Sign in with
Sign up | Sign in
Your question

Need Memory Advice for FSB of 1,333

Last response: in Memory
Share
October 22, 2007 1:02:49 AM

I'm building a computer and need advice on what memory to get. Systems specs are:
CPU: Core 2 Quad Q9550 with a FSB of 1,333MHz (due in 1/08)
Motherboard: GA-X38-DQ6; supports 1,066MHz memory and 1,333MHz FSB
RAM: 4 Gb of DDR2

I will NOT be overclocking. I've heard that it is preferable to run memory synchronously with FSB, especially on the Core 2. In this case that means DDR2 667MHz if I'm running dual channel, I believe. My questions are:

1. Should I go DDR2 1,066MHz given its speed differential over DDR2 667MHz, even though it means running the memory asynchronously. And if so, why?

2. For 4Gb, the obvious choice would be 2x2Gb sticks. However, latency on the 2Gb sticks is higher (CAS 4/5) vs. the 1Gb sticks (CAS 3/4). But I've also heard that, all other things equal, two sticks generally perform better than four sticks, which would partially offset the lower latency of the smaller sticks. Any advice?

3. What is the difference between PC2 5400 and PC2 5300 memory?

More about : memory advice fsb 333

October 22, 2007 12:55:40 PM

1333 / 4 = 333.25 * 2 = 666.5

At stock, a 1333fsb cpu + 667mhz ddr2 ram are a good match. What you heard is correct. Because, see, you can't unlink cpu & ram on X38. They'll run at a ratio. Whatever it is. It's by design. If you want to unlink them, go ahead & maybe you'll win a million dollar contract with Intel. *joke*

1. No. Don't. Get 667. Be done with it.

2. Means nothing to you since you're not overclocking. Get the cheapest.

3. PC2-XXXX is the max bandwidth. i.e. PC2-5300 means 5300 MB/s. It's an estimate. Not guaranteed. Sometimes it's 5300. Other times it's slower or faster. So the diff. is 100 MB/s. Not worth any extra money at stock.

http://en.wikipedia.org/wiki/DDR2_SDRAM
October 22, 2007 2:07:28 PM

As I understand, the large cache sizes on the current gen of Intel chips make performance not depend heavily on RAM speed, and in general, there is some small boost by running synchronously. Thus, I suggest DDR667.
Related resources
October 22, 2007 2:18:30 PM

Seems like sound advice and I'll probably go that route. But to gain better understanding (and be a devil's advocate), if DDR2 1,066MHz is faster, wouldn't that provide better peformance. Same with the lower latency.
October 22, 2007 10:25:59 PM

Yes, it would def. But you sound like you don't want to even touch the bios. You can keep the cpu at stock & run the ram near its specs. i.e.

1333 / 4 = 333.25 * 4 = 1333
1333 / 4 = 333.25 * 3 = 999.75

Just change the cpu/ram ratio to x4, known as system memory multiplier on Gigabyte boards. The ram would run at 1000mhz+, faster than 667mhz. See middle of the photo:

http://techgage.com/viewimg/?img=/reviews/gigabyte/x38_...

The thing is you wouldn't notice a speed jump unless you do benchmarks or any other intensive time-consuming tasks as such databases, calculations, multimedia encoding/decoding, etc. You would notice the less time it takes to do such tasks.

You wouldn't see anything different if you browse the web, read a document, etc. To speed up those kinds of tasks, get a faster cpu or like I do, overclock it.
October 23, 2007 12:22:53 AM

Id get 800mhz as there are generally more of these available at ruffly same price as the 667 stuff.
I would also go with 2x2gb for upgrade options later on.

All advise here is pretty sound.
October 23, 2007 3:21:44 AM

So, if I get the 800MHz sticks, how do I run them at 667MHz if I want to synchronous?
October 23, 2007 5:15:43 AM

You would just go into your BIOS and set them to run at DDR667. Depending on the motherboard, the setting maybe labelled differently (that's the way it is on my P5W DH Deluxe).

As for getting higher binned RAM to run with your system, there are two reasons to do so. Both are in reference to running synchronously.

1. First if you wish to overclock (the most obvious reason). If say you get DDR800, you can run the system clock up to 400 without overclocking your RAM. That is of course if your CPU can take, though you could lower your multiplier to accomodate the higher system clock and run your CPU at or near it's normal operating clock frequency.

2. The other less obvious reason, is that you can possibly tighten your RAM timings if you are running them at a slower rate than they are rated for. Lets use the same example (DDR800 running as DDR667). Lets say you get this RAM with timings of 5-5-5-15, you may find that you can run it at 4-5-4-12, or even (if your lucky) 4-4-4-10.

Now with these timings, it won't actually make a huge real world difference as the large cache on the C2D's/C2Q's hide some of the benefits to lower latency. You will see it on some select synthetic benchmarks, but other than that not a huge realworld benefit. Even memory bandwidth in the first example won't have huge impacts on system performance. You will see a big impact though in the first example if you are in fact OC'ing your CPU.

