• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

How much does higher clocked memory improve performance?

S

static21

Gawd
Joined
Aug 1, 2004
Messages
524
I have a gigabyte ep45-ds3r motherboard that does ddr2 1200mhz memory. Currently I have 4 gigs of corsair 6400c4 in the system at 1:1 at i think 860mhz
I was thinking of getting some pc9600 1200 mhz memory,This stuff, but how much would this really effect performance? I honestly dont know a whole lot about memory. The only thing I really do is game, which is where I would want to see the performance. I also have a 5950 on the way.
 
Probably not enough to notice unless there is a significant gap in memory speeds. Higher rated memory speeds are more beneficial in overclocking headroom on boards when you have to run a high FSB because of the low multiplier. A higher clockspeed on the CPU is going to improve performance more than just a higher memory speed generally speaking.
 
If you only increase memory speed and do not also increase CPU speed, the most you'll ever see in real world usage is typically going to be about 2% - 5% at most. If your memory is holding back your CPU FSB speed, then the faster memory makes sense.
 
Ended up selling my old ram for 60 bucks, and getting some corsair dominator pc8500 on craigslist for 50 bucks =p Couldn't pass that up.
 
i also wonder that for myself. does having faster memory makes any difference, today?
 
On systems where the memory controller is not built into the CPU, you won't see a whole lot of gain, at least not nearly as much as you see with a system that has the memory controller built onto the CPU.

Also, if you do get faster RAM on a system that doesn't have the memory controller built onto the CPU, it really doesn't help to run the RAM faster than 1:1.. at least not enough to warrant buying faster RAM. If you can run high FSB and keep the RAM at 1:1 you will see a much higher gain.

And for systems that have the memory controller built onto the CPU, you see a definite gain when running faster RAM even if you leave the CPU at a lower speed.

For example, with the RAM in my system in my sig running at 3.6Ghz and the RAM running at DDR3-1800, I have a very memory intensive program that is faster then my bro's i7 system running at 4.2Ghz with DDR3-1600 RAM.

In either case, faster RAM will make the system seem snappier and smoother.
 
No! You will not "feel it". On the other hand if the faster RAM really is only $20 to $40 then .. meh. That is less than a tank of gas. Buy the better RAM

I have run some comparative benchmarks with a friend that used to work in Intel server division, once upon a time. He was/is all over max RAM speed. I have never seen or experience any significant improvement in any system because of RAM speed only ... well, ok, benchmarks only. My position it that it is all about CPU speed.

In the i7 systems because of multipliers, you often have to back off RAM speed to something ridiculously slow because the multipliers are pushing the RAM too high if going for max CPU OC. Max CPU OC wins.
 
often the case if you say 2000mhz ram and run it at 1500mhz it will run alot faster than stock 1500mhz ram because of timings. i once bought ram that had almost 0 0 0 2 timings it was very quick
 
aweusx2 said:
often the case if you say 2000mhz ram and run it at 1500mhz it will run alot faster than stock 1500mhz ram because of timings. i once bought ram that had almost 0 0 0 2 timings it was very quick
Click to expand...

Best joke I heard in the last 3 min :p
 
aweusx2 said:
often the case if you say 2000mhz ram and run it at 1500mhz it will run alot faster than stock 1500mhz ram because of timings. i once bought ram that had almost 0 0 0 2 timings it was very quick
Click to expand...

That's not true. I would say you would not be able to tell the difference outside of synthetic memory bandwidth benchmarks between 2000 and 1500mhz, even with tighter timings on the 1500mhz. Both are going to be plenty fast.
 
ar09 said:
bold statement!? is it true?
Click to expand...

Been building systems for 19 years now... yes it is true.

And I have tested it on my i7 system.

Lowering the RAM multiplier while leavinfg everything else the same makes it slower at loading up windows, switching between programs, etc.

Latency goes down quite a bit on the i7 platform with faster RAM.

Even with my Q6600 setup, it was fairly easy to tell the difference between slower and faster RAM when the RAM was at a FSB:RAM speed ratio of 1:1.
 
