Is there a huge difference between DDR2-667 and DDR2-800

[JEDIYoda]

Originally posted by: newb54
What are you smoking Jedi? There is no appreciable difference in performance between DDR2-667 and DDR2-800 when you run both at 533 speeds.

If you read my post that is what I said........

but they key is running both at 533 or even 667.......lol

 

[magreen]

AT reported on the benefits of memory bandwidth on overall system performance for both C2D and X2 last summer when AMD switched to DDR2 and the AM2 socket. It's a great article, well worth reading the whole thing.

On a C2D the higher memory bandwidth with DDR2 667/800 vs. DDR2 533 makes little difference in real world performance (including gaming)... maybe 3% performance gain going from 533 to 667 and another 3% from 667 to 800. So switching to DDR2 800 and configuring the memory multiplier to run your mem at DDR2 800 speed might gain you a whopping 5% or 6% increase.

Whereas on an X2 socket AM2 machine, it makes a much larger difference. 10% performance gain going from DDR2 533 to 667, and another 10% gain from 667 to 800. So you could see a 20% faster system if you configure for DDR2 800. It's due to the memory controller integrated in the chip, as someone mentioned above.

Bottom line:
DDR2 800 on a C2D only gives you headroom to OC your system, as has been stated above. That's all... the rest is negligible.

 

[Atheus]

Are you all saying that if I put DDR2-800 into a C2D, and the SPD is telling the system to run it at 800, the system will simply ignore it? Strange. Why would they do this? Regardless, anyone buying such memory would be an 'enthusiast' system builder, and would likely set it manually to it's rated speed.

Originally posted by: Roguestar
:thumbsdown::thumbsdown::thumbsdown::thumbsdown::thumbsdown::thumbsdown:

I don't know what you are smoking, but I hereby demand you provide benchmarks, specifically in Quake 4 seeing as you picked that example, that one person using a Core 2 Duo unoverclocked with DDR2-533 will experience a noticeable difference in framerate that will make an actual difference in the game. I also require you to prove that the difference in memory bandwidth in DDR2-533 and DDR2-800 will cause a significant bottleneck in gaming performance (Quake 4 again, thanks), with all other factors including components and gaming skill being equal. In fact, the only time it would make a difference ever would be if it somehow reduced the playability of the lower speed RAM's PC to beneath a playable framerate whilst the higher speeded RAM maintained a playable game. If there is in fact such a large difference in DDR2-533 and DDR2-800 in a real world situation, limited only to differing RAM speeds then I will conceed that you are correct.

Until that time I and the previous sane posters will maintain that you are infact both wrong and insane, and unaware of what a "noticeable difference" is in such a small factor (compared to CPU/GPU difference), and consider everything that comes out of your mouth on the subject to be FUD.

:thumbsdown::thumbsdown::thumbsdown::thumbsdown::thumbsdown::thumbsdown:

You are pretty much right, however you are also highly arrogant and very rude, so I feel obliged to point out that in the Anandtech benchmarks (http://www.anandtech.com/memory/showdoc.aspx?i=2800&p=7 - thanks magreen) Quake 4 gains a full 20FPS by going from the slowest to the fastest available memory. You can also see here (http://www.anandtech.com/memory/showdoc.aspx?i=2800&p=5) that the test system gets about a 30% improvement in memory latency - do you claim this will never translate into application performance? Sure, it won't affect everything, but there are definitely programs in which it *will* make a large difference.

This is of course running the memory at it's rated speed rather than the default 533, but 'demanding' people show a difference between 2 sets of memory both running at 533 is a ridiculous strawman argument.

 

[bigsnyder]

In the Quake4, the 20fps is only achieved when jumping from DDR2-400 to DDR2-1067 which is beyond the topic of the thread. Hopefully no Core2 user is running DDR2
at 400Mhz. Comparing DDR2-533 to DDR2-800 which is more realistic is only about 5fps. If you still compare to the fastest speed on the chart from DDR2-533, the difference then is about 11fps. 11fps is not enough of a return IMO for the amount of energy and expense needed to push speeds to DDR2-1112. Also keep in mind that the benches use
a X6800 processor. Since the multiplier can be adjusted on this CPU, it allows for higher memory clocks that would not be possible on the non-extreme models.

 

[Atheus]

Originally posted by: bigsnyder
In the Quake4, the 20fps is only achieved when jumping from DDR2-400 to DDR2-1067 which is beyond the topic of the thread. Hopefully no Core2 user is running DDR2
at 400Mhz. Comparing DDR2-533 to DDR2-800 which is more realistic is only about 5fps. If you still compare to the fastest speed on the chart from DDR2-533, the difference then is about 11fps. 11fps is not enough of a return IMO for the amount of energy and expense needed to push speeds to DDR2-1112. Also keep in mind that the benches use
a X6800 processor. Since the multiplier can be adjusted on this CPU, it allows for higher memory clocks that would not be possible on the non-extreme models.

Yea I know. As I said Roguestar is pretty much correct, but there's still no need for him to be an ass about it - it's not that clear cut.

 

[AnotherGuy]

LOL @ JediYoda

Dude I have the impression u have no clue what so ever how the computer works... and no im not talking about 2 players with different ram speeds who will win for sure 🙂
If u are not overclocking ur cpu will run at 266 x number of Multis in ur c2d = ...mhz. Ur memory will run at 1:1 with ur fsb... that means twice fsb in terms of DDR speed... so 266x2=ddr2-533.... If u were to overclock ur cpu... like fsb=400... then u have the ram at ddr2-800 .... at stock ur not running ur ddr2-800 at full speed... only 533

 

[Roguestar]

Originally posted by: Atheus
Yea I know. As I said Roguestar is pretty much correct, but there's still no need for him to be an ass about it - it's not that clear cut.