As far as running your RAM asynchronously, you may see some small affect on bandwidth benchmarks, but real world benefit.. not sooo much. So though in your case (not overclocking) you won't see any benefit, the fact that DDR800 is nearly the same price as DDR667, I would get DDR800. Even if you don't use it right now, your next CPU might have an FSB of 1600. This gives you the ability to carry that RAM with you on your next upgrade. One caveat to that is that by the time desktop CPU's come out with FSB's of 1600, the motherboards that have chipsets that officially support them will likely to be DDR3.

October 24, 2007 4:13:12 AM

techgeek said:
As far as running your RAM asynchronously, you may see some small affect on bandwidth benchmarks, but real world benefit.. not sooo much. So though in your case (not overclocking) you won't see any benefit, the fact that DDR800 is nearly the same price as DDR667, I would get DDR800. Even if you don't use it right now, your next CPU might have an FSB of 1600. This gives you the ability to carry that RAM with you on your next upgrade. One caveat to that is that by the time desktop CPU's come out with FSB's of 1600, the motherboards that have chipsets that officially support them will likely to be DDR3.


Thanks for the info. Just what I wanted to know. It seems that FSB's are lagging. That is to say, FSB 1333 is pretty fast, but they run memory snynchronously at only DDR 667 (even though DDR 1,066 is all over the place). Is this state of imbalance typically the case?

Also, I hate to sound like a bonehead, but what relationship is defined by the multiplier?
October 24, 2007 5:55:51 AM

DDR2 is double data rate... therefore 667 is multiplied by 2 giving you 1334 hence a 1333FSB CPU uses 667 DDR2 at a ratio of 1:1. Then if you wanted to clock higher you go for the 800mhz, although the 1066 is good for decreasing CAS timings running at under spec.
October 24, 2007 6:51:26 AM

chookman said:
DDR2 is double data rate... therefore 667 is multiplied by 2 giving you 1334 hence a 1333FSB CPU uses 667 DDR2 at a ratio of 1:1. Then if you wanted to clock higher you go for the 800mhz, although the 1066 is good for decreasing CAS timings running at under spec.


No, sometimes I see a higher multiplier, like 12x. What is 12x what?
October 24, 2007 7:42:48 AM

chookman said:
DDR2 is double data rate... therefore 667 is multiplied by 2 giving you 1334 hence a 1333FSB CPU uses 667 DDR2 at a ratio of 1:1. Then if you wanted to clock higher you go for the 800mhz, although the 1066 is good for decreasing CAS timings running at under spec.
That is incorrect. DDR2 667 runs at 666 333 x 2 = 666 and DDR2 800 runs at 800 400 x 2 = 800. The RAM runs at half the speed of the FSB, go figure. The 1:1 is determined from the base clock 333 is 666, 400 is 800, 533 is 1066 etc. I spent a lot of time pinning this down, I always saw it, I just ignored it.

@OP, I would get DDR2 800. That is an OCing chip. I know you aren't going to OC, but you will. If you OC they will come. You will be able to get a nice OC and still stay 1:1. The 1066 isn't worth the money because performance isn't really memory intensive. The only problem is out running the speed of the RAM which DDR2 800 will handle, unless you want more than a moderate OC of 3.4G/1600FSB.
October 25, 2007 2:10:26 AM

jaydub868 said:
No, sometimes I see a higher multiplier, like 12x. What is 12x what?



This will possibly be referring the the multiplier on the CPU to gain the other Mhz rating. Q6600 runs at 1066FSB (quad pumped 266mhz) 9multiplier to give 2.4Ghz (266x9).

Quote:
That is incorrect. DDR2 667 runs at 666 333 x 2 = 666 and DDR2 800 runs at 800 400 x 2 = 800. The RAM runs at half the speed of the FSB, go figure. The 1:1 is determined from the base clock 333 is 666, 400 is 800, 533 is 1066 etc. I spent a lot of time pinning this down, I always saw it, I just ignored it.


I might be wrong in the way of double data rate, but it would then be brought back up with Dual Channel utilisation. Thats the way i know it anyways... If you have a better explaination please let me know
October 26, 2007 7:49:46 PM

chookman said:
I might be wrong in the way of double data rate, but it would then be brought back up with Dual Channel utilisation. Thats the way i know it anyways... If you have a better explaination please let me know
I thought that DDR2 doubled it to the FSB as well. I was wrong, it only doubles to an "effective" speed of half the FSB assuming 1:1. You can't really use the dual channel argument because what that does in theory is widen the bus to From the single channel 64 bit to dual channel 128 bit. I does increase the throughput but not in the same way that sending data on the rising and falling edge does. Dual channel is a wider bus and DDR is an apparent doubling of the clock speed. I had read some articles on the dual channel theory about how it would almost double throughput, net of the additional overhead. So I ran some benches and was dismayed to find a small amount of actual increase in the throughput. I was going to post a thread about this, but I was concerned that I didn't tie down the testing methodology. I don't think this had any affect, but I wanted to be sure before I posted anything. I haven't had time to go back and run the benches again to be sure everything was held constant. I'll get to it sooner or later, unless someone beats me to it. Then I won't have to do it at all. Anyone?
a c 97 } Memory
October 26, 2007 8:40:19 PM

jaydub868 said:
Seems like sound advice and I'll probably go that route. But to gain better understanding (and be a devil's advocate), if DDR2 1,066MHz is faster, wouldn't that provide better peformance. Same with the lower latency.



jaydub; you are absolutely correct, faster memory and lower latencies perform better...BUT, by how much?

