Intel Centrino Duo vs. AMD Turion? 64 X2 for Linux

Currently HP is offering many of its laptops with AMD Turion™ 64 X2. I'm wondering whether this has any advantages or drawbacks if you use it in Linux in comparison to a laptop with a Centrino Duo processor? Did anyone have experience installing Linux on a machine with AMD Turion™ 64 X2?

 

That is going to be hard to compare. Right now the Core Duo offers better performance.

 

Im afraid i have no experience with linux, however the Intel processor is faster per clock speed, and is more energy efficient than the athlon, however the ahtlon is 64bit, so it depends on whether u run 64bit linux

 

I would think hard about buying at 32bit laptop right now, even more so since you are a linux user and can get the 64bit performance boost now.

 

The AMD Turion 64 X2 is faster than the Intel Duo Core.
Though the clock speed is slower, AMD has fast bus and hypertransportation, so don't need large caches. Don't be distracted by the large cache of the Intel, it a conpensation for the lack of memory controller (AMD has a higher benchmark on RAM using efficiency).
Also to mention that the Duo Core comsumes too much power, and too hot, too.
The 64-bit linux is now available, you can have it anytime, and of course it's for AMD 64 (no Intel 64 is available now except for the Pentium Extreme Edition, and those for servers).

 

Yea ok, if you can even find X2. So far there is MSI, but they got owned big time. And they are selling for over $1500 for the slowest X2, more than price of typical Core Duo 1.83 and even 2.0 ghz. Oh yea HP are making them too, but did you check the delivery date yet, how about another few weeks?

People keep saying 64 bit. 99% of application out there is still 32 bit. And even if they had 64 bit Windows Explorer or OFFICE, I can't tell a **** difference - can you???? I use it to type reports and essays, it does not make ONE spit of difference.

And there is noway that X2 uses less power than Core Duo. They are based on 90 nm chips, basic physics. That's why they come with bigger cells for the same amount of battery life.

Pound for pound, you get more performance out of Core Duo than X2. Manufacturers have had over 6 months of lead time to streamline and perfect the Core Duos.

Until I see a X2 based SONY SZ or W7J, giving MORE battery life than Core Duos, it's ALL PIPEDREAMS.

By the time 64 bit becomes MAJORITY of application, I'd be getting Quad Cores and perhaps Turion X4's if AMD mananges to leap ahead in the laptop race.

AMD are great desktop chips [Well except for the Pentium 805 dual core that just came out so cheap but can be overclocked to be Athlon FX killers]. Unless AMD do something fast, they will become fossils in a few years.

YES I waited until X2 came out before deciding to buy Core Duo. I was NOT impressed by the MSI at all.

 

Wow is that real? core duo beat turion X2
I remember when I built my first desktop , I pick AMD Athlon because of its
low power consumption and cool&quite technology

but now for my first laptop I ordered a core duo right before HP release their
turion X2 series laptop
I haven't seen the benchmark between comparing this two yet so where did
you find out about this "crdiner"

 

im afraid ur wrong here. Many test have proven it isnt faster, sure it has hyper transport etc... but the core duo has twice as much L2 cache, its definitely faster than the X2

 

This article proves my point to a degree

http://crave.cnet.co.uk/laptops/0,39029450,49273534,00.htm

 

Here's one very important thing to consider: the Windows multimedia codecs are not available for 64-bit, so you have to run 32-bit anyways if you want to use those. Any performance advantage of 64-bit (and for most applications, it's not much) goes out the window if you're forced to run it at 32-bit. So you can either run 64-bit or you can stream videos from the net, but not both. Until Windows goes 64-bit, there's much incentive for desktop users to stay 32-bit. Add to that Intel chips are still outperforming AMD chips on the notebook market and the choice should be fairly obvious...

 

Hey folks - not to disagree with what you're saying in terms of Windows, but the OP was specifically asking about Linux. So, what might the relative advantages/disadvantages be in choosing Turion 64 X2 for Linux?

I suspect the answer will depend on the extent to which the OP uses 64 bit applications in Linux. For example, some scientific computing might benefit. Anyone know more?

 

Unfortunately, it looks like they used significantly different hardware configurations in those tests. Here is a review (in French, unfortunately) which used more similar configurations: http://www.matbe.com/articles/lire/299/amd-turion-64-x2-socket-s1-et-ddr2/page3.php

The French review also suggests that the Core Duo is faster, especially in games and SuperPi. However, the very low gaming performance seems to be related to a CPU bug which AMD has since released a patch for: http://www.dailytech.com/article.aspx?newsid=3152. And of course, SuperPi is heavily L2 cache dependent (see this article for example: http://www.pcstats.com/articleview.cfm?articleid=1815&page=3), and the Core Duo has four times the cache that the X2 has.