^^^ at 1:1, the memory is running the same speed with "fast" memory and "slow" memory, how can you tell the difference?
 
cyclone3d said:
Been building systems for 19 years now... yes it is true.

And I have tested it on my i7 system.

Lowering the RAM multiplier while leavinfg everything else the same makes it slower at loading up windows, switching between programs, etc.

Latency goes down quite a bit on the i7 platform with faster RAM.

Even with my Q6600 setup, it was fairly easy to tell the difference between slower and faster RAM when the RAM was at a FSB:RAM speed ratio of 1:1.
Click to expand...

Would you call DDR3 1333 MHz SLOW and DDR3 1600 MHz FAST on an i7-930 system?

Looking at a 1333 Mhz set for $101. 1600 Mhz is about $20 more.

It's not very sexy RAM but that price is enticing me big time (w/ rebate & BCB). It's just looking at it this is the "SLOW" ddr3 memory. I'm not looking for my comp to bleed here.
http://www.tigerdirect.com/applications/SearchTools/item-details.asp?EdpNo=4282840&CatId=4093
 
Higher clocked RAM is going to be a real benefit in only a few instances.

Overclocking where you RAM is holding back your max overclock is one of those instances. Having the higher clocked RAM allowing a higher overclock will show real world performance. The memory bandwidth increase is not often the increase in performance, though.

Memory bandwidth limited applications will be another example where higher clocked RAM will show real world performance increases. However, keep in mind that the rest of the architecture such as the CPU and chipset (in cases where the memory controller is still located on the northbridge) also have to be able to make use of the additional bandwidth. Also, memory bandwidth intensive applications are few and far between for the average user or even a power user like many here. Applications which need a lot of RAM but are not bandwidth limited are not included here. Those type of applications want high amounts of RAM.

With the current dual and triple channel RAM setups and on-die CPU memory controllers, memory bandwidth is rarely limited outside of specialized applications. Sure, you can "see" a difference in some benchmarks but that very rarely translates into noticeable real world performance.

I won't tell anyone not to get higher clocked RAM but suggest they look at their needs and see if higher clocked RAM is actually needed or not. If it's not needed, I generally suggest going for the likely cheaper lower clocked RAM which fits their needs.

Also, for anyone who thinks that running higher clocked RAM at lower speeds which match lower clocked RAM will be faster, you are deluding yourself. They are going to have identical performance. The only time you might get better performance is if the higher clocked RAM will be able to run tighter timings than the lower clocked RAM at the same speeds. Even then, the tighter timings generally won't show up as increased performance in real world scenarios outside of bandwidth intensive applications.

While the newest platform I'm using is socket 775 with a Q6600, I have done a lot of testing of RAM bandwidth and latency over the years on different platforms. While some platforms show a bit more difference in benchmarks than others, I have never seen a real world performance difference. That said, I've never run memory bandwidth intensive applications which would benefit from running much higher bandwidth RAM. Generally speaking, most platforms have much higher memory bandwidth than the platform can properly use.

 
atreidesgoldenpath said:
Would you call DDR3 1333 MHz SLOW and DDR3 1600 MHz FAST on an i7-930 system?

Looking at a 1333 Mhz set for $101. 1600 Mhz is about $20 more.

It's not very sexy RAM but that price is enticing me big time (w/ rebate & BCB). It's just looking at it this is the "SLOW" ddr3 memory. I'm not looking for my comp to bleed here.
http://www.tigerdirect.com/applications/SearchTools/item-details.asp?EdpNo=4282840&CatId=4093
Click to expand...

Personally I would go with the 1600Mhz set if it is only $20 more.

And if you overclock - almost a "crime" if you don't, it will be a lot easier to stay within the RAMs specs than with the 1333 set.
 
so if you overclock the i7 930 will 1333mhz ram limit how much you can overclock?

I am wanting to do a basic overclock without changing voltages, eg 3.4ghz hopefully
Would 1333 be ok or should I opt for something higher such as 1600?
 
cyclone3d said:
On systems where the memory controller is not built into the CPU, you won't see a whole lot of gain, at least not nearly as much as you see with a system that has the memory controller built onto the CPU.

