DDR2-1066 performance

[Starrbuck]

Hiya,

I just built a new gaming system based around the EVGA nForce 750i SLI FTW and an E8500, and I am currently testing two different sets of memory. The original memory I got is OCZ OCZ2N800SR4GK (DDR2-800, 5-4-4-15, 2.1v). The board is capable of running DDR2-1066, so I got a Mushkin 996599 set (DDR2-1066, 5-5-5-15, 2.1v).

With the Mushkin memory configured at the proper manual settings in the BIOS, it is barely faster than the OCZ, using SiSoft Sandra's memory bandwidth benchmark. The Mushkin was around $40 more than the OCZ, and I am just not seeing the benefits at this point.

Is there anything I can do to improve the performance? Am I expecting too much of a jump from 800 to 1066?

Thanks in advance for any advice.

 

[jimbo6846]

How much is "barely"? You might be expecting too much of a difference between 800 and 1066. Also, as you posted, the OCZ had slightly better timings than what you have the Mushkin set at, currently. That will eliminate some of the difference. If you set the Mushkin to 5-4-4-15, like the OCZ was, you'd see slightly better numbers in Sandra.

Honestly, you can probably tighten those Mushkin timings pretty nicely. See if you can change them to 4-4-4-12. If not, start moving the numbers back up, one at a time.

 

[Starrbuck]

Hey, thanks for the reply.

With the 800, I was getting just over 6 GB/s, and when I put the 1066 in at stock speeds, I only got 6.5 GB/s. I was expecting a larger improvement.

I just overclocked the CPU to 440 (4180 effective) and overclocked the memory to 550 (1100 effective) and Sandra says at these speeds I am getting around 7.80 GB/s. I'm going to push the memory a little harder and see how high I can get it to go without blue screening or not posting.

 

[Starrbuck]

Yeah, I can't seem to push the memory past 1100. At 1120 it won't boot all the way into Vista. I did get just over 8 GB/s on the final memory benchmark...

I don't really want to spend hours and hours tweaking the overclocking settings. After about an hour. I seem to have it stable at 1100 on the memory and 4180 on the CPU. Temps are under 50 on the CPU and the memory does not appear very hot to the touch.

 

[Nenu]

Relax the memory timings if you want to clock it higher.
If the memory isnt too warm (to the touch), you might get a higher clock by upping the voltage.

You wont see much (if any) performance benefit from higher clocked ram on a C2D system.
Its best to choose the memory you need that will run at the FSB you intend to use, running the memory 1:1 with the FSB (ie 500FSB needs DDR2 1000)

 

[KILLER_K]

You wont see much (if any) performance benefit from higher clocked ram on a C2D system.
Its best to choose the memory you need that will run at the FSB you intend to use, running the memory 1:1 with the FSB (ie 500FSB needs DDR2 1000)

I really hope you was joking when you said that????

Try using Transaction booster to enable and set it to boost 2 and see what your readings are then. As you should get a nice performance boost with that and it may do boost 3 also just give it a shot and see.

Here are some examples using Corsair Dominator 1066 with and without boost.
Q6600 @ 3.6 with boost 2

Q6600 @ 3.6 boost 3

Q6600 stock no boost

CellShock PC6400 @ 1200mhz Q6600 @ 3.6 using a 8x multi with boost 2

Hopefully the pictures aren't too big for you, if let me know i will resize them.

Good Day

 

[Starrbuck]

I don't have Transaction Booster on this EVGA board.

Something that's confusing me... With the memory ratio set to 1:1, with the FSB at 1740 (435), the memory should be 870, but it's 1392. So I currently have the memory set manually to 1100, which seems to set it at 1087.5 automatically.

 

[Nenu]

I really hope you was joking when you said that????

Try using Transaction booster to enable and set it to boost 2 and see what your readings are then. As you should get a nice performance boost with that and it may do boost 3 also just give it a shot and see.

