upcoming "Turion" mobile AMD 64

[sgtpokey]

So it's either 2.1 GHz @ 21 W vs. 2.2 GHz @ 25 W, or
2.26 GHz @ 27 W vs. 2.2 GHz @ 25 W.

Only one problem:

There is no 2ghz or 2.2ghz 25watt Turion. The 25watt line tops out at 1.8ghz:

http://www.silentpcreview.com/article224-page6.html

---------

In any case, we'll see how they price the 25watt line, I've only seen price estimates for the 35 watt line. I do not expect the 25watters to be cheap (I bet the price = Pentium M) and I kind of wonder how widely-available the 25watt cpus will be.

From a desktop perpective, both mobile cpus are interesting.

[smilingcrow]

In any case, we'll see how they price the 25watt line, I've only seen price estimates for the 35 watt line. I do not expect the 25watters to be cheap (I bet the price = Pentium M) and I kind of wonder how widely-available the 25watt cpus will be.

Jump back a few posts and you'll have a link to AMD's pricing. There's very little between the ML & MT pricing, about $5.

The assumption I make is that they will struggle with yields of 25 watt processors, hence the need for the 35 watt units. Strangely though, the low voltage mobile Sempron is 25 watt only!

Turion is no Pentium M, but given the starting point of the K8, I think they’ve done well. Although it’s still all a matter of delivery. I wonder how many low voltage Athlon 64 notebooks have been sold? The price differential there was more like $50, which didn’t help.

[frostedflakes]

The 25w line will be extended to 2.2GHz by Q3, though (as long as AMD doesn't have any unexpected yield issues).

I'm waiting for Turion, but am faced with the dilemma of 25w 1.8GHz vs 35w 2.0GHz? The 10x multi of the ML-37 would be nice, but I'll probably end up going with the MT-34, which is about $100 cheaper.

Just hope it's compatible with my DFI.

[Mats]

Only one problem:

There is no 2ghz or 2.2ghz 25watt Turion. The 25watt line tops out at 1.8ghz:

Wrong.
You didn't complain about there's no 2.26 GHz PM either (Intel fanboy?), for now.
AFAIK, TDP is based on he fastest planned CPU, in this case the MT-40 running at 2.2 GHz. See my first post.

I do not expect the 25watters to be cheap (I bet the price = Pentium M)

Wrong.
The price difference is $5 between ML and MT according to AMD.

Edit: Now I saw the other posts, oh well...

[yeha]

Wrong.
You didn't complain about there's no 2.26 GHz PM either (Intel fanboy?), for now.
AFAIK, TDP is based on he fastest planned CPU, in this case the MT-40 running at 2.2 GHz. See my first post.

woah now! remember to keep it friendly around here

though i can see how people's moods get riled up about cpus - amazing how fascinated we can get over 0.5 grams of silicon.

[Mats]

Just hope it's compatible with my DFI. :lol:

I better post it here as well (Though fr0stedflakes probably saw my post at XS):
Athlon 64 revision E won't work on some Nforce 3/4 boards

Yep, that probably includes Turion as well.
This is a nightmare for Turion dreamers, since fery few mobos work with 90 nm mobiles already(I've only seen two).

[frostedflakes]

It's a gamble I'm ready to take. If anybody will release a compatible BIOS, it will be DFI.

[Mats]

If anybody will release a compatible BIOS, it will be DFI.

True. And besides, you already got the mobo.

[Mats]

From XtremeSystems forums.

amd told me theres a difference, there are different transistors you can chose when building a cpu, some consume more power and switch faster, some consume less power and switch slower, just an easy example.

amd told me that the turion uses a different set of transistors than the regular a64s in 90nm and uses around 30% less power in some situations than a desktop a64 with the same vcore and clockspeed.

i dont know if this i true though, or if its just marketing
after all having a different mask for the turion cpus ould mean a lot of extra work...

[Mikael]

From XtremeSystems forums.

amd told me theres a difference, there are different transistors you can chose when building a cpu, some consume more power and switch faster, some consume less power and switch slower, just an easy example.

amd told me that the turion uses a different set of transistors than the regular a64s in 90nm and uses around 30% less power in some situations than a desktop a64 with the same vcore and clockspeed.

i dont know if this i true though, or if its just marketing
after all having a different mask for the turion cpus ould mean a lot of extra work...

