how to have a DESKTOP Athlon 64 2800+ @38W max.

[mki]

Disclaimer:
I haven't tested this, but I hope someone will.

I've read that many people have had problems running the Mobile Athlon 64's in desktop motherboards. So, why not buy a faster desktop model and run it underclocked and undervolted instead?

Now there is an Athlon 64 3400+ with a 2.4GHz Newcastle core (ADA3400AEP4AX) available. Because all Athlon 64's are unlocked downwards, one should be able to run a 2.4GHz 3400+ at 1.8GHz. Based on AMD's specifications, the processor should run stable at that frequency with only 1.2V voltage (see the note below).

The maximum power requirements for ADA3400AEP4AX at 1.8GHz/1.2V are:
(VID_VDD * IDD Max + I/O power)
1.2V * 30.1A + 2.2W = 38.32W

What is a processor with a Newcastle core that runs at 1.8GHz? It's Athlon 64 2800+! But this one runs almost as cool as the low voltage Mobile Athlon 64 2800+ (AMD2800BQX4AX has a 35W thermal design power). The big benefit in using a desktop processor is that there are no BIOS support issues or problems fitting the cooler.

Running the Newcastle 3400+ at 1.8GHz/1.2V should be completely ok, because that frequency/voltage pair is listed in the processor specifications. 1.8GHz/1.2V is one of the P-states that Cool'n'Quiet uses. Remember, the low voltage mobile 2800+ is a Newcastle that runs at 1.8GHz/1.2V too.

Note:
I haven't found thermal and electrical specifications for ADA3400AEP4AX yet. It is one of the models for which AMD hasn't released the specifications (why?). I'm assuming that the thermal specifications are identical to the 2.4GHz socket-939 3800+ model (ADA3800DEP4AW). The specifications for 2.2GHz socket-939 3500+ (ADA3500DEP4AW) and 2.2GHz socket-754 3200+ (ADA3200AEP4AX) are nearly identical, so there is no reason why the specifications would be radically different between the 2.4GHz models either. The only difference between the 3200+ and 3500+ models is that 3500+ requires more I/O power than 3200+ (maybe the faster HyperTransport link needs more power). The only real requirement is that the 2.4GHz 3400+ has got the same 1.8GHz/1.2V P-state that the 3800+ has. The 2.4GHz socket-754 3700+ has that P-state too.

Can someone with a 2.4GHz 3400+ test this?

[Jan Kivar]

Well, I'm running my 3000+ 1 GHz@0,85V. If I need performance, I'll bump the multi&vcore to 2 GHz@1,3V (using ClockGen). I haven't run 24h Prime95 yet on the "performance" mode, but this AMD .PDF says that there are some CPUs with that P-state. The 1 GHz/0,85V setting has been Prime95-proofed.

Running 2 GHz@1,25V makes Prime95 crash almost immediately (tried twice, both times less than 5 mins), but I did run 2 GHz@1,275V for three hours just fine.

Cheers,

Jan

[sorenbro]

I'm curious Jan, why run the CPU at less than the specified voltage and speed, seeing that your Zalman CNPS7000 can handle the heat quite fine at 5V?
Does the lower speed reduce the strain on the PSU and enable the PSU fan to run slower, or is the only benefit a lower temperature of the CPU?

[Mats]

mki:I don't get your point. You're just talking about regular underclocking and undervolting! Nothing new.
I run my AXP 2500+ @ 1.375 V and 2442 MHz @ 1.65 V when I need it. No underclocking right now though. People have been doing it for many years.

[burcakb]

I'm curious Jan, why run the CPU at less than the specified voltage and speed, seeing that your Zalman CNPS7000 can handle the heat quite fine at 5V?
Does the lower speed reduce the strain on the PSU and enable the PSU fan to run slower, or is the only benefit a lower temperature of the CPU?

My 3000+ runs at default 1.5V with 56C at full load. If I could lower the voltage (my Abit KV8Pro doesn't allow undervolting), I could run the 7000B on it at less than 5V like I did on my Barton 2500+.

A 7000 at 5V is quiet but still audiable. a 7000 at 3.5V is not.

[Dobby]

And why create unnecessary heat and consume energy at times when you don't really need all the power, but computer has to be online, do scheluded tasks etc.

I use 8rdavcore and my poorly undervoltable tbreb b mostly [email protected] V and raise it when I actually do work to [email protected] V.