Also, if you do get faster RAM on a system that doesn't have the memory controller built onto the CPU, it really doesn't help to run the RAM faster than 1:1.. at least not enough to warrant buying faster RAM. If you can run high FSB and keep the RAM at 1:1 you will see a much higher gain.

And for systems that have the memory controller built onto the CPU, you see a definite gain when running faster RAM even if you leave the CPU at a lower speed.

For example, with the RAM in my system in my sig running at 3.6Ghz and the RAM running at DDR3-1800, I have a very memory intensive program that is faster then my bro's i7 system running at 4.2Ghz with DDR3-1600 RAM.

In either case, faster RAM will make the system seem snappier and smoother.
Click to expand...

In my particular i7 system, there is very little difference in stock-CPU-speed (2.66GHz) performance between DDR3-1600 RAM and DDR3-1066 RAM: Encoding 16 seconds worth of HD 1920x1080/59.94i content to Blu-Ray-compatible AVC took 92 seconds with DDR3-1066 memory while DDR3-1600 memory only reduced it to 89 seconds. Three seconds out of 90+ seconds - meh. (And this with the DDR3-1066 memory running at 8-8-8 while the DDR3-1600 memory ran at 7-7-7, which confirms my findings that are described in the next paragraph, below.) Hence, overclocking the CPU (in my case, to 3.5GHz) makes more sense in this case.

Also, you were running that memory with an FSB : DRAM ratio of 2:10. (This is the ratio style on all processors with on-CPU memory controllers.) But once the memory controller's ratio goes beyond 2:10 (such as the 2:12 ratio that I ran with memory at 1600 and processor at 2.66), performance will actually start to be reduced rather than enhanced. This is because stretching out the multiplier too far will increase the total latency of the memory controller. Also, the DDR3-1600 memory that your brother's system is running is most likely to be of the higher-latency type (9-9-9 at that speed), while your DDR3-1800 memory actually ran at 7-7-7 or 8-8-8 at that speed.
 
Last edited:
cyclone3d said:
Personally I would go with the 1600Mhz set if it is only $20 more.

And if you overclock - almost a "crime" if you don't, it will be a lot easier to stay within the RAMs specs than with the 1333 set.
Click to expand...

$20 more is 20% more cost. I'm doubt you'll get 20% more performance.

Modern CPUs have been optimized for slow memory by including larger on-die caches. The last time Intel supported a strategy that looked at utilizing faster RAM was before 2000 as part of the Rambus fiasco. With Core2Duo/DDR2 and i5/i7 DDR3 the incremental increases between speed grades outside of specific memory intensive applications and synthetic benchmarks is <10% (DDR3-1066 C7 to DDR3-1866 C7). Also watch out for higher latencies that come with higher speed grades.

Read this article and save your money unless your application fits in the specialized categories.

http://www.anandtech.com/memory/showdoc.aspx?i=3589&p=13

I miss the days you could get 4GB DDR2 800MHz for <$50. I don't see why I should spend $200 for a 3 channel kit 6GB kit. I ended up getting OCZ3P1333LV4GK (2x 2GB) for an i7 930 build that I'll OC to 4GHz. Price $68 after MIR and BCB shipped. I'll get 2 kits at the same price when they go on sale again. Yes I do video editing and the higher OC + 64bit OS and soon >8GB memory will help more than spending money on a higher speed grade which will be <2-4% faster in real world applications.
 
sfsilicon said:
$20 more is 20% more cost. I'm doubt you'll get 20% more performance.

Modern CPUs have been optimized for slow memory by including larger on-die caches. The last time Intel supported a strategy that looked at utilizing faster RAM was before 2000 as part of the Rambus fiasco. With Core2Duo/DDR2 and i5/i7 DDR3 the incremental increases between speed grades outside of specific memory intensive applications and synthetic benchmarks is <10% (DDR3-1066 C7 to DDR3-1866 C7). Also watch out for higher latencies that come with higher speed grades.