In the rest of the benchmarks, the Core Duo wins more benchmarks than the X2, but they each win some, and often the results are very close. Personally, I think it's hard to say which is faster, especially if you're going to use a 64-bit OS. Athlons running in 64-bit mode get a speed boost from the extra registers which they can use in this mode (see here for a few benchmarks: http://www.pcstats.com/articleview.cfm?articleid=1665&page=6). Having said that, not every piece of software will get this benefit. Commercial games will probably still be 32-bit, for example. Also, programmers may use SSE2 for their 64-bit math, eliminating several of the Athlon's advantages.

So in summary, it's a big mess. I would recommend that you choose your CPU for Linux based on other concerns, such as power consumption, price, and support (e.g. does your favorite distro have a dual core athlon kernel yet -- I know mine doesn't, yet).

 

Not that Core 2 Duos are in notebooks yet, but you may want to look at this:

http://reviews.cnet.com/Intel_Core_2_Duo_E6700/4505-3086_7-31973836-2.html?tag=nav

Also, 64-bit in anything is not really a necessity (yet)...most people do not need 64 bit precision. That said, I'd agree with Mr. Foolish and say, don't pick based on the processor brand, pick based on overall feature set and notebook brand.

I have heard some complaints about the Windows codecs used in Linux when running 64-bit, but the forums don't seem to indicate that 64-bit really is the issue. AMD has had 64-bit out for some time now, and as far as I've heard (notice I'm running a T2400 in my sig) there aren't [m]any issues. Even ATI supports 64-bit operation in their closed-source drivers.

 

I second that. I'm running 64-bit Linux on an Athlon 64 laptop. Video/audio codecs aren't a problem. They probably are 32-bit, but they work. In addition, I too have ATI's 64-bit video drivers working with my Radeon 9600. However, Shockwave Flash doesn't work for me. I'm not sure that I miss that so much, but it is worth noting.

 

Guys, I figured that we maybe setup unfair test for AMD with Intel almost always for SuperPi. Because of large cache in Core Duo, surely 2M digit of SuperPi will lean to Intel.

How about we retest it at march larger digits. Let's say 16M, how is that? And since we need test 2 cores, even in Linux, lets start two terminal to run these two big tasks, and comparing the result.

 

I'm not familiar with the algorithm used in SuperPi, but in order for cache (L2) to have an impact, the next digit in the pi calculation would have to rely on the previously calculated digits (a lot of them in order for 1 or 2 MB to fill and start dropping digits out of cache). I assume SuperPi is actually not a difficult algorithm so the algorithm itself is typically L1 cache.

 

If I'm not mistaken, all pi-calculating algorithms involve sums of infinite series. All of the digits you have requested are calculated right away, but inaccurately, then are approximated more and more carefully until there are "correct" by some standard (e.g. X digits are guaranteed to be correct after Y iterations or something). So I don't think the algorithm works the way you think it does.

Of course, on a real computer SuperPi wouldn't have free run of the L2 cache either. The OS itself has lots of libraries loaded and system level processes running. So SuperPi may just be allowed to use a fraction of the L2 cache, and presumably it gets more when there is more L2.

Really though, the main problem with SuperPi is just that it's a single benchmark, both in terms of just being one benchmark and in terms of benchmarking one simple algorithm. Does anyone else remember when RC5 key cracking was a popular benchmark about 6-7 years ago, and the AMD K5 turned out to be the fastest thing out there per Mhz, like 20-25% faster than the K6 or P6 (Pentium Pro/II/III) architectures because it executed the bitwise rotate function fastest? There are just too many subtle differences between CPUs for simple benchmarks like RC5 cracking or SuperPi to be meaningful.

 

Hi, I dig out two benchmarkers that suppose to test multiple core performance

http://www.cinebench.com/

http://www.majorgeeks.com/CliBench_mk_III_smp_d121.html

And also Pifast result might really like higher memory bandwidth which AMD fans should love it.

 

I'm actually surprised that we don't see raw processor performance benchmarks under DOS. What better way to benchmark cpu operations than under a single-threaded, no overhead, light-weight OS? Of course if you start wanting to benchmark things like main memory or graphics, it would be much harder.

Maybe a more realistic platform for benchmarking would be a soft real-time Linux variant running from a Live CD.

