Thu (thunng8.delete@this.hotmail.com) on September 28, 2013 8:52 pm wrote:
> Michael S (already5chosen.delete@this.yahoo.com) on September 28, 2013 1:11 pm wrote:
>
> > 3D Graphics - in some tests A7 more than twice faster, but in many other tests they
> > are about equal, may be 10% advantage to A7. Of course, we still don't know whether
>
> That's just plain false. And you are comparing a Tablet processor to a smartphone processor on a 4"
> phone with vastly different power usage. Also Bay Trail is on a more advanced process 22nm vs 28nm.

For tablet SoC, Bay Trail is relatively low power. So, I am not sure that power usage under majority of tests in question is vastly different. The only cases where I expect bay Trail to burn significantly more power than A7 are multithreaded CPU tests with all 4 cores active. As to process - why should we care?

>
> T-Rex 1080p off screen 15 25
> Egypt 1080p offscreen 30 57
> 3DMark Graphics test 13435 17669
>
> So 67%, 90% and 32% difference. Hardly 10% as you claim.
>
>
>

You misunderstood what I was saying. I never claimed 10% difference in geomean or something. I just said that it looks like there are two groups of tests that behave as if they measure different aspect of GPU performance. In one group A7 is more than twice faster than Bay Trail (actually, more like trice faster and almost as fast as A6X). In the other group there is little difference. It was the other group where I said that A7 is, may be, 10% faster.
Here are scores compiled from two latest Anandtech reviews. Bay Trail 3DMark numbers are in brackets, because they are measured with previous version of the benchmark suit. However, that should not make a big difference.

Benchmark                                                        A7        A6     Bay Trail

GFXBench 2.7 Fill Test           (offscreen 1080p)               3,304.5  1,670.4  1,165.6   

GFXBench 2.7 Triangle Throughput (offscreen 1080p)               68.20    94.90    69.80

GFXBench 2.7 Triangle Throughput, Fragment Lit (offscreen 1080p) 34.80    78.90    61.20

GFXBench 2.7 T-Rex HD                                            37.0     14.0     11.0

GFXBench 2.7 T-Rex HD (offscreen 1080p)                          25.0     6.9      16.0

GFXBench 2.7 Egypt HD                                            53       41       30.0

GFXBench 2.7 Egypt HD (offscreen 1080p)                          57       30       41.0

3DMark Unlimited - Ice Storm                                     13,869   5,656    [14,500]

3DMark Unlimited - Graphics                                      17,669   5,215    [13,435]

3DMark Unlimited - Graphics Test 1                               102.0    27.2     [59.50]

3DMark Unlimited - Graphics Test 2                               61.6     19.4     [57.40]

3DMark Unlimited - Physics                                       7,913    8,029    [20,070]

3DMark Unlimited - Physics Test                                  25.1     25.5     -

Basemark X - On Screen                                           27.374   18.563   6.399

So, in few tests Bay Trail is not just not slower, but much faster, than A7.
Arguably, it is possible that Physics score does not test GPU at all. But what about "Triangle Throughput, Fragment Lit"?