show me a mini-itx AMD Turion 64 motherboard and i'll buy it. otherwise i'm stuck with either an underpowered VIA or an overpriced Pentium-M

i'll stick with Pentium-M

This must be... a non sequitor...

goes to dictionary.com to look up sequitor.........


not their... hmm. did you mean "sequitur"?? "Definition: a logical conclusion from the premises; a logical consequence "

so, my reply was a "non logical consequence"? hmm...

it looks logical to me. no mini-itx motherboards are currently in production for a turion processor.

Why should a socket change be expected?

Great, great review of the Turion on the desktop. It's less a review and more a 'magnum opus' summarising anything you could ever want to know about the Turion and a pretty good how-to guide too. Silent computers have come a long way, baby. I think for the average person who doesn't have some extreme gaming setup we can safely say that they can transition their computer to the sub-20dB level with the right components (ie mobile processor, seasonic PSU etc all the SPCR standard gear) without having to sacrifice anything in terms of performance, and at a price that is affordable for just about everybody. Silent computing for the masses indeed.

It's also interesting that Intel and AMD were happy to dish out samples for this review; maybe they are recognising that the silence/power-efficiency niche is growing more important every day?

My Turion is famous.

Very interesting article, thanks for taking the time to do proper testing. I think a lot of unprofessional reviews of Turion64 when they first came out really gave people a bad impression of this chip. Having owned one myself, this was super-frustrating, as I didn't think this chip was any less efficient than Pentium-M. Looks like your measurements back this assumption up. The 25w Turion64s appear to be very competitive with a Pentium-M of comparable speed.

Just that the comment is kind of off topic, as mini-ITX is so specific, while as the artilce is about just running a Turion 64 on the desktop, something that hasn't been written about specifically before in the context of an article.

Yes, frostedflakes, that was your MT-34.

One interesting point we may not have made strongly enough... is that there appears to be a minimum power draw for the motherboard + memory + onboard VGA: 36W AC was the lowest we got, but the highest in the group was just 40W (w/CnQ or SpeedStep). In any case, it's clear that A64, T64 and P-M all have such low power draw at idle that motherboard voltage regulation efficiency is far more important for total power consumption (at idle) than the CPU.

Dual-core Turions will come in a new socket, Socket S1. It'll have DDR2 support like the AM2.

--

Great, great article. This is what makes SPCR stand out from the rest

Great article that offers some additional information on a desktop Turion since I've been trying to researching this topic since last fall/summer. Alas, I've decided to forgo upgrading my HTPC with a Turion though. I still have to upgrade my primary rig.

With the never ending advance in CPU development, I look forward to doing some research to upgrade my HTPC around the Merom, Socket AM2 Athlon, or Socket S1 Turion when the time comes.

Its just somthing that should be noted i think. Me personally, i will never go back to BIG and LOUD computers. IMO, large cumbersom computers are to visual disturbances as loud computers are to audible disturbances.

Theirs just not enough attention i think to reducing size as well as noise around here. Mabey if it were brought up more the issue would get resolved and we could see some mini-itx turion boards.

Great article. I just want to point out an error on the 1st page:

Athlon 64 Equivalent & TDP - The equivalent processor is 'Clawhammer' for the ML-44, ML-40, and MT-37 which feature 1MB of L2 cache.

Also, is there any chance you can undervolt a desktop A64 and compare against the mobile chips? I just want to know if undervolting a desktop chip to the same voltage as the mobile part will give you the same power-draw figures.

Mike, the 2.2GHz Turion has no speed equivalent Pentium-M yet you are saying that the 2.13GHz Pentium-M is speed equivalent when you say that the 2.26GHz Pentium-M has no speed equivalent Turion. Looking at the numbers, there is a 0.07GHz difference between the 2.13GHz Pentium-M and the 2.2GHz Turion, and a 0.06GHz difference between the 2.26GHz Pentium-M and the 2.2GHz Turion.

