About the Weird Performance of ThinkPad X300

Lenovo's recently-released 13" ultraportable ThinkPad X300 has drawn considerable attention and also has raised different discussions, especially from those die-hard ThinkPad fans. One important but bothering issue is about its not-so-promising performance: the built-in Core 2 Duo L7100 is even outperformed by its ultra-low-voltage brother U7600 with the same frequency. However, by collecting pieces of information from the Internet, the author found that this apparent dispointment is probably not a big deal, as the real performance of L7100 is more likely to have been under-estimated due to the tester's mistaken or careless setting of the benchmark software.

1. The Fact

So far there are only two original sources of ThinkPad X300 performance testing available from the Internet. One comes from notebookreview ([url]http://www.notebookreview.com/default.asp?newsID=4270[/url]). The other is by pconline ([url]http://www.pconline.com.cn/notebook/testing/occident/thinkpad/0803/1236421.html[/url]). The rest reports are just forwarded versions or reposts of these two. In both sources, the testers are using a multi-threading benchemark software named wPrime, to test and also to compared the performance of L7100 that is equipped on X300. Dispointingly, they are reporting a low outcome of 118s and 126s in calculating the 32M-bit pi value. These scores are dispointing because they imply that the new and fascinating X300 is much slower than most of its competitors (in fact in that comparison it only outperfoms Asus EEE 701 which is equipped with a low-end Celeron M ULV CPU running at 900MHz). Nevertheless, even a ThinkPad X31 with Pentium M 1.4GHz can score 130s. So what does this mean? A brand new X300 performs almost equally as a five-year-old X31? You must be kidding...but this is what shows up here. This weird score not only bothers many ThinkPad fans, but also triggers further digging of the reason behind the apparent puzzle.

Merom L7100 wPrime 32M test score by pconline:
[img]http://forum.51nb.com/attachment.php?aid=524365&noupdate=yes[/img]

Banias 1.4GHz wPrime 32M test score on X31:
[img]http://forum.51nb.com/attachment.php?aid=523290&noupdate=yes[/img]

2. Speculations

There are quite a few posts from different laptop-related forums, questioning the above-mentioned wPrime scores. In this forum, there is a thread initialized recently ([url]http://forum.51nb.com/viewthread.php?tid=649810[/url]), collecting volunteer reports of wPrime test results from different CPUs. As more results accumulated, the problem is becoming easier to identify.

The Core 2 Duo L7100 used in X300 is a newly-released SFF (Small Form Factor) version that is targeted at a lower power consumption (12W) and a smaller size. Besides that, it has no difference from a regular low-voltage Merom CPU in its specifications: 65nm, 1.2GHz, 4MB L2 Cache, and 800MHz FSB. All these specs can easily beat those of a Pentium M on X31: 130nm, 1.4GHz, 1MB L2 Cache, and 400MHz FSB. It is also notable that intel's advanced manufacturing process has been amazingly improving the unit performance per clock. For example, a 90nm Dothan outperforms 20% than a same-frequency 130nm Banias; a 65nm Yohan outperforms even more than 20% when compared with a 90nm Dothan of the same frequency. There is hence no reason that a 65nm Merom has equal score of a 130nm Banias, even the Banias has a 17% faster frequency. Instead, from the above analysis, a single-core Merom 1.2GHz should score at 105s, taking the 1.4GHz Banias's 130s into consideration. This, however, only concerns the manufacturing process, let alone the quad-sized L2 Cache and double-pumped FSB. On the other hand, L7100 is a dual-core CPU, so its reasonable score is expected to be as half of its single-core counterpart, say around 50s - 60s.

See how weird this is?

3. Issues and Analysis

The benchemark software wPrime itself is multi-thread enabled and is able to recognize different CPUs (as it claims). However, there are several issues affecting the testing scores.

a. version. As pointed out by many posts (e.g. [url]http://forum.51nb.com/viewthread.php?tid=649250&page=2#pid9167654[/url]), wPrime has different scores if tested with different versions. The most up-to-date version 1.6 has much slower score than the widely-used versions 1.54 or 1.55, and the difference could be as large as 60% or even 90%.

b. setting. The software itself requires a proper installation before use, in particular, the so-called "missing libraries" issue is better resovled beforehand. This will make it better to recognize detailed CPU information and to run smoothly as expected.

c. multi-threading. There is an "Advanced Setting", which enables a user to manually enforce wPrime to instantiate a certain number of threads, instead of the pre-determined number of threads according to its own knowledge-base of CPUs.

d. OS. It is surprizingly confimed by many posts that wPrime performs better under Windows Vista than Windows XP, though the difference is in a scale of a few seconds, which is not that much noticable.

Both sources use a wPrime version 1.54, so the first issue is no longer an issue. Do they properly install the missing libraries? Well, we hope so, that's what they are supposed to do...but who knows. Do they notice the multi-thread setting and make sure they are truly running two threads? Probably not, as we can tell from their testing scores as well as the above analysis. Do they run on Vista? It doesn't matter that much, anyway.

Shall we suspect the wrong setting of the number of threads? In other words, the testers only ran one thread in fact and took the corresponding score for their reviews? It is more likely. Note that a single-threaded wPrime only tests the performance of a single core. Moreover, given the fact that the SFF L7100 is kind of special, it is possible that wPrime doesn't have the accurate information of that CPU, and recognizes only one core during its initialization and then takes that by default for the subsequent testing.

4. Suggestion

Should anybody reading this thread obtain an X300, please conduct a wPrime testing with an enforced dual-thread setting and kindly post the score here. That result will be most likely to jusify what has been speculated above, and will eventually peace up the minds of ThinkPad fans puzzled by this performance issue.

This  is the first digest of this area^u^

[quote]Posted by [i]juno[/i] on 2008-3-15 12:05 [url=http://www.thinkpad.cn/forum/redirect.php?goto=findpost&pid=9216046&ptid=650110][img]http://www.thinkpad.cn/forum/images/common/back.gif[/img][/url]
This  is the first digest of this area^u^ [/quote]

maybe the second is coming soon.

The Justification

It has been just reported by HOPE ([url]http://forum.51nb.com/viewthread.php?tid=656471&page=1#pid9265047[/url]) that ThinkPad X300 scored at 76s in running wPrime version 1.55. This justfies the above suspection about the real performance of Merom L7100. Though a score like 76s is not very promising as expected, it at least demonstrates the power of dual core (instead of making people worry about its being defective or mulfactioning). On the other hand, it also means that the computation of wPrime is not that simply linear as speculated previously. The following is the original screenshot posted on that thread:

[img]http://forum.51nb.com/attachment.php?aid=531596&noupdate=yes[/img]

Plus, there is also information about its single core computing capability: it finished calculation of Super Pi 1M in 40s. This seems much normal for an LV Merom (giving the fact that a ULV Merom U7600 running also at 1.2GHz has a score of 49s), and is almost comparable with a 2.0GHz Dothan 760. Since the Super Pi computation depends more on the size of L2 cache, it is not very suprising that L7100 with 4MB L2 can run this fast.