My OC Snapshot; at a glance. Am I, "Okay"??

[Witt78]

Okay, one question. I'm hotter than I would like, but am I okay?

Intel Core2 Quad Q9450 (Lapped)
Arctic Silver 5 (Fresh application)
Big Ass Zalman somethin' or other.

The heat sink is a "Very good", "Older" Zalman air cooler. These days it's prolly the upper half of mediocre.

This screenie will give you the rest of the information you'll need to answer my question.

Am I okay?

 

[Mugz]

A good start. Nice one.

 

[Witt78]

Are my heat levels acceptable to a Q9450?

 

[Mugz]

My rule of thumb for heat is 70ºC at highest under maximum loading. You are still good, but getting close to marginal. Graphics cards are usually safe up to 85ºC.

 

[Witt78]

Yeah, I'm not sure if I'm going to OC the video cards or not. I have two ATI Radeon HD4870's in Crossfire mode. But I have a "Finiky" hardware arrangement, and the game I'm usually playing is "Everquest 2", which is "Finiky" itself.

Typically when I put a load on my video cards, with Everquest 2; I'll see one of my cards kick up to 750 Mhz, with a 30-50% load. =\

I don't think I've got quite the right motherboard for crossfire. I'm getting the impression that my two PCI-Express slots don't run at the same top speed. It's a bummer, but a small one. In the end, I get quite good graphics performance.

I am downloading 3dMark as we speak. I'm going to bench the video cards as they stand now. I'll evaluate how they perform now that I have all these monitoring tools active.

--> Torture testing for the CPU to come. Let's see how stable this system is.

 

[Mugz]

Single-slot is at the full 16x PCIE. Two slots usually have the second at 8x. Only the really expensive Crossfire-capable motherboards have the full 16 lanes on both slots - like the X48-based boards as opposed to the P45.

4870s are good cards, unfortunately they are also small furnaces. The amount of heat they dump into the surrounding air is pretty hefty. Try to aim a fan to direct cooler air at them. You will need one with a reasonable flow.

Food for thought: something I encountered during the P4 era onwards is that the CPU can bottleneck the graphics card(s) quite dramatically. I took my 3GHz chip up to 5.2 stable, benching the system every 200MHz, and up to 4.4 or 4.6 the 3D scores went up in huge jumps (usually by ~3,500 points each time), then from 4.6 on to 5.2 it only went up another 600 or so. That was with 3DMark 2001. That system spent most of its time at about 4.8, because the extra 400MHz equated to an extra 15º on the CPU. Graphics card was an 8800GTS. That extra 400MHz however was worth an extra ~6,000 points with the GPU overclocked as well.

I actually had to build a shroud over the card with its own dedicated 120mm fan to keep internal case temps within reason.

Incidentally, I performed the same tests with an 8400 as well, and the huge jumps in performance stopped at about 3.4GHz. It is all about how fast the CPU can feed the GPU(s).

The point of this is it suggests a pretty good OC strategy: push the CPU up until it stops bottlenecking the card(s), take it another two steps, then start pushing up the card(s).

Highest temp I hit was at 5.5GHz, 78ºC from cold (25º) within fifteen seconds, and the machine bombed on Windows startup. That was also the highest speed the system would switch on at... normal loaded temp at 4.6 was 68º, idle at 40º.

The finicky hardware is probably due to the northbridge or southbridge getting hot, or there is a glitch. Was it as finicky before the overclocks?

 

[Witt78]

Well, the reason for my "Finikie" hardware situation is a long story. I'll try to make it short.

I made a foolish mistake when I was building this system. I was fully intending to run a dual video card, "Crossfire" setup; but I bought one card with the intention of buying the second in 2 weeks. Well, 2 weeks turned into 8 and by the time I had my second card, the manufacturer had changed the entire architecture of the cooling sync.

The second card I recieved had a MUCH more robust cooling system on it. (Luckily the less advanced card does have a plastic shroud that directs most of the air directly out the back of the PC case.)

Anyway, My situation is this... I have two of the "Same" cards... but they're different.
---------------------------------------------------------------
BTW, since my last post I have installed 3dMark, and run benchmarks. My cards "DO" seem to be running at the same speed. (As far as clock speed goes.) (I do have an X48 chipset, btw.)

Sitting idle, both my cards appear in Everest to be running at 500 Mhz. However, once I tax either (Or both) GPU, they will kick up to 750 Mhz. (Power Saver Mode?)

Now that I have run the benchmarks, and seen consistant behavior out of both cards; I'm starting to feel more confident about over-clocking the Radeons.
----------------------------------------------
So you said, I should attempt to open up my motherboards FSB until it stops bottle-neck'ing my video cards? Wouldn't that mean pushing my FSB from 400 Mhz (Already 70'ish Mhz above defaults) all the way to 750 Mhz?

Wouldn't that like.... open a black hole or something?

 

[Witt78]

No! Wait! I'm all mixed up.

750 Mhz is the CPU clock speed. 900 Mhz is the memory speed. Dual Channel Memory, so that's 450 Mhz per channel.

So I'd be pushing my system from 400 Mhz up to 450 Mhz.
--------------------------------------------------------------------------
I can see... ...Perhaps... My CPU heat sinc handling that. However, wouldn't that make my Ram pretty hot? ( DDR2-800 ) (4X 2 Gig Modules)

What's your 2 cents?

