Setting Performance Expectations

AMD provided this slide of PCMark Vantage and 3DMark Vantage performance of Brazos compared to its existing mobile platforms (Danube and Nile):

If you look at the PCMark Vantage numbers you'll see that AMD's E-350 provides roughly the same performance as an Athlon V120. That's a single core, 45nm chip running at 2.2GHz with a 512KB L2 cache. Or compared to a dual core processor, it's within striking distance of the Athlon Neo K325 which features two cores running at 1.3GHz and 1MB L2 per core. The GPU performance however tells a very important story. While AMD's previous platforms offered a great deal of CPU performance and an arguably imbalanced amount of GPU performance, Brazos almost does the opposite. You get a slower CPU than most existing mainstream platforms, but a much better GPU.

In the sub-$500 market, you're not going to get much in the way of a discrete GPU. What AMD is hoping for is that you'll be happy enough with Brazos' CPU performance and be sold on its GPU performance and total power consumption. From AMD's standpoint, there's not much expense involved in producing a Zacate/Ontario APU, making Brazos a nice way of capitalizing on mainstream platforms. The 75mm2 die itself is smaller than most discrete GPUs as well as anything Intel is selling into these market segments.


AMD's Zacate APU, 19mm x 19mm package, 413 balls, 75mm^2 die

The Comparison

Brazos, like Atom, will fight a two front war. On the one hand you have the price comparison. The E-350 will be found in notebooks in the $400 - $500 range according to AMD. That puts it up against mainstream notebooks with 2.2GHz Intel Pentium DC and 2.26GHz Core i3-350M processors. Against these platforms, Brazos won't stand a chance as far as CPU performance goes but it should do very well in GPU bound games. I've included results from a 2.2GHz Pentium dual-core part (1MB L2 cache) as well as a simulated Core i3-350M in the mobile IGP comparison.

The other front is, of course, the ultraportable space. Here you'll see the E-350 go head to head with dual-core Atom, Core 2 ULV and Arrandale ULV parts. AMD's CPU performance should be much more competitive here. From this camp we've got the Atom D510 (close enough to the N550) and a simulated Core i3-330UM. The expectations here are better CPU performance than Atom, but lower than Arrandale ULV. GPU performance should easily trump both.

Introduction CPU Performance: Better than Atom, 90% of K8 but Slower than Pentium DC
Comments Locked

207 Comments

View All Comments

  • mino - Tuesday, November 16, 2010 - link

    Yeah, as well as some fail to see the forest for the treas.
  • Anand Lal Shimpi - Tuesday, November 16, 2010 - link

    The focus here is obviously Zacate, which is why the text that accompanies graphs is centered around comparisons with reference to Zacate.

    As far as the gaming benchmarks go. I tried to put together a varied list of titles to show performance across a spectrum of game types. Starcraft 2 was the largest non-FPS game release this year, it of course had to be included. DAO simply provided another datapoint for an RPG showing how CPU limits can appear even in 3D games.

    I don't think anyone is arguing that Intel's HD Graphics are faster (note the title on the desktop IGP comparison page). But I do believe it's important to show both sides of the coin.

    Once final hardware is out we'll definitely spend more time with the platform and run through an even wider range of games and benchmarks. This merely serves as a preview until then.

    Take care,
    Anand
  • mino - Tuesday, November 16, 2010 - link

    "Once final hardware is out we'll definitely spend more time with the platform and run through an even wider range of games and benchmarks. This merely serves as a preview until then."

    That is expected and waited for! :D

    Just think about splitting at least the conclusion as I have mentioned above. ;)
  • Dark_Archonis - Tuesday, November 16, 2010 - link

    I find it more strange that you're criticizing the gaming benchmark selections. Consumers don't talk about "this game is very CPU intensive" when they typically buy games.

    The average Joe does not care how CPU intensive Starcraft 2 or Civilization 5 is. They just want to play it.

    These are both popular games, so their inclusion in the benchmarks is perfectly acceptable.

    Crysis is quite CPU intensive. Are you going to criticize Crysis as well for being CPU intensive?

    I guess any games that show Zacate as a poor performer should not be used according to you right?
  • Jamahl - Tuesday, November 16, 2010 - link

    I have no issue with Starcraft 2 being used.

    I do have an issue with Starcraft 2, Civ 5 and Dragons Age being used however. These are supposed to be gaming benchmarks, yet the majority of them are far more cpu dependent.

    This is not the norm in gaming, however it is the norm in Anands benchmarking.