Now that's some interesting stuff! Thanks for the link!

[Mats]

From Anandtech.

Much as we suspected, all of the power optimizations that went "into" Turion 64 are all transistor level optimizations. Basically selecting transistors that provide better thermal and power characteristics at the expense of lower switching frequencies. Given that the Turion 64 runs at multiple speed grades lower than the fastest desktop Athlon 64s, this tradeoff makes sense, but it also means that Turion 64 is no Pentium M killer.

I don't believe that, it's not that simple. It's possible that the PM 770 is faster, but look at the price. The fastest T64 should be a good competitor to an equally priced PM.

Code: Select all

Intel® Pentium® M Processor Mar '05 (03/20)
Mobile (uFCBGA / uFCPGA) Price Price   
770 (2M L2 cache 2.13  GHz 533 MHz FSB 90nm) $637
765 (2M L2 cache 2.10  GHz 400 MHz FSB 90nm) $637
760 (2M L2 cache 2A    GHz 533 MHz FSB 90nm) $423
755 (2M L2 cache 2     GHz 400 MHz FSB 90nm) $423
750 (2M L2 cache 1.86  GHz 533 MHz FSB 90nm) $294
745 (2M L2 cache 1.80  GHz 400 MHz FSB 90nm) $294 
740 (2M L2 cache 1.73  GHz 533 MHz FSB 90nm) $241
735 (2M L2 cache 1.70A GHz 400 MHz FSB 90nm) $241
730 (2M L2 cache 1.60B GHz 533 MHz FSB 90nm) $209
725 (2M L2 cache 1.60A GHz 400 MHz FSB 90nm) $209
715 (2M L2 cache 1.50A GHz 400 MHz FSB 90nm) $209

Code: Select all

AMD Turion 64 ML-37 $354 
AMD Turion 64 ML-34 $263 
AMD Turion 64 ML-32 $220 
AMD Turion 64 ML-30 $184 
AMD Turion 64 MT-34 $268 
AMD Turion 64 MT-32 $225 
AMD Turion 64 MT-30 $189 

[Mariner]

I'd say it looks as though the Turion chips, especially in the mid to low range will outperform their Pentium M price-range equivalents. It looks as though the real top-end Pentium Ms might have a slight performance advantage over the faster Turions (albeit without 64-bit or SSE3 support) but these are much more expensive. By "performance" I mean your usual benchmarks here. In terms of battery life (for mobiles) and overall heat dissipation, Pentium M should have a moderate advantage, moreso over the ML Turions with TDP up to 35W than the 25W MT versions. This assumes that the Turion TDP isn't overstated and the Pentium M TDP isn't understated as has been the case with AMD and Intel chips in the past!

I'm still wondering how well these Turions will be able to undervolt to reduce power consumption. The initial figures linked on Xtremesystems were encouraging, however. One worrying thing is that I seem to remember reading that AMD is asking motherboard vendors not to support Turion on standard desktop motherboards. Not sure where I saw this but if true it could be disappointing for SilentPC-ers!

[Mikael]

One worrying thing is that I seem to remember reading that AMD is asking motherboard vendors not to support Turion on standard desktop motherboards. Not sure where I saw this but if true it could be disappointing for SilentPC-ers!

It's not that big of a problem, since the Venice and San Diego cores seem to undervolt extremely well. Now, I know that Turion is made from some different substrate, but I still think that Venice and San Diego is cool enough to fit in very quiet computers.

[sgtpokey]

Mats wrote:

You didn't complain about there's no 2.26 GHz PM either (Intel fanboy?), for now.

Man, I just read this!! What's up with the personal attack? The only room for "fans" when it comes to SPCR are the 5 volt modded ones.

