Toshiba T235D: AMD's 2010 Ultrathin "Nile" Platform
by Jarred Walton on August 23, 2010 11:00 PM ESTGeneral Performance – Targeting CULV
Our performance testing begins as usual with PCMark Vantage and 05, two comprehensive system benchmarks that stress all areas of performance. As such, it's not only the CPU that but also the hard drive, memory, and graphics chip that determine the result. To go along with PCMark, we have several other general performance tests that focus more on CPU performance. We've included results from a variety of ultraportable laptops for reference. We also have the last gasp of NVIDIA's IGP represented by the old Dell Studio 14z, with a 9400M and P8400 CPU. It doesn't reach the same level of efficiency in terms of battery life (thanks to the non-ULV processor), but keep an eye on its results elsewhere and you can mourn its passing. Intel has ditched the FSB architecture (not a problem, since the FSB was old and in need of an update), but in the process that kills NVIDIA's chance for a current generation IGP supporting modern Intel CPUs.
Right off the bat, we can see where the Nile platform (K625) improves over the old Congo offerings. Compared to the slightly faster clocked (1.6GHz) L335 in the Wind U230, the K625 (1.5GHz) in the T235D is at worst equal to Congo (i.e. Peacekeeper), and at best it's around 35% faster (PCMark Vantage and Cinebench). For typical tasks, then, somewhere between 15 to 35 percent faster is a pretty good boost in performance. And Atom is naturally down in the cellar, providing less than half the performance of K625.
Comparing the T235D to the old Intel CULV platform, represented by Acer's 1810T with an SU7300 (1.3GHz), things are closer but the K625 certainly holds its own. Peacekeeper, PCMark Vantage, and x264 (second pass) are basically tied, while Cinebench gives AMD a 7-8% lead and PCMark05 is a 15% lead. Clock for clock Core 2 Duo still looks like it has a small lead, but it's also important to remember that CULV is saddled with a really slow IGP (more on that next). Of course, the competition now isn't between CULV and Nile; instead we have to look to Intel's newer Arrandale ULV parts.
For that comparison, we have the Alienware M11x R2, but with the GeForce GT335M forced off. In this case we're looking at what is practically the top Intel ULV part (the i7-660UM is slightly faster), so this certainly isn't a "fair" comparison. An i7-640UM will run around $300 while the K625 appears to cost closer to $150. Obviously, the M11x R2 is a far more expensive laptop, and the performance bears this out. It's 70% faster in PCMark Vantage (helped no doubt by the 7200RPM HDD), but even when other peripherals aren't involved the i7-640UM is substantially faster than the K625. Cinebench, Peacekeeper, and x264 (second pass) all have the M11x leading by 70 to 75 percent. Intel's Turbo Boost does its job and keeps clocks well above the base 1.2GHz even under heavy loads. So not surprisingly, the expensive i7-640UM blows the K625 away in raw computational performance.
Given the price disparity, it's important to look at other options. We just received an ASUS UL80Jt with an i3-330UM (overclockable, just to make things interesting), and we'll have to wait for that review to get into other Arrandale ULV options. However, we can say that initial testing suggests a stock i3-330UM (1.2GHz but with no Turbo Boost) will be similar to the K625 in lightly threaded loads, while maintaining a 10 to 20 percent lead in heavily threaded benchmarks where Hyper-Threading comes into play. Perhaps even more interesting is that at first glance it appears battery life will be somewhat comparable between the two platforms. When we start to talk about pricing, AMD really has a case to make, but we'll save that for the conclusion. Suffice it to say that in terms of pure application performance, AMD's Nile platform is at least a viable alternative to Intel. It's slightly faster than the old CULV with slightly better pricing, and it's generally less expensive than the Arrandale ULV laptops we've seen so far.
