Ram FSB and Conroe

[The I]

I have a few things I want to sort out conserning the memory sub-system before (hopefully) get my hands on a Core 2 Duo.

First thing: am I correct in believing that any DDR2 speed higher than the FSB will yield no, or only small, performance benefits?
- As I've understood it the bandwidth of the intel FSB is basically the same function of the basic clock as with DDR. For example a 266 FSB is quad pumped so it works like 1066 mhz and it's 64 bit wide so it gives a bandwidth of 8512 MB/s (as 1 byte = 8 bits). DDR at 266 mhz is double data rate and therefore works like 533 mhz, it's dual channel so it essentially has a 128 bit buss and therefore also a 8512 MB/s bandwidth. And since communicating with the ram goes through the FSB there are no benefits in having anything faster than 266/533 mhz when running at the standard 266/1066 FSB? Which is synchronously.
If this is the case then I gather that having fast memory only benefits either if you overclock or have an X6800 with unlocked multipliers so you can boos the FSB. And memory as fast as DDR2-800 for example won?t benefit a non-xtreme-processor unless you over clock the FSB to a fairly impressive 400/1600 mhz. Am I correct?

Second thing: Does anyone know of a good article on how Conroe scales with increased bandwidth?
The closest thing I?ve seen is Anandtech?s core 2-coverage where they simply raise the FSB (and as far as I understand they consider the memory ?bottlenecked? even at that speed). Here a 25 % increase in bandwidth seems to yield averagely 2,4 % increase in performance. But it would be nice with a more thorough investigation if anyone know of one.
And are there any inquiries on the impact of memory timings on Conroe?

 

[harpoon84]

http://firingsquad.com/hardware/core_2_memory_tuning/

DDR2 running faster than the FSB yields a small but benchmarkable improvement in performance. Cas latency also has a small impact on performance. I seriously doubt you'll notice a 1 - 4% boost IRL though, but it's good to have nonetheless for us hardware nuts.

As for Conroe scaling with increased bandwith, here's an article at Xbitlabs addressing E6300 overclocking. The achieved a FSB of 420MHz for a total clockspeed of 2.94GHz.

http://xbitlabs.com/articles/cpu/display/core2duo-e6300.html

From pure observation, it does seem a high (400MHz+) FSB does benefit Conroe, if not significantly then at the very least moderately.

If you look at the benchmarks, a 1.83GHz E6300 is generally slightly (5 - 10%) faster than a stock 2GHz X2 3800+.

However, once the E6300 runs at 2.94GHz vs X2 3800+ @ 3GHz, the margin blows out to 10 - 25%.

Now granted % wise the E6300 o/c was slightly higher but I think the extra clockspeed alone can't be the only factor to the sudden relative jump in performance vs the A64, and I can only conclude that the remainder of the improvement came from the higher bus speed.

 

[The I]

Originally posted by: harpoon84
As for Conroe scaling with increased bandwith, here's an article at Xbitlabs addressing E6300 overclocking. The achieved a FSB of 420MHz for a total clockspeed of 2.94GHz.

http://xbitlabs.com/articles/cpu/display/core2duo-e6300.html

From pure observation, it does seem a high (400MHz+) FSB does benefit Conroe, if not significantly then at the very least moderately.

If you look at the benchmarks, a 1.83GHz E6300 is generally slightly (5 - 10%) faster than a stock 2GHz X2 3800+.

However, once the E6300 runs at 2.94GHz vs X2 3800+ @ 3GHz, the margin blows out to 10 - 25%.

Now granted % wise the E6300 o/c was slightly higher but I think the extra clockspeed alone can't be the only factor to the sudden relative jump in performance vs the A64, and I can only conclude that the remainder of the improvement came from the higher bus speed.

Thanks for the imput on the FSB. Kind of settled that doubt for me.

But I?m not quite sure how to interpret those results, really. By overclocking the Conroe you increase the FSB, yes. But the multiplier remains the same and, assuming memory isn?t a bottleneck, shouldn?t the ?bandwith per mhz? be the same as well on the higher frequency?

Isn?t the case rather that the X3800 can take advantage of the full bandwith of the DDR2-800 memory from the start, so while the ?bandwith per mhz? is constant for the Conroe shouldn?t it, at least theoretically, be lower at 3 ghz?

If that is the case, the article doesn?t prove that the Conroe has gained anything, but only that the x3800 has lost it?s advantage?

 

[harpoon84]

TBH, I'm not entirely sure either, I was just trying to figure out how the E6300 performance jumped so much relative to the X2 3800+ after overclocking.

I think as your original post stated Anandtech touched on the aspect of higher bus speed as well. For a given clockspeed (I think it was 2.66GHz?) 266MHz vs 333MHz yielded a 2.4% performance increase IIRC, so I guess we can safely assume that 266MHz -> 400MHz would yield perhaps 4% or so, keeping in mind the law of diminishing returns as the FSB becomes less of a bottleneck.