Here are some examples using Corsair Dominator 1066 with and without boost.
Q6600 @ 3.6 with boost 2

Q6600 @ 3.6 boost 3

Q6600 stock no boost

CellShock PC6400 @ 1200mhz Q6600 @ 3.6 using a 8x multi with boost 2

Hopefully the pictures aren't too big for you, if let me know i will resize them.

Good Day

I'm not sure what you are saying.
The memory bandwidth does increase with memory speed as you have demonstrated in Everest but you havent shown any real world benefit.
Your Cellshock image is another Everest image, there is no data to suggest that it runs faster.

 

[ALL4AMD]

Hey, thanks for the reply.

With the 800, I was getting just over 6 GB/s, and when I put the 1066 in at stock speeds, I only got 6.5 GB/s. I was expecting a larger improvement.

I just overclocked the CPU to 440 (4180 effective) and overclocked the memory to 550 (1100 effective) and Sandra says at these speeds I am getting around 7.80 GB/s. I'm going to push the memory a little harder and see how high I can get it to go without blue screening or not posting.

Yeah, I can't seem to push the memory past 1100. At 1120 it won't boot all the way into Vista. I did get just over 8 GB/s on the final memory benchmark...

I don't really want to spend hours and hours tweaking the overclocking settings. After about an hour. I seem to have it stable at 1100 on the memory and 4180 on the CPU. Temps are under 50 on the CPU and the memory does not appear very hot to the touch.

I'm not sure what you are saying.
The memory bandwidth does increase with memory speed as you have demonstrated in Everest but you havent shown any real world benefit.
Your Cellshock image is another Everest image, there is no data to suggest that it runs faster.

6-8gb seems low to me. i get over 10GBps on my phenom. could be FSB bottleneck?

 

[Dangman]

I really hope you was joking when you said that????

Try using Transaction booster to enable and set it to boost 2 and see what your readings are then. As you should get a nice performance boost with that and it may do boost 3 also just give it a shot and see.

Uh huh, and the where's the real world benchmarks? That's just a bunch of synthetic benchmarks. As shown in this review:
http://www.xbitlabs.com/articles/mainboards/display/asus-p5k-deluxe-overclock_4.html

The Transaction booster only provided at most a 1% increase in real world apps.

And Nenu is right about no real difference with higher clocked RAM with C2D based systems:
http://www.legionhardware.com/document.php?id=628&p=0

 

[jimbo6846]

Starrbuck, ignore the arguing going on in this thread. If you want to increase your memory performance, you'll have to lower your timings and performance level. This can be done in your BIOS, in the memory section, or with a program called Memset.

If you decide not to worry about it, that's fine.

 

[Starrbuck]

It's not "arguing," it's a "spirited discussion."

I put the DDR2-800 back in and it's doing as well as the 1066 was doing, clocked lower. The nice thing is that I have the CPU and memory both overclocked at 440, so I like that they can be set at matching rates. I'm performing stability tests now.

Thank you all for your help and ideas.

 

[KILLER_K]

SNIP!

This is off your page link you posted....
Enabling DRAM Static Read Control increases the memory controller performance; disabling – decreases.

Plus if you noticed they didn't even use boost level 1,2 nor 3 for any test they did. As all they did was turn on to enable and disable. So really their test wasn't even completed.

Laff, then they shouldn't even had made ddr-2 or ddr-3 and even for that matter increased the fsb of the CD2 from the 1066mhz to the 1,333fsb then. Since intel doesn't use the onboard memory controller the added fsb is always a welcome in the CD2 platform.

And yes everything is synthetic incase you didn't otice but that is all we have to get a grasp on what we are buying. I guess you are going to also say there is no need to do reviews or test on anything hardware also next?

ALL4AMD: You can't really compare AMD to INTEL like that since AMD uses the onboard memory controller. As the AMD will communicate faster with that feature and get better benchmarks.

jimbo6846 : Who is arguing? I'm giving him examples for him to go on. Plus providing pictures of my experience with ram settings. If anything it is helpful...

starrbuck: Glad you got it sorted and you are happy.

