How much of a difference does dual channel make in AMD systems?

Ive got a gig of Kingston HyperX BH-5 PC400 and have been waiting on a decent mobo to come out for 939... but to be honest, im pretty impatient... Newegg is having a nice special on 3000+ 64s @ $163, then paired with something like a DFI UT NF3 250, I could have a pretty nice rig without waiting another month for the 939 version of the DFI board to come out....

I buy a mobo AND a cpu everytime i upgrade anyways, so dont post any "dead upgrade" mess, i've read it plenty already... this just seems to be the best value at the moment unless dual channel has some insane advantage...

So... help? Thanks in advance guys.

From what I've read all over, Dual Channel has little benefit for an AMD system, but is a great benefit for Intel. I can't remember the exact technical reasons why, so maybe someone else can elaborate ?

If you follow AMD's PR scheme, DC = 200mhz. Roughly.

S939 3000+ = 1.8ghz, with Dual Channel.
S754 3000+ = 2.0ghz, with Single Channel

the 3500+ is the odd one out. 2200mhz, which puts it at 3200+ speeds but with DC, it earns 300+ extra rating.

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2249&p=9
Take a look at how the s754 3400 does in comaparison to all the 939 chips. Its not that much in some things, but in other apps it can be almost 15%

it actually does make a substantial difference. i ran sandra's memory bandwidth test in single channel, and i barely broke 1700mb/s. in dual-channel, i got 3000mb/s. that extra 1300mb/s makes a really big difference. i am on an axp system, but i'd imagine it'd make even more of a difference with a memory controller integrated into the cpu, as is the case with the 939s.

Nah, you're getting the XPs mixed up with the A64's.

The Athlon 64's with their integrated memory controller benefit from dual channel RAM quite a bit. Most of the socket 939 Athlon 64's have less cache, yet still have a higher rating than the socket 754 versions. I can tell you which one is quicker! Try running a socket 939 processor with either 1x 512MB stick of RAM or 2x 256MB sticks. You'll find there's approx 15% performance advantage in going dual channel.


The Athlon XP's on the other hand barely benefit from a dual channel memory config with the exception of synthetic benchmarks. I don't put a lot of stock in these.....

From what I've seen the Athlon XP's get maybe a 3% performance boost, not very much really. This is probably to do with the fact that the FSB is running with the same bandwidth as the RAM. What good is the additional RAM bandwidth when its being bottlenecked by the FSB?

Why do you think I'm running only 1x stick of RAM in my system when the mainboard supports dual channel?

Fangs404 said:

it actually does make a substantial difference. i ran sandra's memory bandwidth test in single channel, and i barely broke 1700mb/s. in dual-channel, i got 3000mb/s. that extra 1300mb/s makes a really big difference. i am on an axp system, but i'd imagine it'd make even more of a difference with a memory controller integrated into the cpu, as is the case with the 939s.

Click to expand...

You might want to doublecheck your config, there's no way you should be getting 1700mb/sec, single channel. There shouldn't be almost 85% increase by going to dual channel on the AXP platform.

1700mb/sec, maybe if you were running at PC2100 speeds (133FSB) but at PC3200 speeds (200FSB), you should see roughly 2600-3000mb/sec. That score will depend on your timings but if you saw less than that, there's problems. Single/dual channel, doesn't matter.

Know Nuttin said:

You might want to doublecheck your config, there's no way you should be getting 1700mb/sec, single channel. There shouldn't be almost 85% increase by going to dual channel on the AXP platform.

1700mb/sec, maybe if you were running at PC2100 speeds (133FSB) but at PC3200 speeds (200FSB), you should see roughly 2600-3000mb/sec. That score will depend on your timings but if you saw less than that, there's problems. Single/dual channel, doesn't matter.

Click to expand...

im on an intel rig... heres my bandwith at 200mhz FSB 1:1 with pc3200LL (corsair) 2-3-2-6

single channel: 1500mb/s
dual channel: 2600mb/s

guys at abit are to lazy to check if this is supposed to be this way.. (those numbers came from memtest, sandra gives me the full 5500mb/s.. BUT sandra is synthetic..)

maybe he has the same weird stuff problem as me?

Strykar said:

3000+ 64s @ $163, then paired with something like a DFI UT NF3 250

Click to expand...

^^ Well for starters, your CPU cannot support Dual-channel RAM. It will run with multiple sticks of RAM, but it will never be in dual-channel. *Only* Socket939 Athlon64 CPUs and Pentium4s fully support dual-channel RAM. So for you its a moot point. Unless you plan on going socket 939 get a single 512MB stick of RAM.

I would say to wait and get the 939 system. I mean if the 754 was significantly cheaper than it would be worth it, but when you can get either one for roughly the same price, I see no reason not to go 939, may as well get the most performance you can for your dollar.

ongmomo said:

If u read tomshardware.com's article on 90nm 939 processors...they are VERY overclockable...probably EVEN MORE SO...on a DFI. the 3000+ 939 90nm EASILY o/c past FX-53 speeds.