You are correct though, I didn't note the 2.26 PM was not available, but I thought you were making a roadmap mistake i.e.:

In the roadmap published in this link:
http://www.silentpcreview.com/article224-page6.html

There was no 2.2ghz 25 TDP Turion, in fact the roadmap only showed a 2.0ghz 35 TDP Turion. So I thought you had made an honest mistake and mixed up the roadmap.

Intel had already roadmapped the 2.26 ghz PM so I thought your comparison was the roadmapped top-end PM at 27tdp vs the roadmapped top-end Turion at 25tdp. In both cases the comparison is to mythical, roadmapped parts as neither cpu is/was purchasable at the time (or currently)

No reason to start calling people names!

<end of Long explanation that should not have been necessary>

_______

As for pricing, Some of the info doesn't add up, And I will reserve judgement as to price and availability of the 25watt line: my reasoning is still along the same lines as to what smilingcrow said:

The assumption I make is that they will struggle with yields of 25 watt processors, hence the need for the 35 watt units. Strangely though, the low voltage mobile Sempron is 25 watt only!

the point being, why even have a 35 watt line if they are going to have great yields on the 25watt line? For that matter, why have two seperate lines differentiated by only $5?

[Mats]

Mats wrote:

You didn't complain about there's no 2.26 GHz PM either (Intel fanboy?), for now.

Man, I just read this!! What's up with the personal attack?

It wasn't my intention to make you angry. I am sorry about that.

Actually I didn't even know that "Intel fanboy" was "calling names". Someone called me an Intel fanboy recently here at SPCR,
I just thought it was a bit funny because I prefer AMD mostly. Maybe it means something I don't understand (I'm still not good in English, coming from Sweden), and that's certainly not a good excuse, but maybe an explanation.

As for the roadmaps I was referring to the first ones posted in this thread and at SPCR AFAIK, post #3.

So you see that my post was a reflection of your way of not looking at AMD and Intel in the same way, which you did because you didn't know about the future AMD models. So it's all a mistake!

I really hope this makes you feel better, otherwise tell me so.

[sgtpokey]

Mats,

No problem! I overreacted anyway.

In any case, as soon as these become available and are reviewed, we'll get a better idea of what to do with them...

[Mariner]

I've just been looking at the following thread once again and it has just struck me that we could end up with extremely low heat dissipation with the Turions if underclocking.

Xtremesystems.org

If the sample tested there is representative of most of the chips, very low undervolting might be achievable whilst still keeping stock speeds. As I understand it the standard voltage for the Turions is 1.4V but in the example above, a 2GHz clock was achieved at just a little over 1.0V!

The fastest of the 25W versions, the MT-34 has a clock of 1.8GHz at 1.4V so if it is possible to run at stock speed but reducing voltage to 1.0V, CPU Power indicates the chip would be using just 12.7W! This is even lower than the Dothan tested by SCPR though admittedly the Dothan was clocked higher! Very encouraging and I, for one am looking forward to seeing how the Venice/San Diego chips will compare with Turion. Hopefully Xtremesystems' sample is indicative of the chips.

[Mats]

Mariner: The thing is that they don't run at 1.4 V. It's 1.35 V for ML as you can see in that CPU pic and 1.1 V for MT according to the guys in the thread. You can see my posts speculating about the voltage over there, they corrected me.
But still, 1.35 is not far from 1.4 V, which means 2400 MHz, 35 W, 1.35 V @ 2000 Mhz, 1.0 V = 16 W.

[frostedflakes]

That transistor stuff sounds like BS to me, but who knows. I'm still trying to get my hands on an MT-34 (1.8GHz, 1.1v?). I'll be sure to let the SPCR community know how it works out.

Even if it isn't an uber overclocker, 25w is appealing. If a person could even undervolt it a little, this would be great for passive cooling.

[Mats]

Frostedflakes: You're at XS too, right? Have you heard anything new about availability?

[frostedflakes]

Yup, I post occasionally at XS under a similar name.

I was planning to buy my Turion from memoryexpousa.com, but a couple days the Turion category disappeared from their site. So I'm not completely sure if they intend to sell them anymore. I sent their sales staff an email, and am waiting for an answer. I have yet to see any other store that has these processors advertised. Memoryexpousa.com would be ideal for me to order from, because they are in the states. If it comes to it, I may have to order overseas and have it shipped here.