You might want to correct that to show that the 2.2Ghz Turion and 2.13GHz Pentium-M processors are not speed equivalent.

Also, on page 5, you say:



From what I understand, the northbridge is not integrated with the Turion die, but is/was integrated in a single die with the southbridge. You might also want to correct that.

I'm all for going small form factor, but there are several trends which make it an undesirable/unachievable option for many silencers:

1) The number of micro-ATX boards that undervolt I can count on the fingers of one hand. (OK, I exaggerate, but you get the gist) Undervolting is one of the main tools in the silencers toolbox; without this it's almost impossible to get true inaudibility.

2) Mini-itx options are slow, expensive, lack upgradeability and require expensive RAM and adapters.

3) Almost everything that is smaller than standard is also more expensive: slimline optical drives, 2.5" hard drives, PW200M + power brick, etc. It is much, much cheaper to buy standard size PSU's, optical drives and hard drives.

So going smaller is a desirable goal, but the performance and cost penalties are prohibitive.

I am curious, what are the chances that the AMD motherboards do a more energy efficient DC to DC conversion than the Intel motherboards do? I know that Intel uses an 80% efficient DC to DC conversion on their desktop boards but uses a 90% efficient DC to DC conversion on their server boards so I figure that it could be possible that AMD requires that their partners use 90% efficient DC to DC conversion circuits on motherboards for their processors.

If that is true, it should be possible for anyone comfortable with modifying their motherboards to replace the DC to DC conversion circuitry with super efficient circuitry on Intel motherboards and see a considerably difference in AC power consumption.

Typo patrol: page three says "Turiuon".

Aris: if you want small, I think that laptops are the most reasonable choice. I paid CAD 365 (~USD 320) for a used IBM X21 ultraportable that is likely smaller in volume than the average mini-ITX case. It came with a PIII-M 700 MHz that can kick the crap out of every VIA processor out of the market right now and 256 MB of RAM which is enough to run Windows 2000 quite comfortably. The CPU fan is all but silent - the only audible part of the laptop is the old hard drive, which is quite easily replaced. If the 12" screen is unsatisfactory for you, you can always hook up an external monitor. Laptop keyboards are also quieter than desktop keyboards . Oh, and even the most dead battery will be able to sustain operation long enough for you to save any critical documents should the power go out .

I doubt very much this is controlled or stipulated in any way at low power load. Perhaps some kind of min efficiency is a requirement for server boards at full load but I can't see any of the motherboard makers accepting such standards for desktop boards. (Except maybe with those meant to run >100W CPUs?) Impossible to enforce such a standard anyway, and when you come donw to it, what for? Manufacturers are concerned most of all on reliability and cost. At high power conditions, low efficiency can hurt component longevity, but there's no such risk at low power.

The efficiency of just about any PSU at this low load is probably quite poor as well (my guess is ~50%; it's running at maybe 20-25W total) and this contributes to the AC residual power draw.

<40W AC for CPU, MB, integrated VGA, HDD & CPU at idle is very very low. Without spending big $$$ and effort, I doubt this could be much improved. What for anyway? Noise would not benefit, nor cooling, not really.

It makes more sense to try and control video card power consumption in such systems. Let's face it: A single high end vidcard in these systems would probably double the power consumption at every level.

Great article giving Turion more exposure; even with more people becoming aware of AMD's desktop chips' superiority to Intel's in performance/watt and overall power consumption, I think few people have even heard of the Turion and even for many enthusaists the details were not clear.

However the burning question I still have is how close can you come to Turion's power performance with an undervolted AMD desktop chip? In a laptop, a few watts less will translate into a visible few minutes more running time, but for a desktop even a silent rig, do those few extra watts make any practical difference (will it run any cooler)?
In the review, the 35W ML-40 actually consumed less power than the equivalent clocked 25W MT-40 (2.2W vs 2.6W) when both were set to the same voltage at idle. Granted the 0.4W could be within the margin of measuring error, but this really brings up the question if the ML-40 were set to the same vcore as the MT-40 under load, would we see the same power consumption? MikeC any chance to test this?

