Logla

I just checked for a BIOS update tonight for my Gigabyte GA-8INXP. The latest version is F5 and gigabyte claim it is to allow the BIOS to reflect the true temprature of the CPU.

This supposedly means I have a true CPU temp of 26 degrees celcius at the moment. This is nearly a 10 degrees drop from what I used to have. I can't believe for one minute that the CPU would run that cool.

Can it be true?

I'm just running a coolermaster heatpipe HSF on the CPU - no fancy water cooling - and the CPU is overclocked at the moment to 2.6Ghz from 2.4Ghz

craig588

That does seem insane. My AXP with ALK900A with a vantec tornado only does 31C idle.

Sharkims

That 'is' low, what's the abient temp?

__________________
System-1:
MSI K8N Neo Platinum || A64 3200+(200x11) - Zalman 7000A-AlCu || 1x1gb OCZ PC3200@400(3-3-3-7) || 9800XT || WD Raptors 2x36 RAID0
System-2:
MSI 875P Neo-FIS2R || P4 2.4c(FSB:200) - Intel Cooler || 2x256 GEIL PC3200@400(2-3-3-6) || AIW9000 || WD SE 120x2 RAID0
System-3:
DELL Inspiron 9100 || P4 3.0c(mobile) || 2x256mb PC3200(3-3-3-8) || Mobility Radeon 9700 - 128mb || WSXGA+ 15.4" || 60gb 5400RPM 8mb

Logla

Ambient in here is probably about 26-27 degrees. I'm just looking at my temp sensor now and its reading 23 - no make that 24.
and I've overclocked the CPU to 2.9Ghz with no voltage increase. I have never been able to get that high without at least an extra 1v to the core.

I do have good airflow through the case but the temps have never been this low.

Matth

The only thing which would give CPU < ambient (probably equal allowing for error), is if they activated Stop Grant idling

tek

^

whatever that means ?

__________________
[COLOR=White][/COLOR]

Logla

Which they do - what is it.

Matth

From some time back, Processors have had the option of a "low power HALT", where when the CPU is no longer needed during a timeslice, it enters not only a HALT state, but a low power sleep.

The BUS protocol of the AMD makes this more complex, as while the Intel BUS type retained sync, and the CPU could still trace ant activity which conflicted with its internal cache, the current AMD's sleep state is "dead to the world", so the entire CPU cache is writeback flushed (where required) and voided.

Entry to this state requires assertion of the "Stop Grant" - in effect, the request sleep state must be granted, as it is so much more disruptive than the old-style method.

In this state, very little power is used.


So the difference - in INTEL style (and Cyrix, AMD K6 etc.) - the low power halt is controlled purely by the CPU, and it's "snoozing", whereas with the AMD K7, it's controlled by a bit in the Northbridge chipset - and is "fast asleep".


Using idle cooling in this way, also places significant demands on the transient response of the PSU, and can provoke instability or noise in sound - there's a pretty fair amperage being pulsed rapidly!


One other thing - the Intel style of low power halt was a "free lunch" - other than similar power cycling issues and cases of provoked instability, it saved power, but didn't cost performance.

The AMD method, results in starting every fresh timeslice with a flushed cache - giving a reduced cache hit rate, and a not insignificant demand on memory bandwidth. There are also more problems noted (eg. instability) so the default state is usually off.

http://grassomusic.de/english/frames...lish/amdk7.htm

Logla

I think this is probably what they've done. I don't have any external temp sensors so I don't know if the temps I see in the hardware monitor are true but I have noticed that I can overclock the CPU much more reliably now. I can push the CPU as far as 2.9GHz without needing to increase the core voltage - and it stays stable.

Could the stop grant idling make for a more stable overclock?

Matth

It shouldn't do, other than the thermal issues - I'd guess once it gets under heavy load, it'll become unstable.

Software cooling (of which this is actually an example) is often described as good for overclockers, but once you push it to heavy load, software cooling has minimal or no effect.

And unless you just like to see big numbers for clock speed, why bother overclocking unless you DO run CPU intensive tasks?

Logla

True. Most of the time, I just run the CPU at its stock speed but when benching or sometimes for certain games I like to push it a bit further. I stress tested the machine yesterday for 2.5hours with Prime95 at 2.9GHz and all seemed ok.
I can only assume they did something else in the BIOS which has made it more stable.

Optimummind

I heard that Gigabyte's sensor is not very reliable. IMHO, that temperature reading is false. I would invest some money and buy a CPU temperature thermal diode and stick it on the bottom of the processor or processor edge. If you do buy one, make sure you get an LCD screen to see the temperature.

My CPU is also overclocked--from 2.4GHz (533FSB) to 2.80GHz. I have a CPU temp diode attached to the copper waterblock and my reading is 27-28 degrees Celcius at idle and 32-33 C at load.