Posts by Sam Evans3

Freddie said:If you're running AMD, you definitely want to stick to 1:1; if you've got an Intel build, then go with 5:4.
It’s all well and good saying that when your CPU temps aren't topping out at 28°C on air cooling.

What would be the reason to keep an Intel system at 5:4? Perhaps if the memory is absolute junk. Otherwise, I always keep mine set to 1:1.

It really all comes down to those memory latencies. If you've got them dialed in at super aggressive settings—meaning those low values—then honestly, don't be surprised by what happens. You could try adjusting the CPU-to-DRAM ratio to 5:4, which would drop the memory speed down to somewhere in the 300 MHz range, but you'll definitely see a hit in performance. Alternatively, you could just loosen up the latencies by increasing those values, though that risks tanking both your memory and processor performance even further.

Of course, all of this assumes the latencies are set incorrectly.

If you get lucky, you might actually be able to hit a stable 3.5 GHz.

The stock cooler is honestly garbage; the fan is weak and they use some weird black sludge instead of actual thermal paste, which is terrible.

You've got your answer—just grab the Cooler Master 7000 AlC for $90, but do double-check if it will fit your setup. It’s about 10x10cm wide, so there's a chance it might interfere with the power supply, depending on what kind of case you're running.

In my experience, overclocking to 3.5 GHz doesn't cause a massive temperature spike—maybe just a few degrees. When I put a decent heatsink on mine, my temps dropped from 58°C down to 50°C, whereas I was stuck at those higher numbers using just the basic stock cooler and air cooling.

Brian Moore2 said:Look, I’m running the exact same P4 2.8 GHz C processor you are, and it isn't even overclocked. When I ran that test, my temps shot right up to 68 degrees.😉

And there's no way for me to mount some makeshift shroud to pull in cool air because the power supply is blocking almost the entire CPU fan—I'm working with a mini-tower here.

PS: My rig struggled for 8 hours at 68 degrees and somehow survived.😁

😲 68?? 😲

That's basically cooking it, man. 👎

Cases where the PSU sits directly over the processor are terrible; they have zero airflow. My advice? Pick up a new case for $67 where the power supply mounts properly above the motherboard instead of sitting right on top of everything.
It's no surprise you're hitting 68C. If you do go shopping, try to find one with side venting or maybe an extra fan slot. That should drastically drop those temperatures.

Brian Moore2 said:Sam Evans3, why don't you run Prime95 (the torture test)
and then let us know the temperatures on your rig and if the system stays stable.

Well, it is summer here, so room temperature is sitting at 77F. My CPU peaked at just 118F, zero errors found after about thirty minutes of testing.

3.5GHz

Just to be perfectly clear, I am using liquid cooling 👍

If anyone is seeing temps climb north of 131F, you might try placing a small paper roll against the heatsink to ensure it draws in cool air from outside rather than pulling in the warm air from inside the case (you will need to leave the side panel open for the test 😉).

Here is the breakdown showing the difference between running memory at 400MHz with aggressive versus relaxed timings:

PC Mark 02

2-3-5-2 - 8298
3-4-8-4 - 8088
3-4-8-4 (500MHz) - 9142

Aida32 (read)

2-3-5-2 - 3007
3-4-8-4 - 2883
3-4-8-4 (500MHz) - 3641

Now, I'll let you all draw your own conclusions. Is it actually better to push memory to its absolute limit with tight timings and then bump up the FSB like those "professional" tech sites always suggest? Or is it smarter to relax the latencies and jump to a much higher frequency?

The gain in the first test was 6 points, but in the second test, the improvement is five times larger than what you get by just messing with the latencies.

It isn't just about raw memory speed; it's about how those programs actually utilize resources—use your head a little. Even when you run dedicated tests, tweaking the latencies rarely yields any significant gains. That’s exactly what my next test will prove.

I just finished running the PCMark 02 memory test, keeping all other variables constant:

500MHz 3-4-8-4 = 9142

400MHz 2-3-5-2 = 8298

Take a look at that gap. It's a margin you just don't see as clearly when running 3DMark.

I don't really buy into all those quasi-scientific tests floating around, so I decided to run my own latency test to see what was actually happening.

P4 2.8 GHz C @ 3.5 GHz (250 FSB) configuration.
The FX5900 defaults.
Just one stick of 512MB DDR400.

1) Memory running at 400MHz with a 2-3-5-2 timing profile.

3DMark 2001 score: 15,930.

Aida 32: read- 3007
I am still waiting for more information regarding the current situation. It seems that we have reached a bit of a standstill here. I will continue to monitor the developments closely and provide updates as soon as they become available.

2) Memory running at 500MHz with 3-4-8-4 timings.

3DMark 2001 - 16167

Aida 32: reading at 3641.
write- 1370

So, my conclusion here is that regardless of those deceptive benchmark tests, memory running at 400 MHz with relaxed latencies actually performs significantly faster than memory running at 400 MHz with aggressive timings.

It seems those responsible for that test clearly need a new pair of glasses; they either swapped out the images or simply lack the basic competence required to conduct a standard test.
Furthermore, 3DMark isn't actually designed to measure memory speeds, which just goes to show that some of these people haven't got a clue.

Pentium 4 2.8 GHz pushed to 3.6 GHz at 1.6V
Corsair DDR400 memory clocked at 514 MHz with timings 3-4-8-4
Temperatures stay under 45°C under full load, dropping to about 40°C during the winter months

Aggressive latencies are essentially useless anyway; it is much better to increase the memory frequency than to struggle for marginal gains through tighter timings