Turion 64 vs Pentium M - by Tech Report

[MikeC]

Turion 64 vs Pentium M - by Tech Report

IMO, the comparison is flawed because it pits the 27W-rated 2.13GHz Pentium M against the 35W-rated 2.4GHz Turion. The fact is that the much closer matched 25W-2.0 or 2.2 GHz Turions are available; 25W (MT) Turion parts are available in every model except the top speed ML44 for only $5 more (compared to the 35W part). The power efficiency is virtually identical w/the P-M at the same clock speed.

Pricewise, at the same clock speed, the Pentium M is usually priced at least $100 more, often considerably more. The few true Pentium M desktop boards runs >$200, while any number of $70 754 boards run the Turions fine.

Still worth a read.

[frostedflakes]

Unfortunately, it looks like they tested the ML-44 100mV above it's rated voltage -- should be 1.35V, not 1.45V. Keep this in mind when examining the power consumption measurements.

But the benchmarks look about right, Pentium-M and Turion64 both win some and lose some. All things considered, I think the Turion64 is very competitive.

Still waiting for your review MikeC. And please make sure you test the Turion64 at their rated voltage.

[~El~Jefe~]

any word on how Newark's work or do not work on a 754 board?

[Ruiner]

any word on how Newark's work or do not work on a 754 board?

I'm running a 3400 newark on my DFI infinity. It runs ok and overclocks reasonably well (2640 at 1.6Vcore, watercooled).
I had issues with my mobile barton rig, and this mobile chip has them as well:
you have to hard shut down to restart (i.e. hold down the power button to power off and then turn on again). Any regular restarts (reset button or in windows or even bios) results in a lockup (fans running but no post). This makes windows installs, overclocking, and certain driver updates a PITA.

FWIW, the previous owner of this chip said it would do 2640 at default Vcore on his nf3 board and I need 1.6 for the same speed. It may be my PSU (phantom 350) or perhaps the power regulation on the mobo.

[CoolGav]

When will Turions be widely available? As an alternative to a Skt939 based X2 system, would one be a fair bit slower, but much cooler? Skt754 boards look to be in plentiful supply, just no Turions as yet! (at least not in the UK).

I have an upgrade itch, which I think would be best served by an X2, but it's worth looking round to see what's out there... Perhaps a laptop would be a good idea... LOL, type "Turion Laptop" into google and 1st hit is for Dell...

[Ruiner]

Newegg has a bunch in stock. I can't answer for the UK.

[frostedflakes]

To anybody wanting to use a mobile Socket 754 processor, I'd recommend the DFI Lanparty UT nF3 250Gb board. I used Turion64, Newark, and Oakville (90nm "Winchester" core) on it, all worked flawlessly. Probably the best board for mobile S754s out there.

[stupid]

From a performance perspective I would choose the Pentium M over the Turion 64. The slower Pentium M bested the faster clocked Turion in most of the benchmarks.

Of course price is always a factor. While the CPU price is kinda competitive, the motherboards definitely are not since P-M motherboards are around $200.

I'm keeping an eye on mobile CPU for when I am ready to upgrade my HTPC. The Meron seems to be a worthy CPU, but that just based on preliminary info.

[QuietOC]

To anybody wanting to use a mobile Socket 754 processor, I'd recommend the DFI Lanparty UT nF3 250Gb board. I used Turion64, Newark, and Oakville (90nm "Winchester" core) on it, all worked flawlessly. Probably the best board for mobile S754s out there.

Why? It is seriously outdated. The Nforce4 and newer chipsets do seem to outperform the nForce3 250Gb. Even my cheapo BIOSTAR TForce 6100. Plus you get (cheaper-better) PCIe!

[MikeC]

From a performance perspective I would choose the Pentium M over the Turion 64. The slower Pentium M bested the faster clocked Turion in most of the benchmarks.

It doesn't seem this way to me, but the other way around. Did you notice that about half of the graphs had the notice "LOWER IS BETTER"? And most of those were won by the Turion?

Not that I think the difference are really significant. It seems clear that the dual-ch memory support of the AOpen board gives the P-M an advatage in most mem-intensive apps, while the higher clock speed of the Turion gives it an edge on most other apps.