 

[Mugz]

DDR2 runs at 1.8V for 800MHz, you should be able to run it at up to 1100MHz with a voltage of up to 2.0V. Higher than that, get heatspreaders, and the maximum I would be prepared to run through RAM is about 2.2V.

The graphics card speed changes = power saving & thermal management (like SpeedStep on a CPU).

You should not have too many issues with OCing the GPUs, though if I were you I would investigate options for different (and more effective) heatsinks. Actually, I was wondering why the one card in the screenshots was running so much hotter than the other.

Personally, though, for reliability, effectiveness, and the sheer wow factor, watercooling is the safest way when overclocking. I am actually something of a fanatic WRT liquid cooling.

Actually, both RAM channels will operate at the full speed of 800/900 per channel. The way dual channel works is pretty much the same way as RAID striping, by interleaving the data between the two DIMMs.

 

[Witt78]

Okay, here is where I'm at...

I have increased my FSB to 450 Mhz. This has my CPU at 3.6 Ghz, and both my DDR2 and my Video Memory at a 1:1 ratio with my FSB.

When I entered bios and made these changes, I took the DRAM Voltage setting off of "Auto" and set it at 1.84 V

For the time being I left my CPU Core Voltage on Auto. (Though I think it needs to be raised slightly.)

(With my CPU Core Voltage set to "Auto", I was reading 1.30 V on the core. Highest voltage of any screenie here.

However, When I started stress tests, Prime95 Crashed Immediatly.
--------------------------------------------------------

You said I should stay between 1.8 Volts and 2.0 volts on my ram.

What would you suggest as acceptable voltage tolerances for my CPU Core?

Just for giggles, I'm going to put my dram setting back on "Auto" and see how Prime95 behaves; while I wait on further advice.

PS - Your knowledge is invaluable. Isn't sharing it wonderful? (My way of saying, thanks.)

 

[Witt78]

Okay, with voltages on auto 450 Mhz is very unstable. (Surprise Surprise, right?)

So, I backed off to 425 Mhz and I'm much closer to stable, however Prime95 torture tests repeatedly return an error on core 1. (the hot core)

So, I'm at the point where I need to start playing with voltages.

How would you suggest I proceed from here? I'm running out of room in my heat sinc. At 3.4 Ghz, I'm running between 48 and 53 degrees with a small load.

your 2 cents?

Getting that 1:1 ratio with my video memory would be very cool. But, so is having a CPU that lasts more than 12 more months. LOL

 

[Witt78]

Okay, after a few hours of tinkering...

I found preliminary stability (Prime95 would run the first 3 test with 0 errors) at 450 Mhz FSB, however my temps were hitting 73 degrees on Core 1. No thanks.

It appears, I have also found preliminary stability with 425 Mhz FSB. Now I'm slowly backing off my voltages, to try and find out where this stability ends.

 

[Witt78]

Okay, well. After a day of tinkering, 3.2 Ghz is the best I can get it with my current cooling setup.

These settings have survived 5 hours of torture tests so far.

(When I backed down from 3.6 Ghz to 3.4 Ghz, I was able to ease off my CPU Core Voltage and my RAM Voltage A LOT!)

 

[Mugz]

Hi.

On the 45nm processors, the max voltage I would consider on air is about 1.4V with a big heatsink, water 1.5V.

A heatsink upgrade would be a very good idea. You might need to go liquid at this stage - is the heatsink surface lapped as well?

The hotter the CPU gets, the less time it will last - it is possible with a powerful cooling setup to have an insane overclock that lasts and is stable.

It also seems as if your CPU is happiest around 3.2 ~ 3.4 with the present setup. The only problem is the waste heat from other components, and also this level of overclock is putting a strain on the northbridge, so it is producing extra heat as well.

From here, I would suggest keeping the chip at a stable overclock while concentrating on upgrading the cooling and getting the system rock solid.

I followed a similar path with my P4 chip - it took about six months in total to get from 3GHz to 5.2GHz, with two cooling upgrades (stock - bigger air - liquid), a motherboard upgrade, and a RAM upgrade (667 - 800) along the way.

IIRC the RAM upgrade allowed me higher than 3.6GHz and the motherboard upgrade allowed me to go over 4.2GHz. The cooling upgrades happened at 3.4GHz and 3.8GHz. Technically there was a second cooling upgrade after the switch to water, although that was only larger piping and a more efficient CPU block.

Incidentally, that P4 was the only CPU I have had to date that has managed a >50% overclock, or even >33%.

Your figures look good, so far, although it would be interesting to see what they look like with the GPU loads at 100% as well... particularly from a cooling/stability perspective.

Oh, crap! Something else I have remembered about the Arctic Silver compound is it requires around 200 hours of varying heat loads (general use) to cure before it reaches its full effectiveness - so that could also be a contributing factor to your temps...

 

[abswindows7]

Fresh application of thermal paste will false the results. Wait a week until the cpu and the heatsink fusion with each other.

 

[The Gonif]

Did anyone else notice that the OP posted his picture and then said he was hot?

 

[Mugz]

Uh... no?

 

[HansVonOhain]

Let the thermal paste cure. You need to do more thermal cycles to make the paste set and perform as well as it should.