Does speed of memory matter for CS5?

[Randall Leong]

Some people claim that faster memory (while keeping the same overall latency) can improve performance in Premiere Pro CS5. To find out whether or not this claim is true, I tested my main rig with the PPBM5 benchmark with the i7-950 CPU set to its stock 3.2GHz (Turbo) and at DDR3-1066 (Intel's official maximum for the LGA 1366 CPUs), DDR3-1333 and DDR3-1600.

The memory that I used is a plain Corsair XMS3 1600C9 kit (two 6GB triple-channel kits). However, I was unable to run this memory at its full rated spec with all six DIMM slots filled (I had to use 10-11-11-29 timings rather than the advertised 9-9-9-24 at 1600 in order to run stably).

Here are my results (in MPE mode):

• DDR3-1066 (8-8-8-20): 313 seconds total, Relative Performance Index (relative to the top system on the PPBM5 results list as of 6 March 2011) 276.3
• DDR3-1333 (9-9-9-24): 301 seconds total, RPI: 268.1
• DDR3-1600 (10-11-11-29): 297 seconds total, RPI: 260.0

As it turned out, the memory speed does have some impact to the performance of CS5, but not by as much as the difference in memory bandwidth would have implied. In fact, there is a bigger difference in the total test time between 1066 and 1333 than between 1333 and 1600 although the differences in the RPI between the three speeds are about equal between adjacent grades.

As a check, I ran the PPBM5 benchmark with the CPU slightly overclocked to 3.36GHz (with the memory running at DDR3-1120 speed), and came up with about the same result as the stock-speed i7-950 did with the memory running at DDR3-1600.

Therefore, don't expect an upgrade from DDR3-1066 to DDR3-1600 memory to improve the performance of a stock-speed system by more than one to two CPU speed grades in CS5.

[David Chilson]

Randall,

Thanks for all the time you spend in your Bios, so others don't have to.

I think part of your instability problem lies in the fact that you are using 2 6gb kits versus a 12gb kit. Corsair guarantees for their ram to run at published timings and speed PROVIDED you use a binned 12gb kit listed on their web site for your motherboard, for example: CMP12GX3M3A1600C9.

When I was upgrading to an I7 920 and wanted 12GB of ram, two six gb kits was cheaper and I too went that route and had similiar instability problems. Swithched to a proper kit and with a slight overclock of 3.36 GHZ it runs at 1600 at 9-9-9-24 speeds. I didn't have to mess with any voltages, basically just raised the BLK until I reached the desired speed.

Not sure what running the ram at the correct timings and speed would have on the test, if any but if others are having the same problem getting their ram to run at the correct speed and timings, a binned ram kit is the way to go.

[Randall Leong]

Originally Posted by David Chilson

Randall,

Thanks for all the time you spend in your Bios, so others don't have to.

I think part of your instability problem lies in the fact that you are using 2 6gb kits versus a 12gb kit. Corsair guarantees for their ram to run at published timings and speed PROVIDED you use a binned 12gb kit listed on their web site for your motherboard, for example: CMP12GX3M3A1600C9.

When I was upgrading to an I7 920 and wanted 12GB of ram, two six gb kits was cheaper and I too went that route and had similiar instability problems. Swithched to a proper kit and with a slight overclock of 3.36 GHZ it runs at 1600 at 9-9-9-24 speeds. I didn't have to mess with any voltages, basically just raised the BLK until I reached the desired speed.

Not sure what running the ram at the correct timings and speed would have on the test, if any but if others are having the same problem getting their ram to run at the correct speed and timings, a binned ram kit is the way to go.

In my case, when I got the two 6GB kits, they cost $300 at the time while the cheapest 12GB kit still cost more than $500. That's a difference that's far too great for such a minuscule improvement in overall performance. And I would have still needed to spend more than $1,500 for only 12GB at the time just to gain any stability whatsoever because 4GB modules were still immature at the time.

Now that memory prices have fallen to just over $10 per GB (or less than $200 for 12GB), I might consider replacing those six 2GB modules with a 12GB (3 x 4GB) kit.

[Panagiotis Raris]

Randall; i too have Corsair XMS3 ram, 12GB and it was a single kit.

i can hit 9-9-9-24 at PC12800 (1600) speeds, and slightly higher than 1500 Mhz at 8-8-8-20 with mine at my 'standard' 4.0 Ghz overclock (manual overclock on an asus rampage iii gene).

I did not test with PPBM5 at the time i was sorting out memory timings/speeds; i tested on a single 1 hour M2T file export to H.264 Blu Ray with 3 layers of effects. 1500+ Mhz at 8-8-8-20 was within a second or two of 1600 at 9-9-9-24.

i had better results by pushing my i7 overclock to 4.2 Ghz. I can get the RAM to run stable at 1600 Mhz with 8-8-8-20 timings but the voltage is higher than i feel comfortable running, but i can confirm the kit series will do advertised speeds and lower timings; at least this kit will.