No, it's not clear cut. And my entire point was that there was no noticeable difference 😉. The thread title was "Is there a huge difference between DDR2-667 and DDR2-800" and my answer to that was "no", just like the potential question "Is there a huge difference between DDR2-533 and DDR2-800".

PS: @ bigsnyder; no Core 2 CPU other than the 800Mhz FSB Allendales can run DDR2-400 😛.

 

[Wedge1]

Originally posted by: JEDIYoda

Originally posted by: yh125d

There is no way he will have a 400fsb or even 333 because he is NOT overclocking!!!

If he gets 800, it WILL run at 533, as he is NOT overclocking.

you obviously know nothing about ram.......
everthing being equal including not over clocking if we both have mobo`s that support DDR2 667 and DDR2 800 and we have configured one to use 667 and the other to use 800 the 800 wins everytime sorry............

If the RAM runs at 533 on both mobo's, then one isn't faster than the other, be it DDR2 800 and DDR2 667. That's all he was saying. Yet, you disagree with this?

 

[Wedge1]

*deleted my duplicate post*

 

[jae]

so the timing wouldnt improve if ddr2 800 was used in a machine with a lower fsb

 

[soydios]

Just to set things straight: memory controllers have these nifty things called "memory straps" that let you run asynchronous FSB and DRAM clocks. For example, my E6600 is running at 9x266MHz, but my memory is running at 400MHz, both stock. You will get a tiny (less than one percent) increase in performance if you use DDR2-800 (3:2 strap). DDR2-667 (5:4 strap) runs roughly equal too DDR2-533 (1:1) because of the asynchronous memory strap, which isn't offset until DDR2-800.

 

[magreen]

Originally posted by: soydios
Just to set things straight: memory controllers have these nifty things called "memory straps" that let you run asynchronous FSB and DRAM clocks. For example, my E6600 is running at 9x266MHz, but my memory is running at 400MHz, both stock. You will get a tiny (less than one percent) increase in performance if you use DDR2-800 (3:2 strap). DDR2-667 (5:4 strap) runs roughly equal too DDR2-533 (1:1) because of the asynchronous memory strap, which isn't offset until DDR2-800.

The tiny performance increase you mention is true, but only for C2D.

It's because of the high latency inherent in the non-integrated memory controller on the C2D platform. Running the RAM faster does increase the theoretical memory bandwidth, but that high latency nullifies the effect of the extra bandwidth. So the extra theoretical bandwidth doesn't increase the measured memory performance significantly. If you read the article I linked to above, you'll see in fact the C2D would get smoked if it didn't have its intelligent read-ahead predicition method. If it just relied on its reads and writes, its memory access times would cripple the chip. Look at the memory write speed graph -- it's perfectly horizontal over all bandwidths tested -- because read-ahead can't help with writes.

Whereas on the X2 with its integrated mem controller, running the RAM faster has a large effect on actual measured memory performance, since you don't have those high latencies. And so it has a large effect on overall system performance.

 

[ForumMaster]

well without overcling then no. but since overclocking is so easy and safe these days, you should try it. but it's not worth it. C2D cpu's use a 1066Mhz FSB which is double the speed of the ram (533) so unless you plan on overclocking, it won't make a difference.

 

[Wedge1]

Originally posted by: ForumMaster
well without overcling then no. but since overclocking is so easy and safe these days, you should try it. but it's not worth it. C2D cpu's use a 1066Mhz FSB which is double the speed of the ram (533) so unless you plan on overclocking, it won't make a difference.

So what about raising the fsb to achieve these ~3GHz speeds that most Core2Duo's seem to be able do? That is worth it isn't it? Aren't we talking about very perceptible increases in performance?

I realize this question strays from what the OP is asking, but i just want to make sure that there is indeed an argument for buying RAM that is rated at 800MHz. Case in point, i'm looking to build a C2D system soon and I want to take that 6300 or 6400 up to from ~2GHz to ~3GHz. Surely we are talking about gains here that justify the cost. If i'm wrong, tell me so.

 

[apoppin]

yes .. if you OC then you need the fastest RAM that will support the OC you desire

damn it's GOOD to get out of Video

and maybe ... maybe if you upgrade, you just might want the faster RAM ... since there is not too much pricing difference - except for the cheap stuff.

 

[magreen]

Originally posted by: apoppin
yes .. if you OC then you need the fastest RAM that will support the OC you desire

You mean the slowest RAM that will support the OC you desire...

That is, the OC is limited by the RAM speed, so get fast enough RAM to support your OC.

 

[apoppin]

yeah ... more or less 😛

it's that G-D HD2900xt

i dropped out of ALL of those stupid discussions ... 'nice product' but AMD *bungled* the launch ... i hope they don't do the same thing with Barcelona ore there is gonna be just one CPU company 🙁

 

[KeithTalent]

When I put my new mobo (4Core Dual-VSTA) and CPU (E6600) in, there was a an option in the BIOS for RAM speed (or something like that). It was showing 533, but gave me an option of 667. I changed it to that, which is all I assumed I need to do? Do I need to do anything further?

Sorry for the mini-hijack! 😱

KT

 

[apoppin]

nope ... just make sure it really is

CPUz is a great program to confirm your settings

 

[KeithTalent]

Originally posted by: apoppin
nope ... just make sure it really is

CPUz is a great program to confirm your settings

Thanks Poppin! I'll run it when I get home.

Cheers,
KT
:beer: