Now on PCs, normally you need matched sticks, but I've heard of people with Sonoma notebooks that get dual channel with sticks that are different sizes. Anyone personally try this and run some benchmarks, like Sisoft Sandra to determine if you're actually getting the bandwidth? Or just use CPU-Z to see if you're running in dual channel or not.
Sonoma users, what is required to activate dual channel?
- Thread starter Fox5
- Start date
I get around 300mbps less with dual channel and mixed ram sizes (1gb+256) than with to matched sticks. Sandra shows that with matched ram I get 3300mb/s
How does mixed ram sizes dual channel compare to single channel though? And dual channel actually is enabled with the mixed sticks?(the cpu's bandwidth probably won't change at all with dual channel, but theoretically the GMA900 could make use of it)
That's interesting, I wouldn't think "matched" sticks would make a difference. AFAIK, as long as DC works, it works as well as possible
Anyway: mixing RAM sizes SHOULD deactivate dual channel mode. For dual channel to function (in theory), you must have two sticks of equal size (and usually same make/model). I'll test this out later for you on a T43
Thanks in advance for testing it.
Anyhow, the Nforce2 dual channel worked regardless of size, it was very hard to make dual channel on nforce2 cease functioning.(I did notice that measured bandwidth dropped a few hundred megabytes by going from 2 sticks to 3 though, perhaps something similar to the Athlon's dropping to 2T command time happens?)
Since Sonoma's cpu really isn't benefitted by dual channel and it only serves to give the platform as a whole more bandwidth, it would make sense to do a more nforce2 style dual channel.
Anyhow, the Nforce2 dual channel worked regardless of size, it was very hard to make dual channel on nforce2 cease functioning.(I did notice that measured bandwidth dropped a few hundred megabytes by going from 2 sticks to 3 though, perhaps something similar to the Athlon's dropping to 2T command time happens?)
Since Sonoma's cpu really isn't benefitted by dual channel and it only serves to give the platform as a whole more bandwidth, it would make sense to do a more nforce2 style dual channel.
By using non matched sticks it lowers their speed to 400mhz instead of 533 - but it doesnt make alot of difference. It will of course give a nice boost if your gpu supports Turbo cache / hyper memory. My 6600 go apparently supports TC so when I get another gig of ram it should give a minor boost. It arrives tomorrow so I will let you know.
No I have a 6600 go standard (300/600 core/mem) but it still uses TC afaik it just doesnt need it as much as. If you look into system details it gives me something like 6600go 128mb + XXXmb system memory when looking at just the GPU.
I tested it out for ya
1x512MB DDR2 533 + 1x256MB DDR2 533 = 768MB DDR2 533 in Dual Channel Mode according to CPU-Z
I ran a quick Sandra mem bandwidth test and it came out the same as the other dual channel bandwidth tests.
Cool thanks, so looks like I don't need to worry about mismatched memory for dual channel.
I know it was a stated feature that something about stricter standards on DDR2 or something would allow laxer dual channel setups, but I didn't know if it was just dimms with different chips or anything goes. Appears to be anything goes, I can now safely buy a 1GB stick and not worry about getting less performance, thank you.
I know it was a stated feature that something about stricter standards on DDR2 or something would allow laxer dual channel setups, but I didn't know if it was just dimms with different chips or anything goes. Appears to be anything goes, I can now safely buy a 1GB stick and not worry about getting less performance, thank you.
BTW, the sticks I used in the T43 were 1x512MB IBM original (made by Samsung I think) and the other was 1x256MB Crucial (picked using their matching system to ensure compatibility).
Wait, IBM has a matching system?
I was planning on just opening up my IBM(assuming it ever gets here), checking out what ram chips it used, and then buying a DDR2 stick with chips from the same company.
I was planning on just opening up my IBM(assuming it ever gets here), checking out what ram chips it used, and then buying a DDR2 stick with chips from the same company.
I'm using an ASUS W3V with 512 standard. I added in 1GB a few weeks ago, and according to both the BIOS and CPU-Z, I have dual channel active (confused me for a second). I'll try to get Sandra to test out my bandwidth later today
http://forum.notebookreview.com/attachment.php?attachmentid=101
http://forum.notebookreview.com/attachment.php?attachmentid=102
http://forum.notebookreview.com/attachment.php?attachmentid=101
http://forum.notebookreview.com/attachment.php?attachmentid=102
That really doesn't make any sense that unmatched sticks would perform the same, and it would be pretty hard to prove since you can't get a 384+384=768MB setup.
Dual channel works by accessing memory in parallel. If you have 256MB + 512MB, it can only access memory in parallel for the first 512MB. Beyond that, it is only accessing memory from the second stick.
This should be much more obvious in configurations like 256MB + 1024MB where less than half of your memory can be accessed in parallel. It's like having 512MB of dual-channel memory and 768MB of single-channel memory.
It makes sense when you keep in mind that it is just testing memory bandwidth, not performance. If you had some type of program that would read in GB's of data through the RAM, then that would be ideal! Most understand that Sandra doesn't provide real world performance #'s, but instead more theoretical performance #'.s
Wouldn't it still be superior to have 1GB + 512MB over 512MB + 512MB though? You get the same bandwidth for that first gigabyte, then the last 512MB is extra ram at a slower speed.
Certainly. It'd be hard to find a situation where you wouldn't be better off getting a bigger stick, even if it's unmatched.
I suppose, though, when you're talking about laptops, if you have a single 512MB stick, you could argue you might be better off just getting another 512MB instead of a 1024MB, because:
1) cheaper, of course
2) takes less time to hibernate/resume
3) matched for full dual-channel bandwidth
4) uses a little less power
Still, the deciding factor really isn't going to be whether or not it's matched, just whether or not you think the extra capacity is worth the price, so I suppose it's pretty much irrelevant.
It would be interesting to see how this affects the IGP, though. Onboard graphics are one of the only places I've seen much performance difference between single/dual-channel. If they're not smart enough to have the shared memory always come out of the dual-channel part, unmatched memory could hurt video performance.
Working on that article now
Altho its a bit late to add in the mixed size dual channel part, sorry guys Upto 2gb now and no real difference reported by sandra, but ram speed has gone upto 266 instead of 400 with 1gb + 256gb. I am getting 3000mbps.
fbrdphreak If you need any benchmarks I am sure I can oblige.
fbrdphreak If you need any benchmarks I am sure I can oblige.
yeah sorry meant from 200-266, i think my 256 stick was higher latency so it changed them to 200 (400) despite being both 266 (533) back to matched and alls well
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 14K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 9K
-
Discussion Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)- Started by Vattila
- Replies: 13K
-
Facebook
Twitter