[Jan Kivar]

I'm curious Jan, why run the CPU at less than the specified voltage and speed, seeing that your Zalman CNPS7000 can handle the heat quite fine at 5V?
Does the lower speed reduce the strain on the PSU and enable the PSU fan to run slower, or is the only benefit a lower temperature of the CPU?

Partly I run it so low because it can do it...

I was planning to use Cool'n'Quiet, but it has it own issues (not speeding up when necessary etc.). I played a bit with ClockGen, and tried how low voltages were fine. I was quite surprised to hit this low. I actually like that the CPU stays at 1 GHz no matter what I do, and with a click of a button I get more power when I need it (no messy BIOS visits everytime).

IIRC temps weren't affected much from the 1,1V -> 0,85V drop with Zalman running @5V (currently running with BIOS v1.4: 33°C idle/36°C Prime95). I tried running it fanless; after one hour of Prime95 I had temps around 50°C.

I can't really say whether the PSU fan spins slower or not. It is currently the noisemaker in the system though. I'm quite sure that the 1,1V -> 0,85V drop does about nothing on the PSU fan speed. I wish the PSU had an RPM lead, or that I had a Kill-a-Watt or similar...

Cheers,

Jan

[Jan Kivar]

My 3000+ runs at default 1.5V with 56C at full load. If I could lower the voltage (my Abit KV8Pro doesn't allow undervolting), I could run the 7000B on it at less than 5V like I did on my Barton 2500+.

Running at 1,3V I max out at 45-47°C. At 1,5V I also hit about 56-57°C. If the temp readouts are reliable - I need a soft boot to see the "real" values. On a cold boot I see 15-18°C too much. I've read that this is a common problem with Newcastle (CG) cores.

A 7000 at 5V is quiet but still audiable. a 7000 at 3.5V is not.

I have a 6000AlCu on my BX warhorse. It came with a 92 mm fan that is a tad on the loud side @5V. Using a Fan Mate from a 5V rail gives it aforementioned 3,5V, which is fine. It's used to cool my V3 2000 ATM.

Cheers,

Jan

[hvengel]

I have seen reports on Linux sites about dual boot users having problems with the Windows Cool'nQuite drivers (stability mostly) but having no problems with the support for this in the Linux kernel. So problems with the Windows drivers appear to be affecting more than just a few users. Have you downloaded the CnQ drivers from the AMD web site?

[mki]

mki:I don't get your point. You're just talking about regular underclocking and undervolting! Nothing new.

I'm against running the hardware out of specifications. I'm not explaining the normal underclocking/undervolting here, but a completely supported way to run the CPU. That 1.8GHz/1.2V P-state should exist for ADA3400AEP4AX. It means that running that CPU with that frequency/voltage pair is running the CPU in one of its supported stock frequencies.

The thing is that people want the low power Mobile Athlon 64 processors, while they can achieve the same thermal performance with a faster desktop processor and running it in one of its lower P-states. The frequency/voltage pair is supported, so you don't have to worry about stability. And there is no need to wait for BIOS support for the mobiles any more.

Of course, you may be able to run the same processor at 1.8GHz with even a lower voltage, but that's running the processor out of specs. To some people it's ok if it "seems to be stable", but to me it's not. I'm a guy that requires ECC memory, UPS and regular backups. I will never sacrifice stability for silence.

[sthayashi]

I'm against running the hardware out of specifications. I'm not explaining the normal underclocking/undervolting here, but a completely supported way to run the CPU. That 1.8GHz/1.2V P-state should exist for ADA3400AEP4AX. It means that running that CPU with that frequency/voltage pair is running the CPU in one of its supported stock frequencies.

I don't mean to be a dick here, but how do you know that it's an official supported P-state?

[mki]

I don't mean to be a dick here, but how do you know that it's an official supported P-state?

I don't. I have requested that information from AMD, but I haven't received a reply yet. Until then, we can't know it for sure. But, all the other 2.4GHz models have that P-state (except the FX which has different core and P-states than the rest of the processors).

[Schlotkins]

Please be true!

[sthayashi]

Just out of curiosity, what are the other P-states for other AMD processors?

I'm just wondering if it's possible to go a lot lower and still have it be valid/verified.

[mki]

Just out of curiosity, what are the other P-states for other AMD processors?

