CPU and GPU portion to some rather high levels,just via serial bus tuning(45%). He used air cooling. Final OC speeds are 3.77GHz(45%  30% OC) for CPU and 870Mhz(45% OC) for GPU,DDR3 was also OCed to 2320Mhz(45% OC).Memory OC is very important since GPU still depends on memory BW and it's important for ensuring GPU performance scales linear with (GPU)clockspeed increase.

On stock settings he got:
3Dmark 11 P1148 <=in line with discrete 6550M part? no data on this
3Dmark Vantage P4400 <=in line or even better than discrete 6550M part!
3Dmark 2006 7650 <=in line or even better than discrete 6550M part!

Compare these with SandyBridge 2600K stock results in 3 tests:
SB 2600K 3dmark11 :  FAIL (no DX11 support) <= Llano A8-3850 wins by deafult
SB 2600K 3dmarkVant.: 2164 <= Llano A8-3850 wins by a factor of ~2x (4400/2164)
SB 2600K 3dmark06 : ~5000 <= Llano wins by a factor of 1.53x (7650/5000). This test is using a different formula to calculate end score and CPU scores has a great deal in this formula.Llano's GPU actually has a SM3.0 score of 3162 while SB 2600K gets 1926,so pure GPU difference is ~64%.
So as the shader load of modern games increase(3dmark11>3dmarkV>3dmark06 ,level of shader load in tests) ,Llano will command a lead of 2x over HD3000 graphics and probably more when we take into account Aniso filtering and AA.

Where are all those XtremeSystems News section posters who claimed Llano's GPU will suck so bad because it is sharing a lowly Dual Ch. DDR3 memory controller with 4 Hound cores? :D They also claimed AMD is wasting die area on all those 400SP since there is no way all those Stream Processors will have work to do as they will be BW starved all the time.Well,according to these results,not really. iGPU performs exactly like a discrete part***

That's right,Llano has no SandyBridge's "illness" so that it depends so much on the shared cache or shared memory controller.It all works fine and memory BW allocation for GPU and CPU works great.

Now on to OCed results :
3Dmark 11 P1591 (45% GPU OC => 38.5% result increase! )
3Dmark Vantage P6160 (45% GPU OC => 40% result increase! )
3Dmark 2006 10492 (45% GPU OC => 37.1% result increase! )
note: CPU is OCed by 30%,from 2.9 to 3.77Ghz,DDR3 memory is OCed by 45%

As can bee seen,air cooled and OCed by 45%,desktop Llano sets all the new IGP records.And the delta between it(even on default!) and SandyBridge desktop part is huge. Top SB 2600K OC result in Vantage is P3988 while Llano air cooled and OCed for the first time easily  gets 53% higher score. There is just no comparison.

Some pictures from the original thread,enjoy:

***In order to see a high level representation of Llano's magic dust,go here=> garlic and onion interfaces at work.

UPDATE #2 results:
"Hawx 2 DX9 1920x1080

Intel HD3000 1100 Mhz= 26 fps
AMD HD6550D 600? Mhz 46 fps +77%

Streetfighter 1920x1080

Intel HD3000 1100 Mhz= 22.75 fps
AMD HD6550D 600? Mhz 50.32 fps +121%

Resident Evil 5 Directx9

Intel HD3000 1100 Mhz= 15,5 fps
AMD HD6550D 600? Mhz 29,0 fps +87%

Resident Evil 5 Directx10

Intel HD3000 1100 Mhz= 14,5 fps
AMD HD6550D 600? Mhz 27,4 fps +89%
Identical game settings. Intel driver used."


  1. All those numbers demand a table of results or even a bar chart! :-)
    What does stock SB 2600K get?

  2. Hello :). I will try and later add some table with results. As for SB 2600K,i updated the post with SB 2600K default scores in all 3 tests(or 2 since it has no DX11).

  3. This is a great article! Also, very well referenced around the net already!

  4. Thanks for the comment Erick! I tried my best to summarize Llano's gaming performance. As things stand,overall, Llano is much better Fusion-type product than SandyBridge.It will have somewhat slower CPU cores (that will be plenty fast anyway) and will have massively better GPU.Also ,icing on the cake is the Hybrid Crossfire support,a thing SB can't do.So add 60% on top of any difference stock Llano gets over stock 2600K/2500K. Fun times ahead :).