Memory question (dual channel)

[BlueHooHoo]

I could do with some clarification!

I'm looking at some Corsair XMS DDR or some XMS DDR TwinX (the dual channel stuff).

The MoBo supports the latter but it is considerably more expensive. Is there a noticeable performance difference between dual channel and standard memory?

They’ll have the same CAS latency. Will be on a on an A64 3500+,with Asus A8V; I've no real plans to overclock.

More expensive memory will just about tip me over budget would it be worth it?

[Lee @ SCAN]

TwinX is the "paired" memory thats been tested and certified to run in dual channel mode.

You can use 2 sticks of normal XMS - but if theres any issues with the running of Dual DDR don't say we didn't warn you

Personally - I always go for the TwinX in my systems as you get the assurance that they've been pre-tested and that the sticks are 100% compatible with each other.

Go on... spend a little more you know you want too

[Agent]

The normal stuff will run in dual channel mode fine. Order the sticks at the same time, and its almost certain you will get ones that came off the production line one after another anyway.
TwinX is more marketing than anything.

[Agent]

Originally Posted by WildmonkeyUK

TwinX is the "paired" memory thats been tested and certified to run in dual channel mode.

You can use 2 sticks of normal XMS - but if theres any issues with the running of Dual DDR don't say we didn't warn you

Personally - I always go for the TwinX in my systems as you get the assurance that they've been pre-tested and that the sticks are 100% compatible with each other.

Go on... spend a little more you know you want too

And the chances of having 2 identical sticks not work in Dual DDR mode ?
Almost non, considering that its down to the motherboards memory controlers to deal with the memory.
Even then, most issues are sorted out with a BIOS update...

[Ceefer]

Slightly off topic but interesting to note - I was running 2 x 512mb Twinmos (m-tec chips) on my NF7. Had a problem with one memory stick so as a temporary measure replaced it with a spare 256mb stick. This still runs just fine in DDR mode with virtually no performance loss, ie 1 x 512mb, 1 x 256mb. I thought both sticks had to be the same size, but it seems not.

[Agent]

Originally Posted by Ceefer

Slightly off topic but interesting to note - I was running 2 x 512mb Twinmos (m-tec chips) on my NF7. Had a problem with one memory stick so as a temporary measure replaced it with a spare 256mb stick. This still runs just fine in DDR mode with virtually no performance loss, ie 1 x 512mb, 1 x 256mb. I thought both sticks had to be the same size, but it seems not.

Do you know how long ive been trying to explain this to people ?
Everyone seems to think otherwise because of the way its marketed

Mixing DDR ram brands

[Platinum]

TBH if you dont plan to overclock your wasting your money on Corsair, get something cheaper like twinmos or even Corsair value select, the extra speed the lower latencys will give you is negligable.
The high spec stuff like Corsair will allow you to overclock higher than standard ram and possably get tighter timings at higher speeds.
If you want my advice, save moeny on the ram and get a gig of standard stuff and spend the rest on a better vid card / faster cpu / beer ect

[BenW]

or even crucial

I got some DDR400 running at DDR266 with the following timings: 2,2,2,6

[Platinum]

I think Corsair value select is on offer @ Scan atm, ive ordered some for the second rig so ill let you know if its any good

[BUFF]

Originally Posted by Ceefer

Slightly off topic but interesting to note - I was running 2 x 512mb Twinmos (m-tec chips) on my NF7. Had a problem with one memory stick so as a temporary measure replaced it with a spare 256mb stick. This still runs just fine in DDR mode with virtually no performance loss, ie 1 x 512mb, 1 x 256mb. I thought both sticks had to be the same size, but it seems not.

On an SPP board Dual Channel is only good for about an extra 3% anyway.
Now if I understand things correctly then what happens is that in your case it operates as 2x256 Dual Channel with an extra 256 tagged on.

[Agent]

Originally Posted by BUFF

On an SPP board Dual Channel is only good for about an extra 3% anyway.
Now if I understand things correctly then what happens is that in your case it operates as 2x256 Dual Channel with an extra 256 tagged on.

I think its a little more than 3%, but not by much.

The dual chan bit with unequal amounts;, 2*256 should be addressed as 128bit, with the other 256 being addressed as 64bit iirc.

[hartz]

Originally Posted by BUFF

On an SPP board Dual Channel is only good for about an extra 3% anyway.
Now if I understand things correctly then what happens is that in your case it operates as 2x256 Dual Channel with an extra 256 tagged on.