[frostedflakes]

To anybody wanting to use a mobile Socket 754 processor, I'd recommend the DFI Lanparty UT nF3 250Gb board. I used Turion64, Newark, and Oakville (90nm "Winchester" core) on it, all worked flawlessly. Probably the best board for mobile S754s out there.

Why? It is seriously outdated. The Nforce4 and newer chipsets do seem to outperform the nForce3 250Gb. Even my cheapo BIOSTAR TForce 6100. Plus you get (cheaper-better) PCIe!

Obviously if you are a gamer go with a PCI-E board. Just be aware that Turion64 may not work very well with anything other than the DFI Lanparty UT nF3 250Gb. For those who are not hardcore gamers and don't need PCI-E, the AGP nF3 250Gb board would be the best choice IMHO because it will be very easy to set up and work flawlessly once it's running.

[acaurora]

I have a bittersweet opinion of DFI. Their motherboards work amazingly well... when they work. I had a LanParty nForce2 400 Ultra, and after about a week of headaches I never had to touch the thing settings-wise until I got a new computer, this time with the DFI SLI-DR. Again, a few headaches, but other than that this baby is very high performance, and rocks.

Like what was said earlier, Single Channel RAM is the limitation for the Turions, but their higher clockspeeds make up for it.

One little interesting thing I did at work was use two of our laptops from HP of the same model, one with a Turion and one with a Pentium M to play a Quicktime High Definition clip (1080i Trailer - Memoirs of a Geisha and 1080i Trailer - UltraViolet). The Turion came out better, however upon moving to our Sony that had a Centrino Duo could play better than the Turion. All three were nearly identical except for minor differences (hard drive space, screen size, etc.) Again, this may not mean anything specific... but just to let you guys know.

[scandium]

Obviously if you are a gamer go with a PCI-E board. Just be aware that Turion64 may not work very well with anything other than the DFI Lanparty UT nF3 250Gb....

Just curious but what are you basing this on? Well more than curious actually as I'm running a S754 nforce4 PCIe board so Turion64 may be a viable cpu upgrade at some point.

[TomZ]

Just curious but what are you basing this on?

I don't want to speak for frostedflakes, but PCIe is much, much higher bandwidth than PCI or AGP. This is why it is preferred/required for video cards going forward.

PCI and AGP are technically obsolete. Obviously there will be for a while a market for these kinds of cards, but eventually everything will go PCIe. Newer motherboards don't have AGP at all, and have only a couple of "legacy" (their term, not mine) PCI slots. Other slots are all PCIe.

[scandium]

Just curious but what are you basing this on?

I don't want to speak for frostedflakes, but PCIe is much, much higher bandwidth than PCI or AGP. This is why it is preferred/required for video cards going forward.

PCI and AGP are technically obsolete. Obviously there will be for a while a market for these kinds of cards, but eventually everything will go PCIe. Newer motherboards don't have AGP at all, and have only a couple of "legacy" (their term, not mine) PCI slots. Other slots are all PCIe.

I should have edited the quote a bit more as it was actually the 2nd sentence I was wondering about (my bad):

"Just be aware that Turion64 may not work very well with anything other than the DFI Lanparty UT nF3 250Gb...."

I don't have any plans to step back to AGP but I am curious as to why frostedflakes doesn't think Turion64 will work well with anything other than this older DFI board.

[QuietOC]

Just be aware that Turion64 may not work very well with anything other than the DFI Lanparty UT nF3 250Gb.

Comversly, It is more likely that any nForce3 boards will have issues with these chips than nForce4 and newer boards. And, if I wanted to splurge so much on a socket 754 cpu, I'd go whole hog and get a 2.6GHz Mobile Athlon 64 4000+ over the wimpy Turion 64's. 1.35V at 2.6GHz with a 1MB L2 cache sounds like it is the top speed bin of the socket 754 parts. The MT40 with 1.2V at 2.2GHz isn't a whole lot cheaper. For that price though, just get one of the newer core Socket 939 cpus (probably an Opteron) and downvolt it if desired.

Also, I wouldn't put much credibility into what any company says about compatibility. The computer I am currently using has a Tualatin core Celeron 1200 running at 1.5GHz on a 1997 i440BX chipset. Yeah, Abit even explicitly says the BX133 RAID does NOT support Tualatins. Hmmm, seems to have worked just fine for the past several years. Even works fine semi-passively with the stock retail heatsink.