Click to expand...

#1 The article is from Anandtech, not THG
#2 DFI has no plans to make an Nforce motherboard on socket 939 with an AGP slot. So unless you plan on doing your videocard as well you are limited to Giga-byte and MSI.

Overclocking with any AMD64 is a moot point. Almost ALL A64's are capable of reaching 2.4GHz-2.5GHz if done correctly. It matters not which one you pick. The really "great" ones reach 2.6GHz on air. If you want higher you need water or extreme phase change. Even with water, all that does is make 2.5GHz-2.6GHz more reliable with "great lucky" chips reaching 2.7Ghz to 2.8GHz. Phase changing adds in another bump of 100 MHz. The FX line of chips are hand picked to be great and lucky chips, hence why they can squeeze an extra 100MHz out over anything else when it comes to overclocking. Personally, that extra 100MHz isn't worth the premium.

Now why does this hold true for almost every single A64 chip out there? Because of the MEMORY CONTROLLER. That is the one part of the processor that does NOT overclock all that well.

Also, don't believe the PR rating system from AMD, it's quite flawed.

Look, dual channel IS great in some instances. If you plan to turn your computer into a Sandra Benchmark machine and you plan on running the Memory bandwith test 24/7 then yah, that dual channel will make you system look like a champ. Err.. now welcome to the real world. When testing in most applications, the difference is almost moot. Testing many games the difference is usually around 5% or less. Ohh big whuup. I can get 105 versus 100 fps. OH MY GAWD, that is going to allow me to be the fr4g m4st3r 0v3r y0r 4sS.

Now trust me, I'll all about performance and eeking the most out of a system, but I'm a realist as well. Seriously, until a good board like the DFI one comes out for the s939 line I wouldn't even look twice at it. The dual channel is NOT enough of a reason to draw me over there. Right now I can run my chip OVER 2.6 on AIR at UNDER 40C load temps on my DFI board using a DTR processor. Any decent chance of me doing that with a 939 setup without buying an FX-53 processor? No. So will my system at 2.6GHz and no dual channel be equal or even faster in performance then a 939 proc and mobo overclocked to 2.4GHz if you are lucky with dual channel? You betcha.

let's review
2.6GHz with no dual channel >= 2.4GHz with dual channel on a crappy motherboard.

Not to mention, mine cost less in price to boot. If you can hold out for the DFI board or something else that kicks butt for the s939 line then I advise so. In a month the prices for the procs might come down some as well. If you HAVE to upgrade now, then don't even bother thinking about the s939 line. It's not worth it.

You make a lot of sense Doc. I'm actually waiting for the following on my next machine, hopefully Q1 '05:

O Wu (DFI) or another awesome Nforce 4 SLI mobo
939 DTR or Mobile
Price break on SLI certified 6800GT

In the meatime I'll have to limp along with my current set-up.

yep...doc is totally right..

Look at the Sempron 3100 review on Anandtech. Rembmer semprion is socket 754.

Overclocked to 2500 mhz, it beats even the intel 3.2EE. It is just behind FX 53. Like 5 FPS diffeerence....THE PRICE difference? U can calculate that u'reself.

On ALL the benchmarks, it whoops P4 AND AMD64 ***....until memory benchmarks....

Ok, just like Docgiggs says...r u gonna run mem benchmarks 24/7 for the fun of it?

Imo, Sempron 3100/3200 might even be a better choice, letting u get a better graphics card....but the wait for 939? no, not worth it.

Sentential said:

^^ Well for starters, your CPU cannot support Dual-channel RAM. It will run with multiple sticks of RAM, but it will never be in dual-channel. *Only* Socket939 Athlon64 CPUs and Pentium4s fully support dual-channel RAM. So for you its a moot point. Unless you plan on going socket 939 get a single 512MB stick of RAM.

Click to expand...

Now thats not really true. Whether a system can run dual channel is really a function of the memory controller in the chipset. With Pentium 4's and Athlon XP's this has nothing what-so-ever to do with the processor architecture. The processor architecture does however influece how much benefit it gives.

A Pentium 4 paired with a chipset that only has a single channel memory controller will only sive you single channel access with 2x sticks of RAM.

A Athlon XP paired with the regular run of the mill nForce2 chipset will only give you single channel memory access with 2x sticks of RAM.

A Pentium 4 paired with a current generation chipset with a dual channel memory controller will give you double channel memory access with 2x sticks of RAM. Exactly the same is true of the Athlon XP with a nForce2 Ultra.

So when it comes down to it, dual channel memory support with the Pentium 4 and Athlon XP really depends on which mainboard you decide to go with. Don't expect the same performance boost with an Athlon XP as the total bandwidth of the the FSB is only as great (max) as a single channel of PC3200 DDR SDRAM. Architectural differences also affect performance here.