 

What do you mean by single-threaded benchmark. We are mostly talking about dual cores. Surely multiple threads are one of the important factor. Pure single-thread will not show much improve for 2 cores over single core.

 

you just took every single word out of my mouth, and I couldn't said it better myself. many people want to upgrade just for the sake of upgrading. sure who doesn't want the latest and greatest bleeding edge. most keep vying for 64bit but they don't even know what it means, they just think 64 > 32 so it MUST be better. imo, it is so stupid to upgrade when you have no need for it or upgrade to 64bit thinking it'll give you a 2000% boost in performance. and these are the very same people that complain to no end when the new hardware/software doesnt meet their expectations. its not what 64bit is for. its not like comparing a v8 to a 4 banger but most people don't see that.

 

What I mean is that both Windows and Linux are pre-emptive multi-threaded OSes. That means that other threads/processes can take over the current core/processor and make the benchmark take longer because the benchmark didn't have the core/processor for the full time. DOS doesn't have that issue (it will service interrupts, and interrupts are supposed to only do very light work) so if you are interested in pure number crunching, it just seems a better choice. I realize the multi-threading/smp benchmarking is impossible there, but as I said, pure number crunching. You probably are right though, lots of everyday tasks utilitize multi-threading, so perhaps single threaded benchmarking wouldn't scale well into a true reflection of system performance.

 

See, the problem is, this is the Linux forum. We're the smart ones.

 

But really, only Linux can work on both 64 bit and dual cores including day to day applications.

The matter is we don't have much convincing benchmark app available in Linux that can tell us which one is better to another. So many benchmark app in windows vs. so little in Linux. That's su*ks.

 

Every benchmark would be unfair, so it's good there isn't "the one".

You could always build the same kernel on two different laptops and compare that - I think it would be a good overall score for system speed (CPU, RAM, HDD, of course not graphics).
But that's something totally different as e.g. video encoding...

Regards
Aaron

 

What, you can't just cat /proc/cpuinfo and go with the one with higher bogomips????

(yes, I'm completely kidding)

 

That is completely 100% provably wrong.
It's easily possible for a chip based on a smaller process to consume *more* power than one running an older, bigger process.
For a simple example, the 90nm P4 Prescott ate much more power than the 130nm Northwood it replaced. And Intel's 65nm Pentium 4's certainly use more power than any 90nm chip made by AMD.
Power consumption depends on the architecture more than anything else. Both architectures in this case are pretty good at low-power operation. From what I've seen, Turion manages to keep up surprisingly well, considering it's basically a desktop chip, but a Core Duo still beats it (marginally). From what I've seen though, the margin is surprisingly small, and it's not something that'd worry me.
However, just wanted to make the point that you can't use the argument that "It's 90nm, it *must* use more power than a 65nm chip"

 

Ok, here is my AMD desktop 3500+ PiFast result. PiFast seems a lot faster than SuperPi. And you can really test memory I/O since it let you run super big digit in a short time. I have dual channel DDR 400 1GB memory(not fast though).
I've test 40M, 10M, 4M, 2M, and 1M digit. I use compressed mode output. I didn't test if there is any different between no-compress and compress output.
Following are raw output

Computation of 40000000(40M) digits of Pi Method used : Chudnovsky
Size of FFT : 4096 K Physical memory used : ~ 243229 K
============================================================
Total computation time : 234.16 seconds (~ 0.07 hours)
============================================================

Computation of 10000000(10M) digits of Pi Method used : Chudnovsky
Size of FFT : 1024 K Physical memory used : ~ 61355 K
============================================================
Total computation time : 46.11 seconds (~ 0.01 hours)
============================================================

Computation of 4000000(4M) digits of Pi Method used : Chudnovsky
Size of FFT : 512 K Physical memory used : ~ 29283 K
============================================================
Total computation time : 14.94 seconds (~ 0.00 hours)
============================================================

Computation of 2000000(2M) digits of Pi Method used : Chudnovsky
Size of FFT : 256 K Physical memory used : ~ 14874 K
============================================================
Total computation time : 6.47 seconds (~ 0.00 hours)
============================================================

Computation of 1000000 digits of Pi Method used : Chudnovsky
Size of FFT : 128 K Physical memory used : ~ 7769 K
============================================================
Total computation time : 2.75 seconds (~ 0.00 hours)
============================================================

 

gethin, the article proves you are completely wrong...
have look at the system config and graphic cards on the two computers....and also read the comments on the article.

 

Dude, the thread is over a year old. Why would you chime in now, especially on such a moot comparison (the Core Duo has been obsolete for almost a year now)?