This is something I can not believe and also part of the reason why I posted some time ago asking for a special "review" / test of Dual-channel motherboards .

Throughput is always restricted to the slowest point in the "network". The CPU-to-memory bandwidth on Socket-A motherboards are restricted to the CPU front-side bus bandwidth ... but the memory controller can access this memory much faster than the CPU and this spare bandwidth is supposedly available to IO devices. When the CPU and its FSB is busy, DMA could be handling some IO concurrently with this CPU-work, even if there is a lot of memory data transfers happening; and this should give real-world multi-tasking environments a big boost!

Games seems to be written very much single-threaded (I could be wrong about this) so does not gain much from dual-channel memory bandwidth, but programs which could separate out IO-intensive work from CPU-intensive work should be able to gain a lot from the dual-channel memory architecture. In addition current "benchmark" tools are written to test one component at a time, e.g. CPU-to L1 cache, CPU-to memory bandwidth, etc. Most of these synthetic tests intentionally work in a way that does not touch any other components (minimal screen updates while testing CPU, No disk IO while testing memory bandwidth, etc.) These performance tests "intentionally" miss the benefit brought by a dual-channel memory architecture!!!!

I would like to see some MPEG encoding software which leverages this dual channel architecture by using two threads to respectively read data from the source disk into a buffer (a) and write from a buffer (b) to the target file on disk, and a third thread which performs the encoding, reading from the first buffer and storing the resulting stream in the second buffer. The OS should pretty much automatically allocate DMA work to perform the IO in the background. The encoding thread will "pause" only if the buffers over or under-run.

Such a program would make a good test of a system's overall performance, stressing IO and CPU at the same time. I'm not much of a programmer myself, but maybe someone on this thread could compile such a beast <hint-hint>

The other place where dual-channel would come in handy is with paging ... if the work set is multi-threaded and the system is memory bound, then you will have a situation where some IO needs to happen to bring pages back into memory for processes to run. Actually, this is assuming Windows' kernel can schedule a runable process while another is blocked on paging ... I don't know enough of the Windows scheduler to confirm or deny this ability, but I assume a process which is being paged in will block the whole system on a single-CPU host (And I'm a Unix bigot so I am not really interested in learning more ... Windows is just for gaming in any case :wink: )

But in theory once again DMA could take care of much of the paging work in the background. In this regard DMA is almost like an extra CPU dedicated to IO ... except that it could do multiple streams in parallel and it still requires a real CPU to initiate the buffer transfer.

I would really like to see more multi-threaded programs on our desktop systems!

[Nomadd]

I'll echo the thoughts of those suggesting 'cheaper' alternatives. Maybe look at Kingmax, too. I did a ton of research into dual channel RAM before buying mine, and in the end found that the Kingmax gave within 1% of the performance of the top-end Corsair stuff at almost half the price. I'd always used expensive Corsair stuff in the past, but skipped it this time and saved a bomb. Anandtech is a very good place to look up memory reviews and see that the 'real world' differences are very slight between all the brands out there.

Nomadd

[menthel]

No real reason to buy expensive ram unless you have more money than sense and like it to match! Not sure you even really have to have shiny heatspreaders if you have good cooling in you case. Crucial are always good, as are kingston or the corsair value gubbins.

[BUFF]

Originally Posted by hartz

This is something I can not believe and also part of the reason why I posted some time ago asking for a special "review" / test of Dual-channel motherboards .

Throughput is always restricted to the slowest point in the "network". The CPU-to-memory bandwidth on Socket-A motherboards are restricted to the CPU front-side bus bandwidth ... but the memory controller can access this memory much faster than the CPU and this spare bandwidth is supposedly available to IO devices. When the CPU and its FSB is busy, DMA could be handling some IO concurrently with this CPU-work, even if there is a lot of memory data transfers happening; and this should give real-world multi-tasking environments a big boost!

Well on an IGP board using int. gfx Dual Channel is worth about an extra 30% in performance as the gfx chip obviously gets to play with the spare bandwidth to its advantage.
But in gaming on an SPP I've never been able to see much of a difference between single channel & dual channel & in running benchmarks the differences are only slight.

[Captain Fizz]

I would steer clear of TwinMOS though. A mate has a stick of PC3200 which gives errors at 200FSB!

The Corsair Value is good. Unless you are wanting peak performance, then there is no point in getting the best.

And TwinX is really just a marketing ploy as has been stated earlier... Although Gigabyte boards can be fussy.