Good Day

 

[Starrbuck]

So true, KILLER_K. I just switched from an AMD platform to this new Intel E8500 and after having only AMD systems for the last several years, as I was building this one I realized how the two different memory controllers work.

Synthetic benchmarks are all we have to quickly make a determination on what's better. A real world benchmark I have is that FPS went up considerably while standing at the same spot with roughly the same number of characters on the screen in WoW's city of Shattrath.

 

[jimbo6846]

KILLER, I didn't say you weren't helping. I just see this argument so often, it gets old. I don't understand why anyone would advise against OC'ing as simple as boosting RAM speed. Doesn't make any sense...

Star, have you tried boosting that RAM's speed to 1066, and see how much more improvement there is over when it's at 800?

 

[Starrbuck]

Not yet. How loose would you think I might have to go on the timings if I up the clock to 1066?

At this point, I'm just shocked that the OCZ is running as good or better than the Mushkin. I never gave that much credit to OCZ.

 

[Nenu]

KILLER, I didn't say you weren't helping. I just see this argument so often, it gets old. I don't understand why anyone would advise against OC'ing as simple as boosting RAM speed. Doesn't make any sense...

Star, have you tried boosting that RAM's speed to 1066, and see how much more improvement there is over when it's at 800?

It doesnt make sense when it uses more power, increases heat, increases wear and doesnt give a usable performance benefit.
Now if you are doing it just to see how high it can go, thats a different story.

 

[Starrbuck]

Right, I'm not that kind of overclocker. I just want a modest overclock that performs well and is safe.

 

[jimbo6846]

It doesnt make sense when it uses more power,

Not if you don't increase the voltage

increases heat

If you don't increase the voltage, any increase in heat won't be noticeable

increases wear

See responses to the first two items, which cause wear.

and doesnt give a usable performance benefit.

If all you do is surf the web, then you are correct. No performance benefit there. It does make a difference in memory intensive applications, though.

 

[jimbo6846]

Not yet. How loose would you think I might have to go on the timings if I up the clock to 1066?

At this point, I'm just shocked that the OCZ is running as good or better than the Mushkin. I never gave that much credit to OCZ.

Take them up to 5-5-5-15 and just set the RAM to 1066. Make sure the RAM voltage isn't set to Auto, set it to whatever OCZ rated them for. That should be all you need to do.

 

[Starrbuck]

Cool, I may give it a try later this afternoon or tonight, and I'll post the results.

 

[Nenu]

Not if you don't increase the voltage

If you don't increase the voltage, any increase in heat won't be noticeable

See responses to the first two items, which cause wear.

If all you do is surf the web, then you are correct. No performance benefit there. It does make a difference in memory intensive applications, though.

That depends on the difference in overclock, you can still cause a significant increase in power use by raising the clocks.
As you increase the clock speed, you reduce effective internal resistance.
Power = Vsquared / R
Therefore reducing Resistance increases the Power used.

Can you name some of these memory intensive applications that will benefit us if we want to overclock our ram.

 

[KILLER_K]

KILLER, I didn't say you weren't helping. I just see this argument so often, it gets old. I don't understand why anyone would advise against OC'ing as simple as boosting RAM speed. Doesn't make any sense...

Star, have you tried boosting that RAM's speed to 1066, and see how much more improvement there is over when it's at 800?

Okay i was thinking you was attacking me about the defense i was using for 1066 ram and the boost feature. As i still haven't figured out why i was being attacked for it yet.

Overclocking does have it's problems. But the main problem is over-volting and lack of proper cooling. But you can easily fix this by buying high end heatsink and fan combo's or simply go water. As i seen more stuff people has killed not by overclocking it but too much heat and or lack of judgement...