The various C2D application benchmarks I have seen show minimal performance gains in real(vs. synthetic benchmarks) applications, on the order of <2%. If you are going for maximum overclocks, then that is another matter. I think it is better to spend your performance dollars elsewhere.
October 26, 2007 9:05:38 PM

Zorg said:
I thought that DDR2 doubled it to the FSB as well. I was wrong, it only doubles to an "effective" speed of half the FSB assuming 1:1. You can't really use the dual channel argument because what that does in theory is widen the bus to From the single channel 64 bit to dual channel 128 bit. I does increase the throughput but not in the same way that sending data on the rising and falling edge does. Dual channel is a wider bus and DDR is an apparent doubling of the clock speed. I had read some articles on the dual channel theory about how it would almost double throughput, net of the additional overhead. So I ran some benches and was dismayed to find a small amount of actual increase in the throughput. I was going to post a thread about this, but I was concerned that I didn't tie down the testing methodology. I don't think this had any affect, but I wanted to be sure before I posted anything. I haven't had time to go back and run the benches again to be sure everything was held constant. I'll get to it sooner or later, unless someone beats me to it. Then I won't have to do it at all. Anyone?


Ive done a little more research and seems i explained this wrong...Q6600 is at 1066FSB "Quad pumped" this bring clock speed to 266mhz. A DDR2 part at 533mhz has a memory clock of 133mhz, however, as its DDR"2" this is "double pumped" too 266mhz.

Quote from wiki

"The key difference between DDR and DDR2 is that in DDR2 the bus is clocked at twice the speed of the memory cells, so four words of data can be transferred per memory cell cycle. Thus, without speeding up the memory cells themselves, DDR2 can effectively operate at twice the bus speed of DDR."

Making the I/O clock on the memory the same as the CPU clock therefore; 1:1 with Q6600 is DDR2-533.

http://en.wikipedia.org/wiki/DDR2

Sound better?
October 26, 2007 9:34:17 PM

chookman said:
Ive done a little more research and seems i explained this wrong...Q6600 is at 1066FSB "Quad pumped" this bring clock speed to 266mhz. A DDR2 part at 533mhz has a memory clock of 133mhz, however, as its DDR"2" this is "double pumped" too 266mhz.

Quote from wiki

"The key difference between DDR and DDR2 is that in DDR2 the bus is clocked at twice the speed of the memory cells, so four words of data can be transferred per memory cell cycle. Thus, without speeding up the memory cells themselves, DDR2 can effectively operate at twice the bus speed of DDR."

Making the I/O clock on the memory the same as the CPU clock therefore; 1:1 with Q6600 is DDR2-533.

http://en.wikipedia.org/wiki/DDR2

Sound better?
Close but really
Quote:
The key to DDR-II bandwidth is the core is running at 1/2 clock frequency of the I/O buffers
This is how they increase the ram speed over that capable of DDR, the underlying fact still stands that the DDR2 RAM is still running at half of the FSB. The FSB is "quad pumped" and the RAM is "double pumped". Believe me I know that this is not intuitive, it really is counterintuitive, but that's the way it is. This article might help. HEXUS.net - Guide :: DDR-II - How It Works
October 27, 2007 4:27:05 AM

Right that helps a little... now for the big question given a Q6600 at 1066 FSB what do we need to run a 1:1 ratio FSB:RAM you are saying we should still be going against mathematical reasoning and a 1:1 in this instance would be 1066:533?
October 27, 2007 5:08:05 AM

Yup. But if you want to OC to 3G/1333 then 667, which is a done deal with the P35. I like the DDR2 800, In case I get cocky with the clock, although I doubt it, 3G works for me. A lot of people are saying go for the higher clocked RAM to reduce the timings at the lower speeds. I don't know how effective that is with the Core2s. AMD apparently really benefits from the reduced timings though, obviously.

Edit: I'm running my Q6600 @ 3G/1333 w/ 4:5 DDR2 800 @ 800. I don't think that I'm necessarily gaining any major performance increase, although that is based on what I have read. I can get it, so I use it. Gigabyte sets it automatically by default, albeit at looser timings. I suspect the looser timings are for the ridiculous automatic OC that they run in BIOS by default.
!