[MikeC]

The DFI nf3 board is not the only one that plays nice with Turion 64. In case you missed it in the other Turion threads...
http://angelfall.s39.xrea.com/area2ch/turion-e.html -- a Japanese enthusiast site which details 754 boards that don't work and ones that do -- although the latter is simply limited to functionality: "The below list is collected information that it worked. It does not guarantee the full functional operation."

The "works" list is extensive -- at a glance >30 models.

I can say that the DFI nf3 board recognizes the Turion but does not run CnQ w/it; you have to manually set the clock speed & Vore to get best results -- or resort to Crystalcpuid.

The MSI RS482 we're working with supports the Turion 64 as if it is an A64. The only mb we've found so far that does this.

[frostedflakes]

Just curious but what are you basing this on?

I don't want to speak for frostedflakes, but PCIe is much, much higher bandwidth than PCI or AGP. This is why it is preferred/required for video cards going forward.

PCI and AGP are technically obsolete. Obviously there will be for a while a market for these kinds of cards, but eventually everything will go PCIe. Newer motherboards don't have AGP at all, and have only a couple of "legacy" (their term, not mine) PCI slots. Other slots are all PCIe.

I should have edited the quote a bit more as it was actually the 2nd sentence I was wondering about (my bad):

"Just be aware that Turion64 may not work very well with anything other than the DFI Lanparty UT nF3 250Gb...."

I don't have any plans to step back to AGP but I am curious as to why frostedflakes doesn't think Turion64 will work well with anything other than this older DFI board.

Well Ruiner mentioned cold boot problems on his DFI nF4 Infinity, and I've heard of other people having problems on other boards. Not neccessarily that the CPU doesn't work, just that the board acts quirky.

However, the DFI nF3 is tried and true. I used to own one and it worked flawlessly (which is more than I can say about an ASUS K8T800 board I tried my Turion in -- it "worked" too, if you consider a working board one that locks your CPU voltage in at 1.3V, no matter what you change it to in the BIOS).

But I'm also a little out of the loop with Turion, and don't own a ton of gear due to me running a website that I can test parts on. I didn't realize the MSI RS482 board worked so well. After learning this, I'd say it would probably be the best bet if you're going for PCI-E.

[stupid]

From a performance perspective I would choose the Pentium M over the Turion 64. The slower Pentium M bested the faster clocked Turion in most of the benchmarks.

It doesn't seem this way to me, but the other way around. Did you notice that about half of the graphs had the notice "LOWER IS BETTER"? And most of those were won by the Turion?

Not that I think the difference are really significant. It seems clear that the dual-ch memory support of the AOpen board gives the P-M an advatage in most mem-intensive apps, while the higher clock speed of the Turion gives it an edge on most other apps.

Hmm... perhaps I scanned through the review a bit too quickly while waiting for a meeting to begin. I guess I'll need to review the article again.

[NeilBlanchard]

Greetings,

I wonder how an undervolted Athlon 64 (S939 w/ dual channel memory) would have faired? It seems like the 35watt level is do-able?

[stupid]

Greetings,

I wonder how an undervolted Athlon 64 (S939 w/ dual channel memory) would have faired? It seems like the 35watt level is do-able?

I would say that an 90nm Athlon 64 3000+ undervolted to 1.1v (I think that's the lowest allowable voltage) may do the trick. At stock voltage it consumes 39.9w according to this Xbitlabs article under load. Not sure how CNQ would react to an undervolted CPU since I haven't upgrade from my Socket A yet.

The Athlon 64 3200+ uses 43w under load so I somewhat doubt it will achieve 35w when undervolted.

[MikeC]

Not sure how CNQ would react to an undervolted CPU....

Undervolting is not possible when CnQ is engaged. The default settings of the motherboard are used -- including any minor overvolting applied at full clock by some "enthusiast" motherboards. For total control of P-states, CrystalCPUID is the main option-- it's compatible with a huge number of motherboards and CPU-types. As long as load-dependent P-state changes are supported by the board & CPU, CrystalCPUID seems to support it. You can set the Vcore for each power state, including full clock speed. (Of course, if you set too low a voltage, the system will probably lock up or crash... forcing you to go through the sequences so familiar to overclockers everywhere.)