Read this article and save your money unless your application fits in the specialized categories.

http://www.anandtech.com/memory/showdoc.aspx?i=3589&p=13

I miss the days you could get 4GB DDR2 800MHz for <$50. I don't see why I should spend $200 for a 3 channel kit 6GB kit. I ended up getting OCZ3P1333LV4GK (2x 2GB) for an i7 930 build that I'll OC to 4GHz. Price $68 after MIR and BCB shipped. I'll get 2 kits at the same price when they go on sale again. Yes I do video editing and the higher OC + 64bit OS and soon >8GB memory will help more than spending money on a higher speed grade which will be <2-4% faster in real world applications.
Click to expand...

The problem with using four DIMMs on a triple-channel platform is that such a configuration would have forced the memory controller to default to a funky hybrid mixed mode of triple- and single-channel, which would have significantly increased the total latency of the memory controller and thus slow down real-world application performance (in some of the apps, the performance would have dropped by more than 20% compared to having only 6GB in triple-channel). Plus, the way most X58 motherboards are set up, you are required to have one memory module in all three channels before you can install the second module in one of the channels. If you're going to run only 8GB via only four DIMMS, you might as well choose to spend less money and go instead with a lower-end platform, such as a Socket 1156 i7-8xx series or i5-7xx series system.

Now I know why 8GB on a Socket LGA1366 platform runs slower overall than 4GB or 6GB...
 
Last edited:
E4g1e said:
The problem with using four DIMMs on a triple-channel platform is that such a configuration would have forced the memory controller to default to a funky hybrid mixed mode of triple- and single-channel, which would have significantly increased the total latency of the memory controller and thus slow down real-world application performance (in some of the apps, the performance would have dropped by more than 20% compared to having only 6GB in triple-channel). Plus, the way most X58 motherboards are set up, you are required to have one memory module in all three channels before you can install the second module in one of the channels. If you're going to run only 8GB via only four DIMMS, you might as well choose to spend less money and go instead with a lower-end platform, such as a Socket 1156 i7-8xx series or i5-7xx series system.

Now I know why 8GB on a Socket LGA1366 platform runs slower overall than 4GB or 6GB...
Click to expand...

I'm running dual channel on my MSI X58 Pro-E (verified w/ CPU-Z). I checked around about the dual vs triple performance and clearly in synthetics you see a major difference, but in real world apps it doesn't translate to a major difference. If memory was at $12.5 per GB (= $50 for 4GB or $75 for 6GB) I'd go with a triple. Till then I'll wait for good deals and the X58 platform allows me the maximum flexibility to upgrade over time.
 
sfsilicon said:
I'm running dual channel on my MSI X58 Pro-E (verified w/ CPU-Z). I checked around about the dual vs triple performance and clearly in synthetics you see a major difference, but in real world apps it doesn't translate to a major difference. If memory was at $12.5 per GB (= $50 for 4GB or $75 for 6GB) I'd go with a triple. Till then I'll wait for good deals and the X58 platform allows me the maximum flexibility to upgrade over time.
Click to expand...

I stated my fact in my post above. Running four memory modules will NOT put your system in dual-channel mode at all on most X58 motherboards (which require all three slots for triple-channel be filled before any additional memory can be installed). Instead, the memory controller would be forced into the funky hybrid mode in which any memory not being used for triple-channel would run in single-channel mode. It is this hybrid (which Intel calls the "Flex" mode) which increases the total latency of a memory controller and thus reduces performance somewhat. (Maybe I was exaggerating the performance loss a bit, but it is application-dependent.) The loss is very similar to that which occurs when one runs a pair of SLI'd or CrossFired PCI-e graphics cards in an unbalanced x16/x4 mode instead of x8/x8. In other words, such a system would actually perform slower than cheaper, slower systems of the same CPU generation with the same video and mass storage/system drive components.

On these very motherboards, when I tried to ignore this and instead install two pairs of memory in dual-channel, the systems would refuse to even POST. That is a motherboard BIOS limitation, not a memory controller limitation.
 
Last edited:
You must log in or register to reply here.
Back
Top