It's mentioned the silicon is more efficient for the Turions, but with a 90nm desktop chip at similar voltage would we only be talking a insignificant few watts difference under load? Granted there is no guarantee that a CPU will run undervolted, but I believe many people have good experience with the AMD chips (many here managed to run various 90nm desktop chips at stock speeds under 1.25v) and with software like CrystalCPUID doing so becomes quite effortless.

Ideally if someone could do a temp/watt comparison between Turion and a s754 Venice core, or even the much more widely available s754 Semprons. The older 130nm Newcastle used in the article just runs so much hotter and consumes more power in my experience compared with the newer 90nm chips. Again this is not to challenge the Turion as the efficiency champ, but to answer the question whether an undervolted desktop 64 would be the better choice for a silent rig?

A64 (undervolted/clocked) vs T64 is a reasonable question, but it was not within the scope of the article, which already felt too broad. Mostly, we wanted to compare the T64 and the P-M for desktop use. As already mentioned, idle power issues are moot because of other factors that impinge on minimum AC power draw for a system. The more relevant question is power at load.

Yes, we only had a Newcastle A64-3200+ to compare, and this is definitely not as power efficient as the E-stepping core used in the Turions. We do have plans for a Part 2 comparing recent A64 against T64 on the desktop. We need to obtain more samples.

1. you wouldnt need to undervolt if you built the computer on a low power cpu in the first place like a pentium-m, via, or turion processor.

2. they are expensive, and lack upgradeability, but they are not slow, and they use standard ram. lack of upgradeability however is usually compensated by the fact that they are highly integrated boards that usually come with everything you need onboard. you can get boards with pci-e 16x slots, mini-pci slots etc etc for those one or two things you cant get integrated on a boad.

3. smaller is more expensive. quieter is also more expensive. ive seen discussion threads that say the average SPCR'er spends around 300-400 USD on just silencing their computers. thats about the amount it would take extra to shrink the size of the computer down to mini-itx as well.

performance isnt necisarily sacrificed if you pick your components carefully. I'm currently building a gamming rig on a mini-itx system with a pentium-M processor, 2gb dual channel 533 ddr2 memory and the upcomming 7600gt video card. total extra cost of the system over a normal atx system is around 400 bucks. but some of that cost also will reduce system noise, like the 2.5" hard drive, and low power consumption resulting in only a single 80mm undervolted fan. so the extra cost to shrink the system in size also incorperates the cost to make it silent. and since the average SPCR'er spends around 300-400 just making it silent anyhow, they could get a smaller system for the same cost as well if they did it right.



Notebooks as of yet do not have upgradable video cards. when they do, then notebooks will become a viable option for me. not all mini-itx boards utilize VIA cpu's. my upcomming mini-itx system will be a very powerfull gamming rig. read reply above this for specs. as for size, i'm very happy with an enclosure with exterier deminsions of 8"x8"x4".

Very good review. Yeah, comparing A64 with T64 sounds interesting. I think that if you undervolt an A64E you pretty much get the same power usage at the same Vcore and clock speed. A quick check tells me that the 3200+ in the article would use about 9 W more than the MT counterpart if using the same Vcore (if possible). A 90 nm desktop CPU would do better than that. It's still a theoretical point of view though, since I haven't tested it.

Personally I don't see any reason to choose a T64 instead of an A64. I've seen all the trouble (and mobos) people have gone through just to find a working mobo for their mobile A64 in various forums, and I don't think it's easier with a T64. Many times it wasn't just about finding a mobo that supports the latest desktop counterpart. It seemed like it was the differences in the microcode of the CPU that stopped them from working properly. Maybe things have changed though, I don't know.

For those who are looking for an S754 upgrade there are alternatives. The cost effective Sempron for instance, and also the regular A64 revision E "Venus".