[QuietOC]

I would say that an 90nm Athlon 64 3000+ undervolted to 1.1v (I think that's the lowest allowable voltage) may do the trick. At stock voltage it consumes 39.9w according to this Xbitlabs article under load. Not sure how CNQ would react to an undervolted CPU since I haven't upgrade from my Socket A yet.

I've run a 130nm Athlon 64 2800+ off 0.85V. What do you mean by "lowest allowable"? Maybe you need a different motherboard?

[rpsgc]

On that front, we are in possession of two other Turion 64 processors, an MT-42 and an ML-42, and we plan to do some additional testing in order to highlight the differences between the 25W and 35W rated Turions soon. I'm curious to see the results myself.

http://techreport.com/onearticle.x/9385

[jaganath]

I've run a 130nm Athlon 64 2800+ off 0.85V

At stock clock speed? That's pretty impressive. Did that allow you to cool it passively?

What do you mean by "lowest allowable"?

I think stupid may mean 1.1V is the lowest Vcore before you have to start scaling back the clock speed. Certainly my 90nm A64 3000+ refuses to POST at less than 1.1V Vcore at stock clock.

[QuietOC]

I've run a 130nm Athlon 64 2800+ off 0.85V

At stock clock speed? That's pretty impressive. Did that allow you to cool it passively?

No. not at stock clock speed. I am sure it was pretty slow.

As far as cooling requirements: It doesn't exatly work that way in my experience. Lower cpu core voltage generally means greater cpu cooling requirements (specifically a lower C/W) not the other way around. Lower core voltage will mean less heat is produced of course, but that doesn't correlate directly into stability. Running a core off the lowest possibile voltage is as hard or harder than overclocking it with unlimited voltage. CPU stability at a certain clockspeed with a given heatsink (whether passive or not) is generally improved by increasing core voltage, and a better heatsink (like the Ninja) will let you run a lower core voltage for a given clockspeed.

[dragmor]

Not sure how CNQ would react to an undervolted CPU....

Undervolting is not possible when CnQ is engaged. The default settings of the motherboard are used -- including any minor overvolting applied at full clock by some "enthusiast" motherboards. For total control of P-states, CrystalCPUID is the main option-- it's compatible with a huge number of motherboards and CPU-types. As long as load-dependent P-state changes are supported by the board & CPU, CrystalCPUID seems to support it. You can set the Vcore for each power state, including full clock speed. (Of course, if you set too low a voltage, the system will probably lock up or crash... forcing you to go through the sequences so familiar to overclockers everywhere.)

Not true, my SN95G5v3 shuttle has different voltage values with CnQ based on what the CPU's voltage is set at in the bios. Not sure why but it does change the lower value.
1.4v = 1.1v CnQ
1.2v = 0.93v CNQ

[dragmor]

What do you mean by "lowest allowable"?

I think stupid may mean 1.1V is the lowest Vcore before you have to start scaling back the clock speed. Certainly my 90nm A64 3000+ refuses to POST at less than 1.1V Vcore at stock clock.

He means that for some reason the newer versions of Rightmark and CrystalCPUid have a lower limit of 1.1v, they wont let you selected less than this value.

I can't understand this limitation since I can go into the bios on my machine, set the multipler to 4 and the voltage to 0.8 and everything is fine after 24hrs of P95.

[stupid]

What do you mean by "lowest allowable"?

I think stupid may mean 1.1V is the lowest Vcore before you have to start scaling back the clock speed. Certainly my 90nm A64 3000+ refuses to POST at less than 1.1V Vcore at stock clock.

Yes, that is what I meant. Based on several posts I've read A64 CPUs refuse to post at stock speed of the Vcore drops below 1.1v.

I suppose it is possible to drop the speed to post with lower Vcore, but it serves no point when trying to compare an A64 & Turion both running at 1.8GHz.

[rpsgc]

Turion 64, take two.

Second, it was discovered that the MSI motherboard used to test the Turion 64 ML-44 incorrectly configures the upper voltage range for that processor at 1.5V instead of 1.35V. As a result, the "load" power consumption numbers for the Turion 64 were five watts higher than necessary. We've updated that graph, as well.