To Devonavar and MikeC

I understand that we're getting picky so I hope I didn't sound ungrateful for your work. With mobile chips it's always difficult to compare them equally, especially in a labtop setting where every component choice makes a difference. I've always found though, that on a desktop setting, Dothan performs quite well even against full Athlon64s with dual channel memory, which was why I had concerns.

http://www.gamepc.com/labs/view_content ... 780&page=1

I also found it curious why you used the AOpen i915Ga-HFS instead of the i915GMm-HFS you reviewed before, but I suppose it might not be fair to use a power optimized motherboard since I'm not sure if an AMD equivalent is available. It would be interesting to do a follow-up with the i915GM since you noted that it was the overall system power consumption of the Pentium M platform that was high.

On another note, I wanted to suggest a comparison. The new Core Duos now have the LV versions with the same clock speed and FSBs as the full power versions. It would be interesting to compare the 1.66GHz Core Duo with the 1.66GHz LV version to see if there is any performance difference and how much power you really save. I was planning on getting a 1.83GHz LV Core Duo when it's released. I guess this will have to wait for availabilty to catch up though.

ltcommander_data --

BTW, there's a bit of a gotcha w/ the CoreDuo processors.

Yes, they are part of the Centrino mobile platform.
Yes, they have 478-pins, like "479" Pentium M and socket 478 P4s.

...Here's the gotcha...

They are NOT pin-compatible with Pentium M or regular P4-478 sockets. The CoreDuos have a different electrical pinout config, and can only work with new socket "479" boards using the Mobile Intel 945 Express Chipset Family.

This info comes from a chat with a tech at AOpen (one of Intel's close partners, especially w/regard to special products using mobile components) and a close study of CoreDoc docs. I'm 99% sure that it's correct at this time.

Comparisons against earlier P-M processors cannot be done on the same mobo.

This means new classes of Mobile Intel 945 Express boards for the desktop must appear before CoreDuo can be used on the desktop.

I think you can take Intel's word for what the LV version does against the standard V version. For mobile apps & battery life LV is probably preferred. But on the desktop, unless you have a highly efficient VRM on the motherboard and a PSU that's highly efficient at low load, the advantage will only show up at full load.

I really didn't mean it like that. I'm sure you will do a comparison when it's available so I know it's no use in me pestering you about it.


I knew that the i915 was not forward compatible with Core Duo, but I thought that the i945 would at least be backward compatible with Dothan. Supposedly there's a supply shortage of Core Duo's with Apple getting first dibs so I guess Intel is delaying the launch of desktop i945GT motherboards to avoid misdirecting too many labtop chips to the desktop market. I guess that's what you get when you do a process transition and an architecture transistion at the same time.

Since I'll be getting a labtop and running quite a few things in the background, I'll taking your advice and just get the LV.

What's the deal with the AOpen i915Ga-HFS having the PCIe x1 placed too close to the back (about 5 mm) of the motherboard?
Or is it some other kind of slot? Last time I checked it was a PCIe.

Although most of us are unable to break below the 1.1v barrier, seems there are occassional reports of success like here with a Jetway A200GDMS/Semp3000+RevE6 and here with a Biostar TForce6100/Semp2800+. It will be interesting to see how low the Turions can be run stable.

As for fair comparison, I don't think most of us are looking for an exact watt-for-watt, cache-for-cache comparison (heh heh although it would be very interesting if the Turions actually weren't more efficient than the new A64s as we are led to believe). For the most part processing performance between the s754 Venice and equiv clocked Turion should be quite similar, based on various A64vsS64 benchmarks, a larger cahce does not offer that significant a gain, and everyone can decide for themselves if they'll notice/need that extra bit of processing power or not.

There also the S939 Opterons with 1MB cache that will undervolt below 1.1v, but of course there's the different socket, higher HT, and dual-channel memory. Per AMD's own PR ratings, s939 chip should perform on par with a s754 equivalent clocked 200mhz faster (although in actual benchmarks performance does vary depending on what you run). With many manufacturers offering very similar mobos in both s754 and s939 flavors, it would still make for an interesting (if not exact) comparison.

Yeah, for some reason it's very hard to find an A64 with the same specs. The slowest one with S754 runs at 2 GHz, while the fastest T64 with 512 kB L2 cache runs at 1.8 GHz.