MikeC,

A much needed article, hopefully this will let people argue about differences with some actual facts for once. Thanks for your time and effort.

A great article overall.

My only concern is that the power comparisons and price comparisons didn't take into account equivalent performance. A recent review by Tech Report showed that the top of line ML-44 is faster than the 760 although the 760 does win some benchmarks. This is pretty good considering that the ML-44 has a 400MHz advantage over the 760. The faster 770 may be competitive with the ML-44 which changes the price comparison.

http://www.techreport.com/reviews/2006q ... dex.x?pg=1

However, I know that this was mainly done from a power consumption perspective so performance would be less important. I can't wait for you to get your hands on a Core Duo. It would make an interesting comparison to the current Dothans and Turion64.

The 2.4GHz ML44 bests the 2.13GHz 760, so we can surmise the 2.2GHz MT40 would be roughly on par with the 760. Look at their respective pricing. Our premise that the two processors have about the same performance at the same clock speed is correct; hence, the price comparisons are also valid -- Turion64 can give you the about the same performance & power efficiency for a lot less money.

Only problem with comparing CoreDuo is that there don't appear to be any desktop boards or barebones systems for them yet.

wow, amazingly good review from the big Mikey-Cee

That 200 ati chipset is a PERFECT choice for a new system. wow. really cool deal.

The 760 is a 2GHz chip so what we have is a 2.4GHz Turion64 beating a chip 400MHz slower and $60 cheaper. That is why I surmised that the 770 which is actually a 2.13GHz chip, will probably compete well with the ML-44. Their performance is not equal clock for clock, with the Dothan appearing slightly faster.

We are quibbling over very fine differences, and until there is a truly fair contest between these processors at the same clocks speed, it won't be decisively settled. Still, a MT/ML40 would probably be very close to the 770. In any case, the performances differences even between ML44 and ML40 (or 770 and 745) are not perceptably significant in actual usage, based on my hands-on use with systems using these various processors. Just as with the small differences in power draw at idle, the performance differences here are not at all compelling --- which is all the more reason to choose not on peformance parameters, but on price, availablity and ease of use. The last criteria is a bit of a tossup, imo, availability probably favors Pentium M, and Turion 64 has the price edge.

This occurred to me as well. My guess is that the only difference between the ML and the MT is that the MTs are binned for lower voltages. Obviously, when we do part II, we'll have to do some more comparisons, but it was already difficult enough to integrate the large amount of data in this article. I think this article answered some questions and raised a whole bunch more.

Regarding A64 vs T64, keep in mind that the newest A64 chips seem to be artificially limited to a minimum voltage of 1.1V. The Turions don't have this limitation. Also, I believe most A64 chips run at the same stock voltage as the ML T64s: 1.35V. However, we will have to get our hands on some new processors to test this properly. The only E stepping A64 chips available for S754 have 512KB L2 cache, and all our Turion samples have 1 MB. With the parts we have on hand right now, we have no way of doing a fair comparison between an A64 and a T64.

We might just have to fudge it and find an appropriate S939 board.



So "NewCastle" is not the correct core name? I'm sorry, I think I got the name wrong. Rest assured that the comparison did involve the A64 chips with 1MB cache. I was very meticulous about that.

Keep in mind that the clock speed on the 770 is only 6% faster than the 760. IMO, neither of the two "top end" P-Ms are worth considering ... you don't really get anything for your money.

Also, have caution about getting all of your information from one source. The review from Laptop Logic (linked to in the article) also contains detailed performance testing. IIRC, it also does a better job of matching up the processors that they test, and the Turion comes out looking even better. As MikeC has already said, I think the hairs you're splitting are too fine; performance is close enough that 200 MHz or so isn't going to make a huge difference.

Isn't the MSI RS482M2-IL (socket939) nearly identical to the MSI RS482M-IL (socket 754) you used in this test?

Edit: Never mind. I just noticed that that one is replaced by the RS482M4-IL(D) almost everywhere.