The Zacate is not and I repeat this, NOT as fast as a Core 2 Duo or Core i5 MacBook Pro.
I should have been a bit more qualified. I shouldn't have said "current i5s", I should have said some i5. If you are trying to drag in all of the "all the power you want to burn" i5s and C2Ds then no. Likewise, if want to get into a CPU based raytracing and/or double precision floating point number crunching exercise then no. There is only one memory channel and a limited FP pipeline. I'm sure those can be saturated if focus on pushing data through them. Likewise a large enough screen attached to the GPU will probably cause a drop in throughput performance also. [ The AMD demos were on 1024x768 whereas many 11.6 screens are 1366x768 . Also many folks will want to use the mini display port to hook to larger monitor. It won't be a big screen, gamer machine ]
However, the article that has been pointed to a couple of times over at
Anandtech pitted a i5-520M ( unless there is major typo mistake there, that is 'M' not 'UM'. ) against a Zacate. So a couple of points. Only a 'UM' i5 would work for a MBA. Those are substantially
slower than the M models. So when the Zacate runs about at parity with the 520M (after Anantech puts in some fixes) in the web benchmark is indicative that this is a creditable choice. It isn't as "dog slow" as your spinning it to be. The entry level MBP 15" uses a 520M. If the Zacate isn't that far from matching performance with a 'M' , then certainly it is going to be a creditable competitor to the vast majority of the 'UM' line.
Second, I found a reference somewhere on net that these were 1.8GHz Zacate. That is probably the high end of the range. The low and mid range Zacate and Ontario (9W envelope focused) variations of Bobcat probably would suffer with substantially worse numbers. I sure there will be later benchmarks using those more mainstream implementations that get bet bad by mainstream Core-i implementations. The comparison in the article shouldn't be stretched to infer something about the high performance range.
One problem for Apple using this is that perhaps would not be able to do a "better" config with a small clock bump. It has similar issue the C2D ULV has in that Apple is aready at the top end. There is no were to go for a clock speed bump.
It is FASTER as it in some processes that benefit from the GPU + CPU combo, but in a pure CPU sense, the Zacate will most likely fall behind...
Sure, but the vast majority of apps that average folks use involve a usage of both GPU+CPU . Browsers (as in the article's IE tests ) , Office (writing/presentations ) , etc. Similarly, not many folks uses ultramobiles to do run heavy computational jobs. Sure someone who was pressing their MBA into doing AutoCAD like jobs will suffer if switched to Zacate, but is that really the primary market the MBA is aimed at ?
One of the features of the new Zacate APU is that it will be able to really really really underclock both the GPU and CPU to increase battery life depending on how demanding you are using it...
That's a bad feature would not want in a ultramobile offering ?
Intel has done this with it's ULV processors but I think that AMD is trying to do it to a maximum degree with this processor.
As I said previoulsly the Zacate only has to compete with the old SL9400/SL9600 ( not quite ULV , but definitely "Low V" ) offerings and turn in incrementally better performance. The graphics results are obviously substantially better than anything the Core-i 3/5/7 UM series has to offer. That aspect is purely in the loss column in that competition. To 'win' all the Zacate really needs to do is at least 'tie' the ULV offering on the core of the CPU performance metrics. If slightly ahead it should be game over.
The only "win" for Intel would either be a shrunk and speed bumped C2D (so Apple can use Nvidia graphics ) or some Sandy Bridge offering (which has better graphics that may be good enough. ). The latter isn't coming until well into next year. The former just doesn't seem likely ( unless Apple, and others, pushed Intel into a corner. )