what makes one motherboard faster than another ?

[Innoka]

I was looking over at: mainboard roundup, mainly noticing a 5% difference between the fastest and slowest board, when up pops up the A7V266E hammering the slowest board by 25% and the nearest rival KT266A board by 6% in Linux kernel compilation.
So, how do manufacturers either help or hinder a chipset by that much to produce these discrepancies? (Assuming you trust the methodology of the test!)

 

[BD231]

They keep there performance enhancement secret for the most part, but there isent really much they can do to speed up a chipset. Most the performance difference comes from the system BIOS programing, and NOTHING else.

 

[RossGr]

With the high frequencies involved, trace and component layout, power conditioning and dumb luck can vary the speed of a board. There is no pat forumula, so it is more of an art then science. (Were it a science, all mobos would look and run the same!)

 

[Innoka]

I think you're not getting my point. 25% is a huge difference to come out of nowhere. What does Linux kernel compilation demand?

 

[Moohooya]

Excluding any bios differences, are there ANY differences in speed between mobos, assuming that they all used the same components?

If so, then why?

And, back to BIOS differences. What ends up calling code in the BIOS these days? I know getting/setting the time does, but I believe Windows only gets it once, and keeps its own copy of the time for performance. What about keyboard an HD access? I assume that all still go through the bios?

 

[Peter]

First of all a mainboard design must be really neat and clean - if it isn't, you might not be able to
use really fast timings on the RAM array, or you might even need to leave a couple of performance
enhancement features in the chipset disabled that affect AGP, PCI or IDE busses.

Then of course enabling all those performance enhancements and aggressive memory timings
needs thorough verification, and by far not everyone puts this huge amount of work into it -
the geek community is so keen on having the latest stuff, no matter whether it's finished or
not, that you lose quite an amount of customers if you're out "late" (which would be _after_
testing). Just look at how long it takes in the server and high end workstation market between
presentation of a chipset and actual product shipment ... this isn't "problems", its testing.

But whatever market you're in, whatt you release to customers must work with an almost infinite
number of combinations, so quite often you rather go with quite conservative settings in early BIOSes,
and maybe later cautiously tune it upward. In the server corner, there is more time to test and
tune things, so initial release performance is much closer to what the hardware might be able
to do.

Other than in setting up the hardware and assisting in booting up the operating system, BIOS
isn't much involved in running the system. Only in power management handling it gets to do
something, with little to no effect on speed when fully powered.

Finally, with all of the above being done and perfected, all boards using the same stuff should
benchmark the same. So why don't they? Some companies like to cheat a little on the clock
frequency ... e.g. ASUS boards have been caught running at up to 137 MHz when set to 133, and of
course this speeds up everything by 3 percent over a board that doesn't cheat. Kind of like car
manufacturers sending out "special" cars for reviews, with a few extra horsepowers to give theirs
the edge over a "comparable" competitor product.

regards, Peter

 

[Moohooya]

Thanks for the post peter. I understand the clock difference, intentional or unintentional. But not the memory timmings. Perhaps you could explain this a little more.

As I understood it, DRAM manufacturers come up with memory timmings. First there is a maximum bus speed, say 133MHz, and then there is the number of cycles between the various signals. From an overclockers point of view, we increase the bsu speed, reduce the number of cycles between the signals, until our machine crashes, and then back one of them off until it becomes stable.

Who do the mobo manufacturers not all use the standard timmings? Which timming is it that they change? Certinaly if the design is bad the there is too much noise on the bus, then you won't be able to overclock. But surely they all test to ensure all works with the default timmings and then leave it to us to overclock. So it is obvious why one board would overclock more than another, giving better performance.

Thanks,

Moohoo

 

[Peter]

Well if your board has long traces, many DIMM slots, or otherwise less-than-optimal layout,
you'll get a lot of parasitic capacitance and maybe noise onto your high speed busses -
the implication being that you have to go slower than anticipated.

Regarding the overclockability ... some reasonable board makers (like ECS/PC-Chips) concentrate
on achieving optimal performance at specified speeds, and that's it. Some others put extra
toys in it for the overclockers community, but that's a lot of engineering time wasted for practically
no benefit - not for the 99 percent of the consumers that don't fiddle with their stuff anyway, and
also not for the 1 percent who do ... do the math - the $40 to $50 saved by buying a standard board
buy you lots more CPU and graphics punch than you'll be able to squeeze out by OC, and it'll be
safe and stable right away too. Same money, no time wasted, no work lost to unstable toys.

But don't forget, overclocking is very much luck of draw. No component, be it CPU, chipset, mainboard,
or SDRAM, is tested beyond its advertized speed.

regards, Peter

 

[dszd0g]

I agree with Peter, but with an emphasis on timing issues. The timings on the motherboard are really what make the difference in performance. Most manufacturers follow the reference board layout fairly closely. Intel is generally better than most of the motherboard manufacturers at layout from what I've seen. Via isn't so hot (they may have improved?). I think AMD is somewhere in between. One of the keys to increasing performance is minimizing the distance between components and decreasing the length of the traces. One has to make sure to deal with arranging the components to minimize noise. I have not done a full board design, but the mini designs I've done I've had a clock generator and all the signals had to be able to sync with the clock. The key in that case is being able to set as high of a clock speed as possible and still have all the other signals match clock cycles. This is the whole asynchronous vs. synchronous chip design thing. Maybe someone else can go into more detail on this, I don't have the industry experience with VLSI.

 

[Moohooya]

I'm still confussed. I understand that you need clean signals, components to be close to each other etc. But I can't understand how we end up with a motherboard running 4% faster than another.

If one board runs 4% faster than another, excluding an bios differences, where is that 4% comming from if both clocks are running at the same speed? The PCI runs at 33MHz, the EIDE at 100/66/33, and the memory at 133. Where is the 4% difference? For a 4% difference, every 25 clock edges, one is lost. How can 24 signals get across a bus on the clock edge, while the 25th take two clock edges? Sure, if the memory bus was bad, I could see things like the cas taking 3 cycles instead of 2.5 cycles. but isn;t this something that would be obvious by looking at the bios. Board A runs faster than board B because board B cannot handle a cas of 2.5. Shouldn't all boards ship with the same standard memory timmings? Or is it one of the other busses?