On the other hand, you can always change the settings, like using x9 multiplier in the example above. If you're comparing with a S939 you can still run it with single channel RAM and lower the HTT multiplier to x4. The choice of not using the stock settings is not preferrable in a test, but in this case it might answer the question: Why T64 and not A64?

FWIW, I was able to undervolt my old MT-34 to 1.1V completely stable, 1.075V mostly stable. Oddly enough at 1.075V the CPU could handle Prime95 for hours and hours, but I would get random crashes every few days during idle. I think it was because my DFI board would undervolt the processor slightly at idle to ~1.05V.

Also keep in mind that I got my Turion64 almost right after AMD released them, so this was a very early piece of silicon. I'd expect the newer/more mature silicon has better undervolting capability.

Last page:

Don't know if it should be S1?

It should. Will correct later; the server is groaning under the heavy traffic load caused by a front page http://slashdot.org/ reference.

I backed into this fine article soon after I began searching for an SFF to replace an old desktop (no gaming, just simple 2D work at most). mATX is bigger than I want. miniITX looked great, but cost more than I am willing to afford. So I settled on Shuttle's SK21G and discovered the Turion on their support list.

This review doesn't really add anything to what is already common knowledge on the web.

1. The world already knew the Turions are cherry-picked undervolted desktop chips. The "PowerNow" support is simply a microcode change performed when the processors are speedbinned.

2. The world has known for some time that Athlon 64s on the 0.09 micron process are power misers. Tom's Hardware did the same test you did on the Winchester core Athlon 64 way back in November 2004, and found in the range of 1.8 GHz to 2.2 GHz, the cores used 29-32w at full-load, and ~4w at idle. In fact, Tom's tests are the reason I bought a Winchester core at the time.

As for the review, I must question your use of the Newcastle core. The Newcastle core is OLD, 0.13 micron technology, and that is the reason it runs at 1.5v. If you had used a modern desktop Socket 939 procesor, you would have found they have similar power usage to the Turion-ML.

I also took issue with your recommendation of the Turion, calling it COMPETITIVELY priced compared to desktop systems.



Your claims just don't stack up. You started by claiming that a Turion platform would be cheap because some Socket 754 boards can be had for under $50. But then you discovered that most older boards (ie, cheap, used boards) don't support Turions correctly, and that ATI chipset board you settled on costs more in the $70 range (same as some Socket 939 boards).

Secondly, did you people even bother to price Turions before you wrote that crap? They have a huge price premium over equivilant-perfoming desktop chips. For example, the Athlon 64 3200+ s939 costs $140 on Pricewatch. Equivilant performance Turions go for $230 (ML-40) and $270 (MT-40). Considering that the only difference between the ML and desktop chips is a tiny difference in idle wattage, and considering how few places sell Turion chips to the public, how can you justify reccommending Turions unless people are going to shell out for the MT series?.

I mean, really, if you want a CHEAP, low-power machine that can do anything except play newer games, its very simple: buy the cheapest Venice core Socket 939 Athlon 64, or a used Winchester core.. Undervolt that sucker to 1.1v, and see how high you can clock it. You're guaranteed 1.0 GHz as a minimum (thanks to the requirements of Cool 'n Quiet), and hell, even my old Winchester can do 1.2GHz at 1.1v.

At that speed range and voltage, your processor uses 10w or less at full load, and that can easily be passively-cooled.

*puts on flame suit*

I'm just curious how you "know" that Turions are simply undervolted desktop processors? For one thing, all Turion64 parts are SH-E5 revision, which is exclusive only to Turion64 and Athlon64 mobile (Newark core). This suggests that it is based on a different process than other K8, as all previous mobile processors have been the same revision as their desktop counterparts (for example, I owned an Oakville before my Turion64, which was DH-D0, same as the Winchester for Socket 939). Also AMD says that Turion64 uses lower-power transistors, considering the evidence above I see no reason to doubt them. How much of a difference Turion64's lower-power design makes is up for debate, though. I'd assume it's no more than a few watts.

As far as the power consumption measurements, keep in mind that Winchester has roughly 33% fewer transistors than 90nm K8 with 1MB L2, which means that inherently it is going to use less power and dissipate less heat. Look at Tom's power consumption measurements for the 3800+ and 4000+ at load. Both are Socket 939, 2.4GHz, and 130nm, but the 4000+ is a Clawhammer core with 1MB cache. It pulls ~5w more power than the 512KB Newcastle.

