Core i3/i5 review @ anandtech

[lothar]

Lucid Hydra is overrated.

 

[IntelUser2000]

Actually he is telling half the truth by mixing and spinning it with what the REAL GPU vendors do, offload as much work from the CPU to the GPU through their drivers. Intels IGP is so crap that it offloads to the CPU to help bring up the numbers in vantage and gaming.

All revealed by mr.Schuette at LostCircuits.

Enjoy.

"The bottom line here is that depending on the contents/workload, and maybe some driver optimizations, Intel has found a way to offload a major portion of the iGFX workload to the CPU cores, which reflects in a substantial performance increase. If the CPU - for whatever reasons - cannot be recruited by the iGFX, graphics performance drops substantially. For the record, in the case of the Phenom II X4 965 BE/ 785G combo, the CPU utilization stayed at about 0% during the entire course of the benchmarks."


http://www.lostcircuits.com/mambo//...task=view&id=74&Itemid=1&limit=1&limitstart=5

http://www.lostcircuits.com/mambo//...ask=view&id=74&Itemid=42&limit=1&limitstart=6

http://images.nvidia.com/blogs/ntersect/feature_image09.jpg

Bit late, but Lost Circuits has this wrong. More comprehensive testing shows only applications that is able to use 4 threads is faster with Software VS. All other times the Hardware is SIGNIFICANTLY faster than Software.

Here's some examples:
3DMark2005: Software-2245, Hardware-3996

World in Conflict Demo

Software
800x600 Lowest-22
1024x768 Medium-10
1280x1024 Very High-5

Hardware
800x600 Lowest-50
1024x768 Medium-20
1280x1024 Very High-5

Hardware is almost 2x faster than Software. Neither WiC or 3DMark2005 is well threaded, if at all.

Oh, and it doesn't really matter how you achieve the performance. If Intel can use their CPUs to assist the GPU and beat competitors, more power to them. GPUs will offload CPU code, why not the other way around? They are largely not interchangeable, but there are bits of 3D code that will run faster on a CPU and there are bits of general code that will run faster on GPU.

Isn't the theory of Fusion sharing resources between the two to achieve ideal performance? AMD will very likely do the SAME thing Intel does with Clarkdale and use CPU to offload some 3D code for its iGPU.

 

[lopri]

Oh, and it doesn't really matter how you achieve the performance.

I agree with that. But after all that IGP testing, turning around and installing HD 5870 for power consumption comparison (without explicit explanation) is deceptive. And that's what Anand does without a hint of shame these days.

 

[IntelUser2000]

Yea, I don't know why they did that. Though most of the testing was done on the HD5870, so I guess that makes sense.

Another thing: They don't do TRUE dynamic sharing between CPU and GPU yet. It would be good if CPU can BOOST the GPU with vertex processing when required, but all it can now do is replace it. I'm guessing we won't see that until the future. It'll be kinda like hybrid xfire but instead of gpu+gpu its cpu+gpu.

 

[soccerballtux]

I agree with that. But after all that IGP testing, turning around and installing HD 5870 for power consumption comparison (without explicit explanation) is deceptive. And that's what Anand does without a hint of shame these days.

That's a good point.

They did it to make it look less bad for AMD, because we need AMD around.
I don't care about power though, so it doesn't bother me.