If you read most of the users that actually killed something overclocking, something either failed, fan stop spinning or they just didn't use fair judgement. And i have yet seen someone buy a lower end CD2 and not overclock it. As that is the name of the game Best Bang For The Buck.

Good Day

 

[jimbo6846]

That depends on the difference in overclock, you can still cause a significant increase in power use by raising the clocks.
As you increase the clock speed, you reduce effective internal resistance.
Power = Vsquared / R
Therefore reducing Resistance increases the Power used.

No increase worth a second thought. The voltage you set for your RAM is the top end of what is what is applied, not the middle or bottom end. If the motherboard sends more than what it's specifically told to (taking known overvoltage into account), it would only be a small amount and only for a microsecond when the sticks just finished a particularly heavy-duty job. This is covered under the manufacturer's warranty. This is normal.

Can you name some of these memory intensive applications that will benefit us if we want to overclock our ram.

The OP already stated WoW performance improved, so there's one.

 

[jimbo6846]

Okay i was thinking you was attacking me about the defense i was using for 1066 ram and the boost feature. As i still haven't figured out why i was being attacked for it yet.

Overclocking does have it's problems. But the main problem is over-volting and lack of proper cooling. But you can easily fix this by buying high end heatsink and fan combo's or simply go water. As i seen more stuff people has killed not by overclocking it but too much heat and or lack of judgement...

If you read most of the users that actually killed something overclocking, something either failed, fan stop spinning or they just didn't use fair judgement. And i have yet seen someone buy a lower end CD2 and not overclock it. As that is the name of the game Best Bang For The Buck.

Good Day

That's why I never increase the voltage passed the manufacturer's specified amount. That way, if the stick dies there was something wrong with it to begin with.

 

[Starrbuck]

The OP already stated WoW performance improved, so there's one.

I think that is more related to the CPU overclock. When I was testing the overclocking of the CPU, I still had the memory at stock, yet WoW framerates still increased significantly. It is my belief that WoW in general is more limited by CPU than by memory.

 

[jimbo6846]

I think that is more related to the CPU overclock. When I was testing the overclocking of the CPU, I still had the memory at stock, yet WoW framerates still increased significantly. It is my belief that WoW in general is more limited by CPU than by memory.

Oh, I didn't see you say you OC'd your CPU. Games generally see a very slight improvement when your memory latency decreases, though. You'd likely see another slight improvement if you OC'd your RAM to 1066 and tightened your timings + Power Level as low as they could go.

How much did you OC your CPU?

 

[Nenu]

No increase worth a second thought. The voltage you set for your RAM is the top end of what is what is applied, not the middle or bottom end. If the motherboard sends more than what it's specifically told to (taking known overvoltage into account), it would only be a small amount and only for a microsecond when the sticks just finished a particularly heavy-duty job. This is covered under the manufacturer's warranty. This is normal.



The OP already stated WoW performance improved, so there's one.

Read again, it depends how far you can overclock as stated already.
The ONLY proof so far of a real world (?) performance increase is

A real world benchmark I have is that FPS went up considerably while standing at the same spot with roughly the same number of characters on the screen in WoW's city of Shattrath

Thats great if you want to stand still and not play a game, some better benchmarks would assist your case.

 

[jimbo6846]

Read again, it depends how far you can overclock as stated already.
The ONLY proof so far of a real world (?) performance increase is

Thats great if you want to stand still and not play a game, some better benchmarks would assist your case.

http://www.driverheaven.net/reviews.php?reviewid=591&pageid=5

 

[Nenu]

http://www.driverheaven.net/reviews.php?reviewid=591&pageid=5

Those are 800x600 benchmarks, set to lowest detail/quality and are not real world usage, only relevant if you have benchmark fever.
By "better benchmarks" I mean real world use situations.

 

[jimbo6846]

Those are 800x600 benchmarks, set to lowest detail/quality and are not real world usage, only relevant if you have benchmark fever.
By "better benchmarks" I mean real world use situations.

