Mushkin XP2 PC2-5300 DDR2 – Xtreme Performance Memory
by Steve Carmel & Wesley Fink on March 28, 2006 12:01 AM EST- Posted in
- Memory
Overclocking Memory Timings and Concerns
We were able to overclock using the DDR2-400, DDR2-533, DDR2-667 or DDR2-800 memory straps, although using the DDR2-800 memory strap did require looser timings. With an Intel chipset, the main memory always runs at a specified ratio, which is in relation to the clock rate for the system itself (AKA the FSB). Most modern chipset vendors now offer a wide range of memory ratios among the BIOS settings, so there are many choices for stock memory speed and base speed for overclocking.
As mentioned before, both the Intel NB 945 and 955x chipsets offer four distinct clocking ratios: 1:1, 3:4, 3:5 and 2:1. Proceeding from the base rate of 200 MHz (which translates to FSB800 for DDR2-400), this allows DDR2-400, DDR2-533, DDR2-667 and DDR2-800 settings. Now that DDR2 memory is maturing, manufacturers are finally addressing the increasing market for faster DDR2 at tighter memory timings. The DDR2 race for timings and speed will pick up pace with the introduction of the AMD AM2 socket and Intel Conroe later this year.
You can overclock an Intel system by just raising the front side bus speeds while you lower the memory ratio. The CPU clocks also work on a multiple of the front side bus clock speeds.
Since the Presler 955 Extreme Edition processor utilizes a 1066 MHz bus speed, it allows us to use the 1:1 memory to FSB clock ratio when the front side bus speed is set to 266 megahertz. Keep in mind that the Intel FSB is quad-pumped, which is why a setting of 266 yields a FSB of 1066 (4x266). DDR2 Memory on the Intel platform, however, is Double Data Rate by definition, so a 266 base setting is 533 (2x266). The chart below indicates relative scaling with front side bus speed increases for comparative purposes. SiSoft Sandra Professional Buffered and Unbuffered memory scoring were both included in benchmarking, along with Super PI mod 1.4 time to calculate 2 million places in seconds. The Half Life 2 - Lost Coast benchmark was run at a 1024 x 768 resolution at maximum quality settings with HDR enabled. Far Cry was set for ultra quality in the settings with a resolution of 1024 x 768.
You'll see from the chart below that the Mushkin PC2-5300 memory was able to maintain its relatively tight timings of 3-3-3-10 up to DDR 700 speeds with no problems.
There is a very little gaming performance gain using more relaxed memory timings necessary at the higher memory ratios, which has been a known issue with DDR2, especially on an Intel platform.
Although not included on the chart below, other games ran well with no issues using the Mushkin modules. Doom 3, at default settings, ran at 76.5 frames per second using the built-in time demo; while at 350 x 13, with a 1:1 ratio at 3-3-3-10, Doom 3 scored 89.3 frames per second. With the same settings, Quake 4 benchmarked at default at 139.1 FPS and reached 190.9 FPS at 350 x 13.
In most cases, the 13x350 timings yielded the best 1:1 memory performance, with higher game frame rates and unbuffered memory performance with the 3-3-3-10 timings. For this reason, the Highest Memory Speed (1:1) reported in the charts on the next page are based on the highest FSB speed at the 13 multiplier.
We were able to overclock using the DDR2-400, DDR2-533, DDR2-667 or DDR2-800 memory straps, although using the DDR2-800 memory strap did require looser timings. With an Intel chipset, the main memory always runs at a specified ratio, which is in relation to the clock rate for the system itself (AKA the FSB). Most modern chipset vendors now offer a wide range of memory ratios among the BIOS settings, so there are many choices for stock memory speed and base speed for overclocking.
