CPU/Bus/RAM synchronization

[SEANT]

I have a theoretical question about computer specs with regard to computationally intensive rendering programs.

 

While researching a new computer purchase I noticed Dell offers several CPU speed, Bus speed, and Memory speed options for their Precision 490 model. The question is: How beneficial is it to coordinate those speeds in common intervals?

 

To clarify, if one chose the 1333 bus, should they match it to the 2.66 GHz Cpu and 667 MHz Ram. I have to say this does appeal to my sense of symmetry and harmonics. It may be an erroneous assumption, however. If there were any "hiccup" in sychronization it would guarantee a delay of a full clock cycle. Perhaps mixing the speed would insure only a fractional clock cycle wasted.

 

Addmitedly, I'm splitting hairs here, and neither of those delays would likely make a significant difference. It's just that I'm weighing the option of a single CPU model as listed above - or Dual 1.60 Ghz CPU, 1066 Bus (CPU factor 1.5), 533 MHz Ram.

 

I believe all of the other applications I use would perform well, even with the slower CPU. The rendering application, with the potential of billions of iterations, may benefit from a the proper synchronization. It would definitely benefit from the second CPU.

 

Any thoughts?

 

Sean

[AJLynn]

You really don't need to worry about the relationship between the RAM, bus and CPU - so log as they're spec'ed to work together, you won't experience a slowdown from using fractional multipliers, since there is buffering built into the process. Get the highest numbers that fit your budget, prioritizing CPU clock speed (when comparing CPUs in the same product line - e.g., don't opt for a Pentium D over a Core2 Duo because of clock speed) over bus speed because bus speed is less a factor in render speed than CPU.

 

WRT the single/dual vs slower dual/dual issue, you can compare total MHz speed allowing for a few percet efficiency loss adding more CPUs, for applications that multi-thread well, e.g. a render process. But other don't - Photoshop, for example, or the Max interface, would be slow if each core is 1.6GHz. I think going that low is sarificing too much.

[SEANT]

Thanks for the feedback.

 

If coordinating frequencies is not an issue then I guess the real choice with the given budget is:

 

Single (Dual Core) 2.66 GHz, 1333 bus, 2 GB Ram (667) or

 

Dual (Dual Core) 1.86 GHz, 1066 bus, 2 GB Ram (533)

 

The first provides roughly 5.32 GHz processing, the other 7.44 GHz. Even with the overhead of increased multi processing, that is a significant gain.

 

I can't really complain with the performance of my other apps - AutoCAD, Rhino, Office Suite, CorelDRAW - on my current setup, so the lower single threaded performance of the choices above should suffice.

[AJLynn]

The 1.86 is actually pretty fast, it's as good as a 3GHz P4. With 4 of those I don't think you could have any complaints.

[tecton3d]

I second AJlynn's response for the dual dualcore setup, #4 - 1.86 core2 cores will pay off nicely when rendering and will be more than sufficient for non-smp tasks.

[SEANT]

Thanks again for the feedback. It's nothing but the waiting to stress over now.

[Greg Hess]

You have to be very careful when comparing different numbers in a computer to one another. The amount of dividing and multiplying going on can give you a splitting headache.

 

Case in point, lets use your example.

 

1333 FSB or 1066 FSB

 

The memory should be the same right? Well this is where it all gets really confusing. The FSB is actually quad pumped on the intel boards.

 

The actual FSB prior to being multiplied is...

333 and 266

 

Those numbers look familar to anyone?

 

The memory gets even more confusing...especially given the nomenclature they've decided to use to befuddle the average consumer.

 

First there was ram. It had a nice little rating like PC133. The 133 was the megahertz...oooh so nice and easy.

 

Then you got DDR...double data rate. So you double the number...and va la, now the 133 is 266. Around this time manufacturers start adding odd names to the ram, like PC2700, and PC3200, just to confuse the living #$()&@$ outta everyone...its all marketing bs meant to give you an idea of stellar performance.

 

Well now we have DDR2...that means you double it again. 266x2 =533.

 

But wait, we aren't done yet.

Thats right...you run the DDR2 in a dual channel system. 533x2=1066

 

Ta da! Make sense now? If it does, you should probably be a computer consultant. Cause I'm still confunded as hell.

 

Basically 667 for 1333, and 533 for 1066. Easiest way to figure stuff out is just to divide the intel fsb by 2 and compare to the ram, or multiply the ram speed by 2, and compare to the fsb.

 

Any faster ram is primarily for overclocking purposes right now. If you were building a conroe system, lets say an E6600 (2.4 ghz core duo2) which runs at 1066 FSB, and purchased some PC5300 667 megahertz DDR2 (remember multiply by 2 for 1333), you could then (baring any other problems) run the FSB at 24-25% higher frequency, netting you about 3.0 ghz...right at the ram's actual sweet spot.

 

However if you just bought the faster ram, and didn't fiddle with either the timings or the FSB, you wouldn't gain any actual performance from the more expensive dimms.

[room217]

Nice explanation Greg!

 

Thanks.