    You would be very hard pressed to find two games more cpu dependent than Starcraft 2 and Dragons Age.
  • Dark_Archonis - Tuesday, November 16, 2010 - link

    I will bet you anything that Diablo 3 will be very CPU intensive.

    As I mentioned, Crysis is CPU intensive. What do you have to say regarding that?

    While these are only a few games, these are very popular games.

    What exactly do you mean that it is not "the norm"? Do you mean to say that most sites don't include Starcraft 2 or Dragon Age in benchmarks? I don't see why people would have issues with this. It illustrates a variety of games that stress different parts of a computer.

    RPG and strategy games are typically quite CPU intensive. Also, they are rarely included in gaming benchmarks.

    I for one applaud Anand for including them, as far too many sites lack these kinds of benchmarks.
  • Jamahl - Wednesday, November 17, 2010 - link

    Crysis is cpu AND gpu intensive. The point is most benchmarking suites would not have a MAJORITY of cpu intensive games. There's a reason why people actually spend a lot of money on gpu's.

    Did you even read Anands preview? Let me help you understand as you're making a great job of failing to.

    Page 2 - "Setting Performance Expectations"

    You get a slower CPU than most existing mainstream platforms, but a much better GPU.

    The expectations here are better CPU performance than Atom, but lower than Arrandale ULV. GPU performance should easily trump both.

    Then what does he do? Well lets look at the gaming benchmarks shall we?

    First of all, lets look at the comparison benchmarks.

    http://www.anandtech.com/show/3871/the-sandy-bridg...
    http://www.anandtech.com/show/2952/2

    So the 5450 was benched using a i5 661 and an i7 920 - And you wonder why Bobcat can't keep up? To see if it is actually close to discrete graphics performance, it should have been benched on a 1.5ghz athlon for a fair comparison.

    1 in 3 of the desktop tests was Dragon age - a massively cpu bound game which hugely favours intel cpu's over AMD's...and he benched it in comparison to a 5450 with two fast intel cpu's. Not right however you look at it. MW2 is not particularly heavy on graphics either so basically the only true graphical test is Bioshock 2.

    Next page

    In the mobile gaming comparison, he's gone and added Starcraft 2 - TWICE.

    Lets see what we got then -

    MW 2 - 50/50 split cpu/gpu
    Bioshock 2 - gpu
    Dragon Age - cpu
    Warcraft - 50/50 split
    SC2 TWICE - (yes even the gpu test is more like cpu obviously looking at the results)
    I'm not even gonna talk about using Civ 5. Civ 5 got a huge question mark over it's reliability in the fallout over the 6800 benchmarks, then he goes and uses it here?

    So...yeah. ONE pure recognised gpu benchmark used in all of those tests compared to 3 or 4 cpu gaming benchmarks.

    It's really very simple. benchmark 20 games at random and Bobcat will beat any intel integrated graphics by around 50% in 15 of them. Anand has managed to make it lose more often that not in the very, very few games that massively favour cpu's.

    The fact is, he knows this or should. He said at the start what to expect, then went and benchmarked what he knew would happen.
  • techworm - Tuesday, November 16, 2010 - link

    all of us know anandtech is intel PR site so no complaint
  • nitrousoxide - Tuesday, November 16, 2010 - link

    Watch your words, unless you can give evidence that Anand is indeed an Intel PR.
  • silverblue - Tuesday, November 16, 2010 - link

    Do me a favour and step back before making judgements.

    If Anand was biased towards Intel, why would AMD even entertain him with an offer to benchmark their newest CPU, let alone let him publish the results? He already said he was time constrained so you can't very well do everything. Moreover, he did mention that this was a "preview" - this isn't final silicon, it's an engineering sample, and that the final product should use less power.

    You should be grateful that there's a preview at all.

    It is a shame that the chip seems bandwidth starved, but I don't think AMD ever intended for it to be a triple-channel, SMT-enabled, out-of-order ULV solution - just think of the cost. As long as it does what you want of it, isn't that its point? Atom was "just enough" for browsing, basic media and light workloads, yet Brazos is at the very least an improvement in IPC, and substantially better if you take into account gaming, movies and so on. If Anand truly wanted to give it a bad name, he wouldn't have benched so many games AND highlighted where the chip's true strengths are. He also mentioned that most games should be GPU bound and thus will extol the virtues of the Brazos architecture - how very anti-AMD of him (!).

Log in

Don't have an account? Sign up now