FSB and RAM: 1066 and 533 vs 667

[dtw]

I just read this on wikipedia:

Unlike the previous Pentium 4 and Pentium D design, the Core 2 technology sees a greater benefit from memory running synchronously with the Front Side Bus (FSB). This means that for the Conroe CPUs with FSB of 1066 MT/s, the ideal memory speed is PC2-4200. In some configurations, using PC2-5300 can actually decrease performance.

However, there is no reference. Has anyone stumbled across discussion of this "fact" or is it a load of wikpedia rubbish?

[stigweard]

Viewed strictly from static benchmarks like Sandra and Everest, PC2-6400 has better performance for read / writes than PC2-4200 or PC2-5300. I have two 1GB sticks of CL4-4-4-12 PC2-800 (PC2-6400) that runs CL3-3-3-8 as PC2-533. I get better performance with static benchmarks, but there is really no visible difference in performance in applications. If you have any specific benchmark you want done, I can take a few minutes to try it out at either speed.

[Ryan Norton]

I, too, am curious about this. I'm living in Taiwan for a year and I put together a Core 2 Duo PC with some components I already had (VGA, HD, CPU) and some I bought here (mobo, power, cheapo case). RAM is SUPER expensive in Taiwan, though, so I only bought a single 1GB stick of PC4200 (533MHz) cheapo RAM whereas I was planning on 2x1GB PC6400 "performance" memory. What I wonder about is whether it's worth buying the above-outline 2x1GB PC6400 once I'm back in the US and DDR2 prices fall a bit, or if I should just buy another cheapo PC4200 1GB. I primarily use my computer for 3D games and basic MP3 player/web browsing/office productivity stuff.

[jaganath]

Here's an Xbitlabs article about it:

http://www.xbitlabs.com/articles/memory ... guide.html

As you can see, memory types with different theoretical bandwidths do not differ much in practice. For example, there is a 100% difference in theoretical bandwidth between DDR2-533 and DDR2-1067 whereas the difference between the practical results obtained with those memory types is 17% at maximum.

This poor performance of fast DDR2 SDRAM is due to the architecture of Core 2 Duo systems in which memory is connected to the CPU via the chipset and two sequential buses. In this design it is not the bandwidth of dual-channel high-frequency memory that becomes the bottleneck, but the Quad Pumped Bus that connects the CPU with the chipset's North Bridge. Its maximum theoretical bandwidth is 8.5GB/s in Core 2 Duo systems, which only equals the bandwidth of dual-channel DDR2-533 SDRAM. That's why we don't see a really big performance growth if we use memory faster than DDR2-533.

It seems it doesn't make any sense to use memory faster than DDR2-533 on the Core 2 Duo platform. This is not quite so. Memory access latency decreases along with frequency, which can be seen in practical tests.
...

3D games have always reacted readily to any increase in the speed of the memory subsystem. We see that again here, but the reaction isn't very enthusiastic. However, you can see that higher-frequency memory enjoys a certain advantage over slower-frequency one and allows achieving a higher frame rate whereas the memory timings affect system performance less. We shouldn't overestimate the role of fast memory in gaming applications. For example, the results of DDR2-533 and DDR2-1067 differ by only 5-10%, i.e. installing the twice faster memory leads to a negligible performance increase even in games.

...
The previous section may give you an impression that Core 2 Duo systems do not in fact need fast memory. Using higher-frequency memory modules makes the system costlier, yet doesn't lead to any significant performance increases. This is true, in part: memory faster than DDR2-533 can only provide a maximum of 5% performance growth in a majority of widespread applications. The problem is in the front-side bus which is only clocked at 266MHz as yet.

But it doesn't mean fast memory is completely useless for owners of Core 2 Duo systems. Although Intel has limited the frequency and bandwidth of the front-side bus, it's in our power to increase them without Intel's help. So, we'll be talking about overclocking now.

[jackylman]

The article is probably correct. However, I think you'd be hard-pressed to notice a difference in everyday usage.

The best way to run Core 2 Duo is to set the memory speed at 533MHz DDR and then OC the FSB so it matches your memory (333MHz for PC2-5300, 400 MHz for PC2-6400). Of course, this will OC the CPU, but it shouldn't require too much extra voltage to be stable.