[Mats]

Keep us informed about the Turion you'll get, I'm very curious!
One thing most of us would like to see is what the lowest Vcore is for stable stock speed?
Also if there's some difference performance difference to your old 2800+ at same speed?
Since you're at XS too I must ask what kind of cooling you will use?

[frostedflakes]

The MT-34 should be here either 4/20 or 4/21. The first thing I intend to do is see how far it will undervolt at default clock speed. Then I'll try max clock at default voltage (1.2v), max clock at default desktop voltage (either 1.35v or 1.4v), and then max clock at any voltage.

For cooling I will be using a modified XP-90, Panaflo 92mm (medium), and Arctic Silver 5.

[Mariner]

The MT-34 should be here either 4/20 or 4/21. The first thing I intend to do is see how far it will undervolt at default clock speed. Then I'll try max clock at default voltage (1.2v), max clock at default desktop voltage (either 1.35v or 1.4v), and then max clock at any voltage.

For cooling I will be using a modified XP-90, Panaflo 92mm (medium), and Arctic Silver 5.

(Rubbing hands together)

Excellent! Exactly the information I'm interested in!

[vapb400]

The MT-34 should be here either 4/20 or 4/21. The first thing I intend to do is see how far it will undervolt at default clock speed. Then I'll try max clock at default voltage (1.2v), max clock at default desktop voltage (either 1.35v or 1.4v), and then max clock at any voltage.

For cooling I will be using a modified XP-90, Panaflo 92mm (medium), and Arctic Silver 5.

(Rubbing hands together)

Excellent! Exactly the information I'm interested in!

i cant wait!

[~El~Jefe~]

Socket 754 means AGP which kindof sucks! Any PCI-e s754 boards out there? This is simply a rev. E0 winchester core (with single channel s754 1 MB cache etc) right?

PCI-e has absolutely no effect on anything besides being, um a different slot. There are no conceivable vid cards out there that gain a thing from that over agp. AGP is no wheres near maxed out. I rather have it personally. Pci-e is only for when all the pci slots will utilize it. right now, its only for gfx card, which, unless board designs change, a gimic. SLI is on a lot of pci-e, so I guess thats something to consider, but um.... yeah sli is so needed...

[frostedflakes]

So far it's running like a champ.

Instead of posting the same stuff on all the forums I've visited, all my results will be the Turion thread I created at EOCF. Here's the link.

http://forums.extremeoverclocking.com/s ... ge=1&pp=20

[Mats]

I saw the mobo even recognised it as a Turion with that BIOS! No compatibility issues at all? Will you do any benchmarking to compare with your old CPU just check if it got lower performance att same speed? You know what I'm talking about, Saaya mentioned something about slower low power transistors used. Still don't know what to believe. Thanks for keeping us informed!

[frostedflakes]

SANDRA CPU benchies between Oakville and Turion are nearly the same at 1.8GHz. A couple points off at the most, nothing that can't be contributed to normal variation in the tests.

Whether these "slower transistors" (assuming there even are any) effect maximum clock rate has yet to be seen.

[wim]

PCI-e has absolutely no effect on anything besides being, um a different slot. There are no conceivable vid cards out there that gain a thing from that over agp. AGP is no wheres near maxed out. I rather have it personally. Pci-e is only for when all the pci slots will utilize it. right now, its only for gfx card, which, unless board designs change, a gimic. SLI is on a lot of pci-e, so I guess thats something to consider, but um.... yeah sli is so needed...

i heard that modern vid cards eg the 6600s are created with PCI-e in mind and then the agp versions are made with the same circuitry as the PCI-e, except with a bus and necessary circuitry to interface with an agp slot. so the agp versions are kind of a botch job, and the additional circuitry on the card has the following effects

• agp card runs hotter due to the increased ineffeciency
  card is slower due to pci-e -> agp conversion
  agp version is more expensive due to more components

while i'd agree that agp theoretically doesn't need to perform any worse than pci-e (yet), in the real world of 'cost engineering' there is a conceivable gain for avoiding agp now