And in Devon and Mike's defense, I don't think they ever claimed that Turion64 was competitively priced with desktop K8. I think the main focus of the article was comparing power consumption and cost of Turion64 and Pentium-M on the desktop -- the Clawhammer was just thrown in there for kicks.

Wow, nice, you register for the sole purpose of trashing the review (in a rude and offensive manner to boot).

My Winchester definitely puts out more than 4W at idle at stock clock.

Also, Turions can be picked up on Ebay for much the same price as normal A64's.

Defaultluser: First off, if you decide to stick around, Welcome to SPCR!!!!!!!

I think you've missed the basic thrust of the Article, which was to compare the Turions vs. the Pentium M. Some desktop data was added, but, as was already pointed out, it was more "just for kicks" than for objective comparisons.

I think you raise some fair points. We were not aware of how close the Turions and the A64s might be when we first started testing. There is very little (if any) information out there that tests them directly against each other. The T64 vs. A64 question is a valid one, and will be the focus of an upcoming article ... if you're interested.

I would like to respond point by point:

1. The world would be wrong if they knew that. Turions are not undervolted desktop chips. I don't know where you got that, but it wasn't in the article. At the very least, Turions have added support for the C3 sleep state (not in the article, but a difference none the less).

2. Yes, the world has known that for some time. What's your point? The article is about Turions, not Athlons. Also, I would be interested to know how you came to the conclusoin that THG did that same test we did. I was unable to find any information at all about how the processors were tested, what tools were used, or what efforts, if any, were made to examine possible variances between the various test systems.

I might add that the THG article is not trying to do what we are. THG mentions power consumption in passing, does very little in the way of analysis, and doesn't really try to put the results in perspective. If all you want from a review are hard numbers (obtained in an unspecified manner) THG is fine. They have far more resources at their disposal than we do, and can afford to test every processor out there. We don't. However, we do try and review products while keeping in mind how and why they might be useful. This may not mean much to a seasoned hardware junkie like yourself, but for many users who are interested in computers because of what they can do, not just how fast they can clock, a review that is more than a few benchmarks and a textual summary is something that is quite hard to find.

I do not pretend to have all the answers. And, when it comes down to it, this article is very much an opinion piece. I have no problem with you disagreeing with my opinion; I think your criticisms point out some weaknesses in the article that only came to light after we had finished testing.

3. I don't see the correlation between motherboard compatibility and the price of the motherboard. We happened to find one that worked for $70. I'm not going to try and guess how many of the <$50 boards work, but I would imagine that most of the ones from the larger manufacturers do (IIRC, ASUS had one for ~$40). Refer to the list of compatible boards that was linked to in the article if you're really intersted.

4. Yes, I did "bother" to do a price check, on both PriceGrabber and Froogle. I discovered two things. One: Turions are hard to come by. Two: They vary hugely in what people are charging for them. I saw close to a $100 variance for some models. This is why I decided to judge from AMD's official price list (vs. Intel's official list). For the most part, AMD's bulk price was pretty much at the low end of the scale that the chips fetch on the open market.

5. I agree. Turions are not the cheapest way to go. I'm using a Duron 750 to power my HTPC right now, and it works just fine (total cost: $22+shipping for a new motherboard). But this article wasn't written for the lowballs. It was written for those who might previously have been considering a Pentium M (twice the price), and would want to know that there's an equally low-powered equivalent on the market.

I seriously doubt you will stick around to read this, but if you do, I hope you will come to appreciate the rest of our site. We don't look at computers the same way that most sites do; performance is not our raison d'être. Ultimately, it is the ergonomics of computers that interests us. If that's not your bag, feel free to move on. But, please don't judge our work on the basis of what it's not.

Oh, I plan on sticking around your site. I've been reading and enjoying your reviews for several months. Most of your methods and conclusions are reasonable.

I was, however, so disappointed with this review, that I signed up for a forum account. Sure, you compare the Turion to a Pentium M, and that is useful, but then you go beyond the bounds of your review material:

You also compare it to an older Athlon 64 desktop chip, as if this was a difinitive comparision. This lack of modern desktop chips in your test only convinces people (incorrectly) that Turion ML chips are somehow significantly better in power consumption.

You then state quite clearly that you believe that Turion systems can be offered at competitive prices with desktops, and yet you just admitted to me that you know this is not true. If you can even find a low price on Turions, they're either sketchy stores with zero feedback and strange ordering systems like MobileSearch, or usually they're our of stock.

But you didn't mention any of this in your review, and it felt really half-assed compared to anything else you've done around here.

EDIT: And as for the price list, usually popular processors are sold for less than on those lists, because big distributors make big deals for much less than list. But, because of the low demand for selling Turions and Pentium Ms direct (believe me, low-power enthusiasts are still rare these days), and the relative scaricity of notebook chips over desktop chips, there is some markup over desktop chips. So, I'm surprised the prices are as low as they are...the ones I mentioned at $230 and $270 are very close to AMD's list bulk prices.. But, they will NEVER be in the same price range as desktop chips, not even close.

The article clearly refers to the included A64 as using "the less efficient Newcastle core". The article paints comparisons between the P-M and Turion chips, with the Newcastle included more as an example of a "typical" desktop chip than anything else. I don't think most readers would see it as more than that. (and considering that after 25000 reads, you're the only person who seems confused...) A follow-up article comparing the various core revisions would be a logical next step, but including the highest selling A64 core as an example of the "typical" isn't entirely illogical.



It took me all of 15 seconds to find Turions at prices similiar to those in the article, from reputable stores.


Your issues seem to deal with two real topics:
1. Including one desktop chip as an example makes the article a comparison against every desktop chip available.

2. That one conclusionary comment, "...quiet-oriented system integrators could easily offer extremely quiet, high performance, and power-efficient Turion-based desktop systems at prices very competitive with equivalent performance Athlon 64 systems, and much cheaper than equivalent Pentium M systems.", somehow means that Turions cost the same as desktop chips. Which they don't, but the article never says they do either.

Yes, Turions cost a hundred bucks or so more than a equivalent A64 chip, but look at the market data: a "quiet oriented" system that only sells for $100 more than an equivalent "mainstream" system is very competitive.


Perhaps the solution for the one reader who seems confused by this is to pull the A64 references completely, or add a postscript saying something to the effect of, "Sorry, but this article does not answer every question ever asked about CPU wattages. Please stay tuned for future articles".

Is this article going to be fixed? I already pointed out the error -- Only the MT-32 with 512KB cache should be compared to the Newcastle core, while the 1MB chips should be compared to Clawhammer.

After reading this article, I went to newegg and put together a Turion based PC just for fun. What I noticed is yes, on the high end, Turions are more expensive. But on the low end, a Turion MT-32 was $154 and a Venice 3000+ was $160. Both are clocked at 1.8Ghz and have 512KB cache. Sounds pretty resonable to me.

GHz: Thanks, I've already seen and responded to your point above. I'll have to point it out to Mike to make sure he edits the article.

In the article you mention a few motherboards you tried out that didn't properly support the Turion. Are the mobos you mention in the article the only ones you tried?

It would be nice to have a list of the motherboards that fully support the Turions, not just the ones that work partially. The website you mention in the article doesn't go into much detail as to how well each board supports the Turion, just whether or not it seems to work.

I'm putting a Turion system together for a MythTV box and would like to figure out what all the options are.

I think the article that can compare an identical Turion (both MT and ML models) and desktop Athlon64 on the same process (90nm) will answer the question whether or not buying a Turion is worth the price premium. It should compare chips with idential clock speed and cache at stock voltages and also at equal voltages. For example compare the Athlon64 3000+ E6 Venice against a Turion64 MT-32 and run the Athlon64 at a lower multiplier to get 1.8GHz. Compare them once at stock voltage, and then run both chips at identical voltage (1.2v).

I too would be interested in seeing the new Athlon64 3000+ E6 (ADA3000BXBOX) 51W 90nm compared to the MT-32 at both stock and reduced (1.2v) voltage. Especially given that the MT-32 is only 1.8GHz and is ~ $25.00 more, if the Athlon64 3000+ E6 chip can compete, it may be the sweet spot for S754 systems.

On the other hand, I enjoyed the article immensely.

-Derek

Hi!
I own a DFI 250GB with a Turion MT-32 and I want to say that PowerNow! absolutely works, but just related to moltiplicators. Just install the AMD drivers from the site (the last available are 1.2.2) and then the PowerNow! drivers (the last are 3.0.5).
I've owned also a K8N-E Deluxe, it was good, but there wasn't any possibility to enable PowerNow!.
The only problem with this motherboard is that I can't change the Vcore1 (the vcore of CPU). I try to explain: if I use Vcore=4 in CPUZ.INI it seems that Vcore is well recognized by CPUZ, but using other softwares (such as Sandra or Everest or SpeedFan) the Vcore recognized is still the one I had set in BIOS...
I've heard that some others have the same problem with Opteron and DFI NF4.

marcoland: Welcome to SPCR!!