To answer your question, Ryan, if you don't want to OC the CPU, then stick with DDR2-533.

[dtw]

OK, so that's quite useful.

I think the main thing I was concerned about as the statement about a "decrease in performance" - any comments on that?

Can I just sum up then? 667 doesn't provide much bonus over 533, similarly 800 doesn't provide much benefit over 667 - this is due to the 266 FSB bottleneck on the chipset. However, the FSB can be overclocked with proportional benefits for 667 and 800, presumably.

So: over 533 has small benefit without overclocking. That correct?

I can see that 266 is (mathematically) a common factor of 1066, 800 and 533 whereas 667 is not - is that the crux of the issue?

[jackylman]

OK, so that's quite useful.

I think the main thing I was concerned about as the statement about a "decrease in performance" - any comments on that?

As others have said, you can usually tighten the timings at lower frequencies to reduce latency. So in some cases PC2-4200 will be faster than PC2-5300. Here's a good page to illustrate this.
http://www.behardware.com/articles/623- ... -test.html

Can I just sum up then? 667 doesn't provide much bonus over 533, similarly 800 doesn't provide much benefit over 667 - this is due to the 266 FSB bottleneck on the chipset. However, the FSB can be overclocked with proportional benefits for 667 and 800, presumably.

So: over 533 has small benefit without overclocking. That correct?

I can see that 266 is (mathematically) a common factor of 1066, 800 and 533 whereas 667 is not - is that the crux of the issue?

Yes, yes, and yes. You seem to understand things now.

[sjoukew]

I can see that 266 is (mathematically) a common factor of 1066, 800 and 533 whereas 667 is not - is that the crux of the issue?

266 * 2,5 = 667
"the crux of the issue" seems to be FSB bandwidth, nothing else.
Intel uses a quad pumped FSB, so the effective frequency can be seen as 266 * 4 = 1066
The memory is 533 double channel ==> 533 *2 = 1066.
Therefore faster memory has small benefit.
Note: I know, the specs here are a bit 'loosely interpreted', only to illustrate the math, and 'the crux of the issue'.

[stigweard]

Jackylman is right. If you aren't planning to overclock, there is no real performance gain apart from on synthetic benchmarks that test the ram only. Here's a couple of quick tests I ran just now using SuperPi 1.5 XS and wPrime. EIST was disabled because SuperPi wasn't computationally demanding enough to cause the processor to change its multiplier.

SuperPi

[email protected], PC2-533, CL3: Pi to 1M - 22.5s
[email protected], PC2-800, CL4: Pi to 1M - 21.8s

wPrime

[email protected], PC2-533, CL3: 64M - 116s
[email protected], PC2-800, CL4: 64M - 114s

I'm getting about 1-2% performance increase over using the ram as PC2-533.

[dtw]

OK, bit confused by this RAM timings/latency thing.

However, I can see that stig has demonstrated that his PC2-800 runs very slightly faster than it does when he changes the timings so that it runs at PC2-533 (is that under-clocking?). I assume that is to be expected and that stig is basically demonstrating, on the hardware he has available, that an un-overclocked PC2-800 is not a lot faster. Is that the conclusion I should be drawing?

I've now read the whole of the Xbit labs article (opera on a phone can only do so much for you) and fully understand the pros and cons of this debate. However! Most discussion, here has been around 533 vs 800. I happen to have a deposit on 667 and was mainly wondering if a quick upgrade/downgrade was in order. However, a downgrade would save about a £1. An upgrade, on the other hand, would be £16 and, as I understand it, would vastly increase my overclocking potential. I think, therefore, that my question has been well answered.

And finally:

In some configurations, using PC2-5300 can actually decrease performance.

In other words if I set the timings on 667 to the slowest possible, it'll be slower than 533 at regular timings?! Is that the point being made? If so, great, why would I set my timings that slow anyway, but why has it been mentioned? Does the combination of the C2D and memory in this particular case give this result? Is it not normal for higher clocked RAM on slowest timings to perform worse than a lower clocked RAM on quickest timings?

That's my last question. You guys have been of immense help and I'm exceedingly grateful. Cheers!

[jackylman]

I assume that is to be expected and that stig is basically demonstrating, on the hardware he has available, that an un-overclocked PC2-800 is not a lot faster. Is that the conclusion I should be drawing?