Facebook
Twitter
RSS
34 Comments
View All Comments
mczak - Wednesday, August 25, 2010 - link
per-clock performance of HD3200 and HD4200 is pretty much the same. The difference here (aside possibly from clock - not sure what clock the hd3200 used in the other notebooks were running at) is the platform. Most notably the HT speed, which was limited to HT-800 for Congo platform instead of HT1600 (for the K625). This basically halves the memory bandwidth available to the GPU and as you can guess this has a pretty disastrous effect on performance. Other differences are possible as well (e.g. different sideport memory), I believe some congo designs also ran their (ddr2) memory at a very low clock which could further lower scores though unless it was single channel it shouldn't make much of a difference since the HT-800 limits available gpu memory bandwidth still more. Well it could be a combination of slow memory and slow HT but whatever the case it's a bandwidth problem not architectural differences which make the HD3200 and HD4200 perform differently (unless you had an app which would use DX10.1 features).Hyperion1400 - Tuesday, August 24, 2010 - link
"Of course, battery capacities are a bit tricky—I know I have some 2500mAh 1.2V AA Energizer rechargeable batteries that suck compared to some equivalently rated 2500mAh Eneloop rechargeables"IxV gets you rated output(max power output assuming 0 ohm resistance). The only way to know the capacity of a battery is to rely on the manufacturer's word or to test the battery yourself.
Computer Bottleneck - Tuesday, August 24, 2010 - link
I think it is really interesting that older 45nm Intel's like SU7300 score better in battery tests than 32nm Arrandale i7-640um (found in Alienware M11 R2).Does this have anything to do with differences in peripherals (hard-drive, LCD, etc)? I noticed the Alienware has a 7200 rpm drive whereas the Acer Timeline 1810 has a 520 rpm drive. Or is the battery life difference related more to other factors? (I was expecting 32nm to definitely pull away from 45nm all things being equal).
JarredWalton - Tuesday, August 24, 2010 - link
I don't think most of the components make much of a difference. M11x isn't the most power-friendly implementation of Arrandale ULV, though, so we'll see how the ASUS UL80Jt stacks up next week. In general, though, the IGP in Arrandale is far more potent than the old GMA 4500MHD, and perhaps that's part of the difference.Intel specs the Arrandale ULV chips at 18W, which includes graphics. The CULV stuff was 10W, but I don't have a clear number on the chipset+IGP. It would appear that the chipset tops out at around 3W-5W for the IGP based on my testing here:
http://www.anandtech.com/show/2932/intel-core-2-cu...
Idle power draw on CULV may also be better, for whatever reason. Maybe 32nm has higher leakage, or it's just the number of transistors. I keep thinking the next 32nm parts from Intel will probably show much better power numbers, as this is really their first 32nm part. We'll see later this year with Sandy Bridge I guess.
strikeback03 - Tuesday, August 24, 2010 - link
All the testing with any Arrandale (ULV or not) has shown battery life not really improved from the later C2D processors, correct? Maybe there will be an improved stepping at some point.Computer Bottleneck - Wednesday, August 25, 2010 - link
Good point about Arrandale ULV possibly having higher leakage at idle. Maybe this is because i7-640um needs to be built on higher power wafers in order to hit the 2+ Ghz turbo speeds?Kishkumen - Tuesday, August 24, 2010 - link
As usual, the display just kills it all for me. A fine review, I just wish these laptop manufactures would throw us a freakin bone when it comes to some better displays. They're not doing themselves any favors with my pocket book anyway.Taft12 - Tuesday, August 24, 2010 - link
If you're referring to 1366x768, I fully agree!!Dell doesn't allow you to customize the screen nearly as much as they used to. Come on already, I am not paying over $600 for a laptop with that screen resolution!
The Crying Man - Tuesday, August 24, 2010 - link
Nice review. Is there any in the works with the N930 paired with an HD 5650? I have the dv6, but I'm just curious where it falls compared to all the other review laptops.JarredWalton - Tuesday, August 24, 2010 - link
I've got a P920 with 5650 review coming, but I do wish the CPU were faster. It's the Toshiba A665D... except the A665D is now discontinued and so my replacement system is the A660D. It's the same notebook but with a 7200RPM HDD AFAICT. Should have that up in the next week or so.