If you want to change Vcore with any software you can't change the VID in BIOS, something that's extra tempting to do though for T64 users who wants to start it up at full speed. I don't have the same mobo as you but I think they work in the same way with this.
MikeC & Devonavar: For the same reason, it could be why the DFI mobo in the review didn't work with Cool 'n' Quiet.
It all depends on how you set the BIOS. If you didn't do like this then just ignore this post.

AFAIK, there are two Vcore settings in BIOS in most DFI mobos, CPU VID Control (something like: default, 0.8 - 1.6 V) and CPU Voltage Adjusting (+0.0 - +0.7 V).
Just remember to keep VID = default, and for some boards you shouldn't set CPU Voltage Adjusting too high, it may damage the CPU (on Expert mobos for instance).

As a side note: I'd like to see more people to post in the Undervoltable Motherboards thread. It would be even better if you gave a bit more information about the mobo like I did in my post (link above), like number of fan headers and how many of them that works with software like SpeedFan, CnQ, and all kind of voltage settings in BIOS.

It turns out that there's another article with an almost identical title, "Le Turion dans les desktops," written in French, posted in Oct last year by the French site matbe.com. I had missed this before, perhaps because it was in French, but it's definitely worth a read, even if you have to use one of those web translators.

The focus is more on performance benchmarks, along with some power comparisons, not as detailed as ours. They had a Venice-core Athlon 64 3500+ (939) in the mix. Unfortunately, they used it with an Asus A8n-sli Deluxe, which we tested recently and found to have much higher than normal power draw for a motherboard due to its "double" chipset. Naturally, the A64 fares poorly in the power comparison...

Hi mats!
Thanks for your welcome.
I don't know why I had never try it! I had uninstalled PowerNow! drivers, leaving alone the CPU drivers and now the PowerNow! works like CnQ. No differences!!!!!!! It undervolts it self and changes multipliers as well!
I will post some images!
Now I have to go... Ciao!

Good to hear!
Then I guess the reviewers could have got CnQ/PN! if they had done the same. CrystalCPUID could also have been an alternative, and it gives you the possibility to undervolt.

Is it like this with other mobos as well or is it just DFI?

Interesting article, has anyone tried this with an Asus K8V-MX? Reason I ask is thats about the cheapest board I can get here This MSI ATI chipset board isn't very common, and fairly expensive here.

I have a SK6 heatsink on this Turion MT-32 and on yesterday I had decided to make a lapping. I had some problems with the temperatures and I have proceded again to the lapping with more quiteness...
Well, the result is a great surprising...

Before of lapping (idle, min voltage)



Before of the "bad lapping" in full load (screenshot at the sharp moment at the end of 1MB superPi)



AFTER THE GOOD LAPPING IN FULL LOAD (same moment as above for the screenshot)



I had always acclaimed this little monster of heatsink, but I had never beleft that it could reach such as performances.
AND I REPEAT AGAIN THAT CnQ WORKS!

Here are others my scores. I had raised the clock setting the VCore Special up to 126%, the temperatures are not critics at all, and it increases very much the stability. This Turion was able to boot up at 305MHZ, but Prime95 is istantly overstressed (it says generically "for hardware problems"). I had tried to set less aggressive latences for the RAM, but the error persists, also diminishing its frequency. Then I had brought back the RAM to the initial speed and latences, and so I had analyzed the CPU. I had climbed down 1 by 1 each MHzs of the FrontSideBus and I had arrived to the execution of prime95 (one of the best CPU-stress-software I know) at 296MHz without any interruptions for more than 1 day (I had bored about it so I had closed it manually...how long is it? ) with medium temp of 46°C and peaks of 48°C. Therefore, now the CPU runs at 296x9 with voltages that oscillate between 1,616V<1,632V<1,648V in full load, and in idle at 296x4 @1,200V.
The voltages of the full load are indeed "dancers", but the overwhelming majority of the time the voltage is that in the middle.








EDIT:
Prime 95
Two days ago_________Yesterday (after lunch)