Yes.

However, a downgrade would save about a £1. An upgrade, on the other hand, would be £16 and, as I understand it, would vastly increase my overclocking potential. I think, therefore, that my question has been well answered.

I got kind of confused when I built my first Intel system recently too. I used a Biostar 965PT board. It turns out that even though I had DDR2-667 RAM, I had to select DDR2-533 in the BIOS for a 1:1 FSB/Memory ratio. That was it's way of asking me if I wanted to use a memory divider or not. It seemed counterintuitive to select a lower memory speed to OC, but that's what you have to do to get the RAM to run synchronously with the FSB. After I did that, I was able to OC the FSB to 333MHz without OC'ing the RAM.

Basically, what I'm saying is that if you hardly save any money, don't change your purchase because you should be able to set your DDR2-667 RAM to run at 533 DDR (i.e. synchronously with the FSB) with tighter timings if you don't want to OC the CPU. If you want to OC, then DDR2-667 guarantees that your memory won't hold you back until you hit a 333 MHz FSB and likewise DDR2-800 is good to 400MHz FSB without doing any RAM OC'ing.

In other words if I set the timings on 667 to the slowest possible, it'll be slower than 533 at regular timings?! Is that the point being made? If so, great, why would I set my timings that slow anyway, but why has it been mentioned?

Because some memory is rated with the slowest timings as the default. For example, the cheapest DDR2-667 is sold with a CL5 rating (5-5-5-15). It will run that way unless you change it in the BIOS, which may or may not require a small RAM voltage increase to run at CL4 at 667MHz.

Clear as mud? Good.

[wowimsocute]

http://www.madshrimps.be/?action=getarticle&articID=472

So does Core 2 need high speed memory to shine? The answer is a resounding no! This is different from AMD AM2 where more expensive memory is needed to get the most out of the system.

On Intel Core 2 pure synthetic memory bandwidth benchmarks show a 30+% increase, but this does not translate in a noticeable performance bump in games and applications, where the increase, at best, is ~6% and this going from cheap high latency PC3200 to expensive low latency PC6400, and while these expensive modules do take the performance crown, their lead over the mostly cheaper PC4200 rated sticks is smaller than 3% in real world benchmarks.

So it doesn?t matter much what memory speed/timings you buy, the value line will suit the Intel Core 2 system just fine, but do keep one thing in mind, as in our testing we found performance actually decreases a bit going from PC4200 (533) to PC5300 (667)!

The Core 2 has a front side bus (FSB) speed of 266Mhz x 4 (Quadruple) ?1066Mhz?, the ram is running at 266Mhz x2 (Dual Channel) x2 (DDR) = ?1066Mhz?, so with PC4200 memory and FSB are running synchronized. When you use PC5300 you are no longer running synchronous with the FSB and a memory divider of x1.25 (5/4) has to be used. The older Athlon XP from AMD also displayed this decrease in performance when running memory asynchronous due to its short pipeline, where memory latency is more important then memory bandwidth. With the Pentium 4 the pipeline was longer and the effect of running asynchronous which increased latency was masked. Core 2 technology marked Intel's return to a shorter pipeline and thus is more similar to the Athlon XP than the Pentium 4.

By running the memory synchronous to the FSB you have the least amount of latency and thus performance is at its best. Why is PC6400, which is also running asynchronous, faster then? Because the memory speed is now that much higher that it compensates for the loss of running asynchronous and overall performance does increase. PC5300 worked great with Pentium 4 but it should be avoided when running Core 2, except when you plan to overclock. PC5300/6400 will give you the extra headroom to increase the FSB while keeping memory synchronous.

[dtw]

/me sighs

So, I dunno how I missed the Mad Shrimp review before but my previous summation seems correct: 667 still "beats" 533 in all but one case when they are both set to the quickest timings. And it's interesting to note that this one case is shown by the Gavrichenkov review to be the only test where the difference between every combination of timing and clock speed is negligible anyway.

This article suggests that to reach the fastest timings for 667 used in the Mad Shrimp review I'd need to up the Voltage on some Corsair Value Select by 0.2, likewise at a clock speed of 533. So, if I want to avoid any tweaking at all I really do need to upgrade to 800 to negate the asynchronicity or drop to 533 and embrace it.

Fin. Thanks all.