"The resolution and graphics details were set down to the minimum possible in order to maximize the load of the system and not that of the GPU."

In any case, click on the "back" arrow at the bottom of that page. There's a Photoshop benchmark.

 

[Nenu]

"The resolution and graphics details were set down to the minimum possible in order to maximize the load of the system and not that of the GPU."

In any case, click on the "back" arrow at the bottom of that page. There's a Photoshop benchmark.

Photoshop results do show an improvement and it is a real world application.
If you care about those particular uses of Photoshop, 'maybe' you would consider using faster ram.

I'm not sure I find those results believable though as a 10% increase in memory speed liberates a 21.6% increase in performance on the last test.
This isnt possible even under perfect test conditions.
The AVERAGE performance increase for that group of tests is 9.8%, way too close to the theoretical max of 10% to be true.

The game benchmarks you supplied are not representative of real world use, not even for gaming !
For gaming there is no noticeable benefit running faster ram on C2D

 

[jimbo6846]

Photoshop results do show an improvement and it is a real world application.
If you care about those particular uses of Photoshop, 'maybe' you would consider using faster ram.

I'm not sure I find those results believable though as a 10% increase in memory speed liberates a 21.6% increase in performance on the last test.
This isnt possible even under perfect test conditions.
The AVERAGE performance increase for that group of tests is 9.8%, way too close to the theoretical max of 10% to be true.

The game benchmarks you supplied are not representative of real world use, not even for gaming !
For gaming there is no noticeable benefit running faster ram on C2D

Every little bit counts. You could provide benchmarks to prove your side of the case, but I won't even ask. I've already seen the benefits of overclocking.

 

[Nenu]

Every little bit counts. You could provide benchmarks to prove your side of the case, but I won't even ask. I've already seen the benefits of overclocking.

Please report back with your 'benefits' if they further your case.

 

[jimbo6846]

Please report back with your 'benefits' if they further your case.

I wouldn't hold your breath. I don't plan on updating you.

 

[Nenu]

I wouldn't hold your breath. I don't plan on updating you.

 

[Starrbuck]

Thats great if you want to stand still and not play a game, some better benchmarks would assist your case.

That's a rude comment, and I'm not trying to make any case.

 

[Nenu]

That's a rude comment, and I'm not trying to make any case.

Dude, this isnt a girl scouts forum and that wasnt rude, it was matter of fact.
The guy had no foundation for his remarks and was capable of leading people to believe things which are not true.
People come here to get good information so they can wisely purchase new hardware, his comments couldnt really be left alone.
Someone had to take a stand, it was my turn

 

[jimbo6846]

Dude, this isnt a girl scouts forum and that wasnt rude, it was matter of fact.
The guy had no foundation for his remarks and was capable of leading people to believe things which are not true.
People come here to get good information so they can wisely purchase new hardware, his comments couldnt really be left alone.
Someone had to take a stand, it was my turn

The only thing you've added to this thread is claiming faster RAM has no real world benefit. I've proved that it does with my link. You then pulled out the "Those results are skewed, that can't be real" defense. Oh, and the "Who plays games at 800x600 resolution" line. I'm not sure why you think the results would be proportionally different at higher resolutions, but whatever.

This is getting pathetic.

 

[Nenu]

If you werent posting incorrect information I wouldnt have needed to contribute further than my original post #5, where I gave a fairly balanced opinion and advice.

Read the results you posted, they cannot be down to clockspeed alone as the difference is sometimes over twice the theoretical max achievable for a 10% clockspeed shift alone.
You also have taken a non real world scenario of playing games such that they are not GPU limited at all.
That test is to verify the maximum possible benchmarkable benefit between 2 different memory modules, not the same module at different speeds and not for everyday usage scenarios either.
You havent provided any worthwhile proof yet.

If you wish to prove that running faster ram on C2D will help the readers of this forum looking for advice on purchasing new hardware, please do.