Hello. Recently I've been looking closer at what I can do with my ram, and I've got a couple of questions. Overclocking my ram seems to be a trade-off between a high frequency and low timings. It's 2Gb of Crucial Ballistix, PC-6400. It's rated to do either 400Mhz with 4-4-4-12, or 500Mhz with 5-5-5-15, both at 2.2V So, is it better to have a high frequency with looser timings, or vice versa? Which sorts of applications benefit from frequency/timings? Is it as simple as freq divided by timings = speed? Eg, 400Mhz and CAS-4 is effectively as fast as 500Mhz and CAS-5? And finally, does anyone else have this ram? What speeds/timings/voltages have you been able to achieve? Thanks
Not certain on the actual question, but you can more than likely have both tbh, CAS 4 with 500Mhz, or at least something very close, with just a little OCing, the best of both worlds as it were.
With such a small difference in mhz I would go with the 800mhz and cas 4. Going up to 5 and only having 1000mhz isn't going to be much better. It will likely be the same to be honest. If it is faster one way or another you would only notice it in a benchmark.
For Intel CPUs a higher frequency will benefit you more than lower timings. Opposite goes for AMD CPUs.
I don't agree at all, in fact I'd argue the exact opposite. Intel CPUs greatly benefit from reduced timings because the system is latency intensive and the CPUs are designed to compensate by having large cache's (Another reason why Nehalem's have been reduced to 256k L2) and advanced prefetchers. We have previously seen an Intel X48 at CAS-6 and 1,600MHz DDR3 perform the same as the NF790i at CAS-9 2,000MHz. It's also a lot to do with tRD timings of the chipset being used and getting the memory in sync counts for a lot so if you're cranking up the FSB having faster memory = better. AMD memory controllers on the other hand are direct connect so the latency between leaving the controller and getting the answer from memory is inherently low (it's the same with Nehalem I recently found out), so benefit from 1,066MHz DDR2, but it is happy at CAS-5. The problem is Phenoms memory controller is delinked from the CPUs so the internal latency is greater than the original Athlon X2s (K8s).
Uh..that isn't what I've read in the past. And in my personal overclocking experience, intel cpus DEFINATELY benefit more from higher clock speeds as opposed to lower timings. Maybe I'm wrong, though.
Greetings! I would argue that, because of those, latency plays a lesser role once a certain speed is achieved. PS: Maybe an article similar to this with Quad-cores could be made by Bit-tech?
To some degree, but there are many aspects to Intel systems and Intel wouldn't push DDR3 without it benefiting their CPUs, however imo there has to be a balance struck. Intel CPUs benefit MORE from higher FSB and efficient transactions through the northbridge just as much as latency-bandwidth. Remember when Intel went DDR3 and everyone saw no benefit? But if you crank it up to 1600MHz low latency or 2000MHz high latency you get a lot more in it. Given the cost and reliability, I'd still go 4GB 1600 LL over 2GB 2000 e-peen To the thread starter: you're running 333MHz FSB on a P5K so there will be a natural memory divider at 800MHz. I'd push 4-3-3-11-1T if you can - try downloading Sandra and Everest to check your performance and latency at both settings and let us know!
Are there any guides on memory timings and what they affect? EDIT: I've been reading up on it, and yes, theoretically 4-4-4-12 400MHz is the same speed as 5-5-5-15 500MHz but then there are little factors such as read to write delays which make faster memory at proportionally higher base timings slightly faster. The main thing you should keep in mind is the Memory speed compared to the FSB.
If you want to fine tune the overclocking of the ram, you should keep the second number one or more higher than the rest. This has worked for me in the past. Though that may even be on a case by case basis, if different manufacturers have better strengths in specific areas. For example, running your stick at 5-5-5-15 and 1000mhz, you should be able to run it at 5-5-4-15 easily, if not much better. I imagine running it at 4-5-4-12 would work. You might even be able to push it a little further by upping the voltage .1v or two, as well as upping the memory speed another 10%. The last time I did this, I had a 2.4GHz Northwood Pentium 4 (2.4C), running up at 3.31GHz. The default fsb on this was 800mhz, and I upped it to 1100mhz. The ram would have been at 400mhz, and then pushed to 550mhz. It was some corsair matched pair, DDR466 and decently low timings from what I remember. Sadly, this box died a good couple years back. It was running 24/7 though and I gave it a good beating with 3D apps, compiling and gaming. When doing this kind of tweaking, it's best to run memtest86 for a good day or so. I prefer to run memtest for a good solid week, simply because unless I move or a UPS fails it'll never be turned off. All memtest does is do different wipe patterns on the memory from start to finish. It is designed to find errors though. But eventually it'll find one. It could take years to do so, but it'll find one. LOADS if you've got ECC memory. You should only worry about errors with ECC memory if memtest reports them as "uncorrectable." More recent versions of memtest only run it a few passes or some fixed time length (I'm afraid I don't recall which.) You'll have to go into the menu and tell it to loop on the full test. Info on the timings of memory is found on the wiki page for CAS_latency at the "RAM Timing" section.
I know back in the Pentium 4 vs Athlon 64, that was entirely correct - Pentiums loved high clock rates, Athlons thrived on tight timings. I'm not sure about now, so I'd be inclined to take Bindi's word for it.
The more I'm playing with DDR3 the more I'm finding it a little different with newer memory available. There's very little difference (sometimes negative) between 1600 C7 and 1900+ C9, unless you use it to jack the CPU speed as well.
techreport had a very good article I can't find, I thinK I might have it saved. IT had a table with common memory timings and latencies and their theoretical bandwidth. From what I've heard as of right now it's better to use lower speed but lower frequency memory as opposed to something faster with looser timings. For the same reason as bindibadgi has mentioned ddr3 won't be advantageous until around 1600 mhz+ mostly due to the higher latency as the cpu will be waiting for instructions. I wish I could find that table. Originally I was going to go for 1066 DDR2 but switched back to 800mhz and I might put a tiny o/c on them.
gah, could never find that article that had a great table of timings to frequency http://www.neoseeker.com/Articles/Hardware/Reviews/ddr2_vs_ddr3/8.html that gives a bit of the picture although you weren't asking about ddr2 vs ddr3 all in all though you should remember that the overall performance difference between say ddr2 800 at cas 4 and ddr 1066 at cas 5 is very negligble I. Where as the price margin is wider. I came to the conlcusion that I would go for ddr2 800 and spend the money on a better CPU cooler or CPU as that would make a much bigger performance difference per $