[Randall Leong]

Originally Posted by Panagiotis Raris

Randall; i too have Corsair XMS3 ram, 12GB and it was a single kit.

i can hit 9-9-9-24 at PC12800 (1600) speeds, and slightly higher than 1500 Mhz at 8-8-8-20 with mine at my 'standard' 4.0 Ghz overclock (manual overclock on an asus rampage iii gene).

I did not test with PPBM5 at the time i was sorting out memory timings/speeds; i tested on a single 1 hour M2T file export to H.264 Blu Ray with 3 layers of effects. 1500+ Mhz at 8-8-8-20 was within a second or two of 1600 at 9-9-9-24.

i had better results by pushing my i7 overclock to 4.2 Ghz. I can get the RAM to run stable at 1600 Mhz with 8-8-8-20 timings but the voltage is higher than i feel comfortable running, but i can confirm the kit series will do advertised speeds and lower timings; at least this kit will.

The difference here is that you have a single 6 x 2GB kit (HX3X12G1600C9, also sold as CMX12GX3M6A1600C9) while I have two separate 3 x 2GB kits (2 x CMX6GX3M3A1600C9) of identical revisions.

[Panagiotis Raris]

Randall, i am aware of that. I was thinking about running some PPBM tests similar to yours to see what differences timings vs speeds makes.

with mildly hiked timings, i could probably push my RAM to around 1800Mhz under 1.64v. I just sorted out and performed a rather lengthy series of benchmarks on my RAID/hard drive setups, and am going to retest my system with RAM at various speeds and timings,

thanks for the work thus far! Have you had any issues with your system at even multi's? odd question but i cannot get it to boot if i OC past 3.4 at either 22 or 24 multi's, and 21 is flaky as well. thus far ive only had luck at 23 multi

[Randall Leong]

Originally Posted by Panagiotis Raris

Randall, i am aware of that. I was thinking about running some PPBM tests similar to yours to see what differences timings vs speeds makes.

with mildly hiked timings, i could probably push my RAM to around 1800Mhz under 1.64v. I just sorted out and performed a rather lengthy series of benchmarks on my RAID/hard drive setups, and am going to retest my system with RAM at various speeds and timings,

thanks for the work thus far! Have you had any issues with your system at even multi's? odd question but i cannot get it to boot if i OC past 3.4 at either 22 or 24 multi's, and 21 is flaky as well. thus far ive only had luck at 23 multi

The problem with my system is the available memory multipliers. I have only discrete RAM multiplier settings in increments of 2x, starting with 6x. At 4.0GHz (174 x 23), I could not set anything between 1392MHz and 1740MHz for the RAM, so I get my RAM running either well below the RAM's advertised spec or beyond the advertised spec - with no happy in between at all whatsoever. If I want to run memory at anywhere close to the full 1600MHz spec while keeping the 23x CPU multiplier, I would have had to use a BCLK speed no higher than 160MHz (which results in the CPU speed that's below 3.7GHz) or a BCLK higher than 183MHz (resulting in a CPU speed that's high enough to require a massive increase in the Vcore and thus result in a massive increase in the CPU temperature enough to require exotic, super-expensive CPU cooling). My system would not even POST with the existing memory modules at 9-9-9-24 settings at 1600MHz even at 1.66V and a VTT of 1.375V (I had to use 10-11-11-29).

[Randall Leong]

An update:

I finally got this memory running stably at the advertised 9-9-9-24 @ 1600MHz. But it took a lot of tinkering with various voltages just to achieve this result. I had to set the VTT to 1.355V and the CPU PLL to 1.88V and disable C1/C3/C5/C7 states.

I ran PPBM5 at these new settings with the CPU running at both its stock 3.2GHz Turbo speed and an overclocked 4.01GHz.

The result? Virtually no difference in performance between 9-9-9-24 and 10-11-11-29 at either CPU speed.

[Paul Matwiy]

Any opinions on quantity of RAM? I'm running 8 GB (CS5 is using 6.5 GB) on a 12 core Opteron system running at 2.6 GHz with Win 7/64,

[Harm Millaard]

Opterons are far from ideal, but even those chips would benefit from more memory.

[Paul Matwiy]

Thanks for the information.

[Randall Leong]

Originally Posted by Randall Leong

An update:

I finally got this memory running stably at the advertised 9-9-9-24 @ 1600MHz. But it took a lot of tinkering with various voltages just to achieve this result. I had to set the VTT to 1.355V and the CPU PLL to 1.88V and disable C1/C3/C5/C7 states.

I ran PPBM5 at these new settings with the CPU running at both its stock 3.2GHz Turbo speed and an overclocked 4.01GHz.

The result? Virtually no difference in performance between 9-9-9-24 and 10-11-11-29 at either CPU speed.

I spoke too soon. The system still locked up and BSOD'd at 9-9-9-24 at 1600MHz. I could only get the full load stable at 1600MHz only with 10-11-11-29 timings. I might have to ditch these modules (sell them on you-know-what auction site) and order one or two 12GB triple-channel kits or six 4GB DDR3-1333 modules.