With the Athlon 64 the memory controller is no longer in the chipset, but integrated into the CPU, cutting down on memory access latencies. Socket 754 processors have a single channel memory controller and Socket 939 processors have a dual channel memory controller.

That is why I don't agree with your above statement at all. If you're arguing that a Athlon XP doesn't really support a dual channel config because it doesn't have an integrated memory controller, that doesn't hold water either. Neither does the Pentium 4!


The main reason why Pentium 4's benefit so much from a dual channel memory config is exactly the same reason why they perform better with a larger level 2 cache than Athlon XP's and Athlon 64's.

Due to the extremely long pipeline of the P4, there is a much greater chance for branch misprediction, and hence, often data stored in the L2 cache is not required and must be pulled from the main system RAM.

A Dual channel memory config allows the P4 to continue to operate more smoothly as it doesn't have to wait as often for data to be fed to it. This is also the reason why larger level 2 caches benefit the P4 so much and noth Athlon XP/64's. A detailed explanation of the Pentium 4 architecture can be found over at Ace's Hardware. I'd recommend you check it out here: http://www.aceshardware.com/list.jsp?id=4

True, when I had 2 stix of PC2100 in my Nforce2ultra board it WAS running dual channel!!!!

Oops, sorry about my technical rant above. I didn't read your comment in context. You were ttalking soley about the Athlon 64's.

This technically still isn't correct. With the Athlon 64's AMD has included the abillity for chipset makers to implement memory controllers in their chipset and disable the integrated memory controller in the CPU.

This has been the case ever sincece the Athlon 64 and Operon were released!

Thats why technically, the DDR2 RAM argument with socket compatabillity is BS. However, you lose a large part of the Athlon 64's performance capabillity when you go with a non-integrated memory controller. So technically, a Socket 754 processor IS dual channel capable, but only if the chipset supports it.

Fangs404 said:

it actually does make a substantial difference. i ran sandra's memory bandwidth test in single channel, and i barely broke 1700mb/s. in dual-channel, i got 3000mb/s. that extra 1300mb/s makes a really big difference. i am on an axp system, but i'd imagine it'd make even more of a difference with a memory controller integrated into the cpu, as is the case with the 939s.

Click to expand...

I have a NF7-S also and thier is no way in fcking hell you got a 1500 point increase with duel channle on a AXP system, the diff is more like 200 points in sandra. I think youre confusing youreself here.


Duel channle for the A64 is like Intel's now where it actuall doubles the score. so if you get 3000 you'll get like 6000. Also what Doc gibes is saying a bout the cpu ocing to 2.5ghz and water only making it more reliable and just cooler while gaining no increase in overclock, is what i have been saying forever now way before he ever got an A64 computer, just no one belived me. Dropping the temps 30c will not in most cases increase youre overclock one bit. I just went watercooling cuss i already had a watercooler on my axp and it was going to waist to i hooked it up to my A64 system.

Hitechjb1 still claims that a 1-2c and a 30c drop in temp will make youre overclock alot higher, he writes alot of fancy stuff all the time but still seems to refuse that the Athlon64's dont gain anything like the AXP'S did when dropping temps

If Intel went with a integrated memory controller it would really scr*w AMD. Intel's architecture would benefit from it more.

Amusing really. AMD was first out with dual channel: it benefitted Intels P4 architecture more.

AMD is once more first out with an integrated memory controller: it will once more benefit Intel's P4 architecture more.

I suspect Intel is waiting or their FB-DIMM memory module technology to be implemented before going with an integrated memory controller. Likewise, I suspect AMD is also waiting for the FB-DIMM instead of going to the FB-DIMM as this will mean the end of all socket changes instead of changing whenever memory technology changes.

It would mean that regardless of the RAM type being used, the memory controller would not need to be modified and neither would the mainboard. It would not matter whether the RAM type is DDR, DDR2, Rambus XDR or QBM RAM, the controller only has to talk to the FB-DIMM.

The FB-DIMM itself talks to whatever type of chips are used on the module.

Check out the following two links for details:

http://www.theinquirer.net/?article=15214
http://tech-report.com/onearticle.x/6563

This will definitely help out the better performing but currently unsupported memory types such as QBM (6.4Gbit/module) or XDR RAM(12Gbit/module).

Funnily enough, this technology combined with faster memory types might actually benefit AMD more than Intel as the Athlon 64 currently uses a 2GHz HyperTransport Point to Point interconnect offering far more bandwidth, and fewer bottlenecks than Intel's 1066MHz shared system bus. I guess only time will tell.

oh, crap. i don't know what i was thinking earlier today. i meant to type 2700mb/s to 3000mb/s instead of 1700mb/s. i need to lay off the crack.... still, that extra 300mb/s does seem to make a good amount of difference in benches like 3dmark and such.