Although SiS introduced the first dual-chip platforms (processor + chipset) to the market over a decade ago, they hadn't been very popular until recently. Even moving memory controller into processor (done by AMD in 2003) didn't help much as manufacturers still preferred the cheaper and more flexible triple-chip solutions. But Intel, having moved memory controllers into its processors as well, knuckled down to reverse the situation. Firstly, they rolled out the LGA1156 platform that only had a Southbridge, because PCIe controller and then GPU were all moved into processor. A few months later, the Pine Trail platform underwent the same change.
In turn, AMD continued to use the triple-chip design, because being more flexible it allowed building any configurations — from multi-core processors with integrated graphics and simple Southbridge to mediocre processors with excellent multi-GPU support and matching expandability. That was good for the desktop market, because motherboards of that class had a lot of space. But for the mobile market reducing the number of chips was more reasonable, and that's what manufacturers are going now. So, the Brazos platform is the first to enter the mobile market. It doesn't offer anything radically new — AMD moved PCIe controller and GPU into processor as well. Note that they started combining CPU and GPU before Intel did, but the latter was the first to introduce ready products.
So it's not a surprise that AMD positions the platform not just as an example of unification, but as a completely new class of devices it calls "Accelerated Processing Unit" (APU), where processor and graphics cores form heterogeneous multi-core architecture, helping each other in handling various tasks, including those unrelated to graphics. Yes, it's the very GPGPU people have been talking about so much recently. And it seems to be as useful for common tasks as all that hot air for a windmill. All the more so that AMD doesn't have much time in reserve: integrated GMA HD could decode high-definition video already (one thing people do need). And it wasn't even an APU, because GPU and CPU were on separate dies. And the newer GMA HD 2000/3000 of the LGA1155 platform could assist processor in transcoding of the same video, too.
In other words, manufacturers still tread the same path, occasionally coming up with nice marketing names which have no real meaning. But that isn't so bad. The new solution works and that's what's important. Besides, the novelty from AMD is more advanced in terms of graphics (it supports DirectX 11), is also more powerful theoretically (has 80 stream processors), and Intel Atom, its primary rival, still doesn't even have something like GMA HD, not to mention the newest GMA HD 2000/3000. The new platform has a bit more interesting processor part, too, because it's an out-of-order device. Finally, thanks to finer process technology, Zacate and Ontario are more compact than Atom is and have comparable power consumption.
Now let's proceed from theory to practice as we have a sample to test. In this review (one of many we plan) we'll compare the performance of Zacate (namely AMD E-350, the highest-end model in the series) with that of a higher-end Atom and a couple of dual-core Celerons (one of the first and one of the best today).
Testbeds
CPU |
E-350 |
Atom D525 |
Celeron E1400 |
Celeron E3500 |
Core |
Zacate |
Pineview |
Allendale |
Wolfdale-2M |
Process technology, nm |
40 |
45 |
65 |
45 |
Core clock rate, GHz |
1.6 |
1.8 |
2.0 |
2.7 |
Multiplier |
- |
13.5 |
10 |
13.5 |
Cores/threads |
2/2 |
2/4 |
2/2 |
2/2 |
L1 cache, I/D, KB |
32/32 |
32/24 |
32/32 |
32/32 |
L2 cache, KB |
2x512 |
2x512 |
512 |
1024 |
Integrated graphics |
Radeon HD 6390 |
GMA 3150 |
- |
- |
RAM |
1xDDR3-1066 |
1xDDR3/DDR2-800 |
Depends on chipset |
Socket |
BGA413 |
BGA559 |
LGA775 |
LGA775 |
TDP, W |
18 |
13 |
65 |
65 |
So, as stated above, E-350 is the main character, Atom D525 is its primary rival, the old Celeron E1400 is yet to be outperformed by the newest Atom, and Celeron E3500 is a fast low-end desktop processor. According to our estimates, the latter should win. But we'd like to see by how much exactly low-end desktop CPUs outperform these Atom and Zacate processors.
|
Motherboard |
RAM |
BGA559 |
Gigabyte D525TUD |
Kingston KVR1333D3N9K3/6G (1x800; 6-6-6-15) |
LGA775 |
Gigabyte G41M-ES2H (G41) |
Crucial Ballistix BL2KIT25664AA80A (2x800; 4-4-4-12) |
BGA413 |
Sapphire Fusion Mini E350 (AMD A50M) |
Kingston KVR1333D3S39/2G (1x1066; 8-8-8-19) |
For obvious reasons, it would be impossible to test all these processors on the same testbed. Especially as such devices are usually soldered to motherboards. But we tried to reduce the number of motherboards involved to a minimum. In particular, both Celerons had to do with an older board based on the formally modern but functionally outdated G41 chipset (at least it's cheap). But the test method used today doesn't involve graphics, because we're only interested in the processor part (graphics would require a dedicated review). Next, we used a mobile hard drive Seagate Momentus 5400.5 ST9320320AS (2.5-inch, 5400 rpm, 320GB, 8MB cache, SATA-300) to negate its potential influence. We used different memory modules simply because of system requirements, but at least we installed 4GB in every testbed. That would seem to much for a netbook or a nettop, but it's cheap and it also goes better with our test method.
Tests
For your convenience the results are represented in percent, 100% being the result of AMD Athlon II X4 620. Absolute results are provided in this Excel spreadsheet.
We had to remove four groups of tests, but all scores on the diagrams below, excluding the overall one, comply with those of our full-fledged tests, so you can quickly compare chosen processors yourself. That's actually why we had to remove certain groups of tests completely instead of fine-tuning specific applications. The latter could've been more useful in a restricted sense, because some of the programs we didn't use would've probably worked with Atom, not to mention Zacate, but it also would've made results incompatible in the wide sense.
Image editing
Three of the four applications do well with the two cores of E-350. Photoshop prefers the higher efficiency of two threads to the lower efficiency of four threads, so AMD E-350 outperforms Atom D525 by 20%. A good result for a netbook, but a mediocre one for a nettop — E-350 is outperformed by the three-year-old low-end Celeron by 1.5 times. Well, Atom is 2 times slower, but that's all the same — a newest nettop can't even catch up with an old low-end desktop. Sad. On this background, Celeron E3500, one of the slowest desktop processors, is not just faster but fundamentally faster.
Data compression/decompression
Atom wins in 7-Zip, thanks to HT, but loses in the other two tests, even despite the larger L2 cache. So it lags a bit in general. Low-end desktop solutions are still unreacheable, as you can see. Well, performance increases slowly in the low-end segment (even slower than in the mainstream and high-end segments), so if both Intel and AMD try hard, they might reduce the 'time lapse' between solutions above down to two or three years.
Compiling
This test favors cache (as well as everything else actually), so Celeron E1400 is in trouble with its 512K L2 cache per both cores. Other processors have 1MB of L2, although E-350 has it split into two parts (still a cache, though), and Atom processes four threads at the same time. Considering all that, competitors are still far behind Celeron E3500, but are closer to E1400. The results are almost the same — Atom wins a few seconds (out of 30 minutes), but that doesn't affect the diagram.
Write a comment below. No registration needed!