I'm just wondering if it's possible to go a lot lower and still have it be valid/verified.

For Athlon 64, Sempron and some Athlon XP-M's that information is available here.

The coolest P-state in the new CG-core Athlon 64's is 1.0GHz/1.1V. In that P-state the processors have maximum of 22W thermal design power (full load) and maximum of 9.9/10.6W (socket-754/939) heat dissipation during halt+stop grant. That P-state is available in all but the first "-AP" Athlon 64's and the FX.

If I just roughly interpolate the AMD's performance ratings, at 1.0GHz frequency they would be:

• socket-754 Athlon 64 with 512kB of cache: 1450+ ... 1550+
• socket-754 Athlon 64 with 1MB of cache: 1550+ ... 1600+
• socket-939 Athlon 64 with 512kB of cache: 1580+ ... 1590+

The real performance is probably better than that, because the difference between the memory clock and the cpu clock is smaller than with full speed. So, from CPU's point of view the memory becomes faster, because a single cache miss means fewer lost processor cycles. If that kind of performance is enough for you, just buy an Athlon 64 2800+ and run it at 1.0GHz/1.1V. Cool and cheap.

The 1.8GHz/1.2V P-state is the next best P-state, and it's available in 3700+, 3800+ and (as I told earlier, most likely) in the 2.4GHz 3400+. The TDP is 38-39W in that P-state.

[Ingis]

I did some testing with a 3700+ I just installed. It is performing quite nicely: 2GHz@1,15V

I haven't worked out the heat dissipation, but I noticed it runs at the same voltage as the Opteron EE (although at higher frequency).

[mki]

I did some testing with a 3700+ I just installed. It is performing quite nicely: 2GHz@1,15V

I haven't worked out the heat dissipation, but I noticed it runs at the same voltage as the Opteron EE (although at higher frequency).

According to my calculations, the thermal design power of 3700+ at 2GHz/1.15V is:
1.2V * 31.0A * (2000MHz/1800MHz) * (1.15V/1.2V)^2 + 2.2W = 40.2W
or
1.3V * 39.0A * (1.15V/1.3V)^2 + 2.2W = 41.9W

... depending on which P-state specifications I use in the calculations. The default voltage for 3700+ at 2GHz is 1.3V, but nothing says it couldn't work fine with lower voltages though. With interpolation, the power rating of 3700+ at 2GHz would be 3080+, but in reality it's probably more.

That's fast and cool.

[Jan Kivar]

I did some testing with a 3700+ I just installed. It is performing quite nicely: 2GHz@1,15V

I haven't worked out the heat dissipation, but I noticed it runs at the same voltage as the Opteron EE (although at higher frequency).

Dang, that's shweet! I only manage 2 GHz@1,275V with my 3000+. What's the max. temp while running Prime95?

What motherboard and RAM You're using?

Cheers,

Jan

[Ingis]

40W? Sweet. Thanks mki.

The screenshot was taken about 2 secs after i stopped Prime. Max.temp was about 35C with ambient at ~25C using SLK948U w/fairly low noise 92mm. I'll try a 12dB Papst later on...

System: Asus K8N-E with with a 1GB set of OCZ 3500EB.

[Schlotkins]

1) Tomshardware has a little blurb on the new 3000+ mobile:

http://www.tomshardware.com/hardnews/20 ... 83106.html

Is that 12W they quote?!

2) It seems to me that we should maybe wait for the 3400+ 90nm parts to come out. Even if there is a modest cut in voltage at say 1.8, we could be looking closer to 25W than 40W.

I'm dreaming of a 25W A64, 2.4W Samsung mobile drive, my 10W or so ATI9600 np and about 2gigs of ram with the fanless power supply. That would be a pretty quiet system with good performance!

Chris

[mki]

1) Tomshardware has a little blurb on the new 3000+ mobile:

http://www.tomshardware.com/hardnews/20 ... 83106.html

Is that 12W they quote?!

There must be a mistake. The old low-power models have got a TDP of 35W, just like the new model too.

The old models have got a TDP of 12W at 800MHz, in their lowest power state. Maybe the new 3000+ mobile has got a 12W TDP in its lowest power state too.

[Jan Kivar]

The old models have got a TDP of 12W at 800MHz, in their lowest power state. Maybe the new 3000+ mobile has got a 12W TDP in its lowest power state too.

I'd vote for this also.

Jan