As mentioned before, both the Intel NB 945 and 955x chipsets offer four distinct clocking ratios: 1:1, 3:4, 3:5 and 2:1. Proceeding from the base rate of 200 MHz (which translates to FSB800 for DDR2-400), this allows DDR2-400, DDR2-533, DDR2-667 and DDR2-800 settings. Now that DDR2 memory is maturing, manufacturers are finally addressing the increasing market for faster DDR2 at tighter memory timings. The DDR2 race for timings and speed will pick up pace with the introduction of the AMD AM2 socket and Intel Conroe later this year.
You can overclock an Intel system by just raising the front side bus speeds while you lower the memory ratio. The CPU clocks also work on a multiple of the front side bus clock speeds.
Since the Presler 955 Extreme Edition processor utilizes a 1066 MHz bus speed, it allows us to use the 1:1 memory to FSB clock ratio when the front side bus speed is set to 266 megahertz. Keep in mind that the Intel FSB is quad-pumped, which is why a setting of 266 yields a FSB of 1066 (4x266). DDR2 Memory on the Intel platform, however, is Double Data Rate by definition, so a 266 base setting is 533 (2x266). The chart below indicates relative scaling with front side bus speed increases for comparative purposes. SiSoft Sandra Professional Buffered and Unbuffered memory scoring were both included in benchmarking, along with Super PI mod 1.4 time to calculate 2 million places in seconds. The Half Life 2 - Lost Coast benchmark was run at a 1024 x 768 resolution at maximum quality settings with HDR enabled. Far Cry was set for ultra quality in the settings with a resolution of 1024 x 768.
You'll see from the chart below that the Mushkin PC2-5300 memory was able to maintain its relatively tight timings of 3-3-3-10 up to DDR 700 speeds with no problems.
There is a very little gaming performance gain using more relaxed memory timings necessary at the higher memory ratios, which has been a known issue with DDR2, especially on an Intel platform.
Although not included on the chart below, other games ran well with no issues using the Mushkin modules. Doom 3, at default settings, ran at 76.5 frames per second using the built-in time demo; while at 350 x 13, with a 1:1 ratio at 3-3-3-10, Doom 3 scored 89.3 frames per second. With the same settings, Quake 4 benchmarked at default at 139.1 FPS and reached 190.9 FPS at 350 x 13.
Mushkin XP2 PC2-5300 (DDR2- 533 - 1:1 Overclocked) - 2x1GB Double-Bank | |||||||
CPU Settings | Memory Speed 1:1 |
Memory Timings & Voltage | Far Cry fps | Sandra Unbuffered | Sandra Standard Buffered | Super PI Mod 1.4 2M places (time in sec) | Half Life 2 Lost Coast fps |
13x267 | 533 DDR | 3-3-3-10 1T 2.1V |
62.07 | INT 3402 FLT 3417 |
INT 6448 FLT 6451 |
85.7 | 78.2 |
13x333 | 666 DDR | 3-3-3-10 1T 2.3V |
76.26 | INT 4295 FLT 4296 |
INT 8109 FLT 8073 |
68.5 | 86.2 |
13x350 | 700 DDR | 3-3-3-10 1T 2.3V |
82.69 | INT 4442 FLT 4516 |
INT 8476 FLT 8493 |
65.2 | 87.9 |
12x370 (4.45 GHz) |
Highest Mem Speed 740 DDR |
5-5-5-15 1T 2.3V |
79.51 | INT 4076 FLT 4116 |
INT 8679 FLT 8662 |
67.6 | 86.3 |
In most cases, the 13x350 timings yielded the best 1:1 memory performance, with higher game frame rates and unbuffered memory performance with the 3-3-3-10 timings. For this reason, the Highest Memory Speed (1:1) reported in the charts on the next page are based on the highest FSB speed at the 13 multiplier.
11 Comments
View All Comments
PrinceGaz - Tuesday, March 28, 2006 - link
"I know that the built in spreaders are fantastic for heat dissapation"There is a lot of debate over whether heatspreaders made any difference at all even with the hotter running DDR modules at 2.5V+. Many people believe the heatspreaders are more to do with making them look good rather than perform better. I really don't know either way as all the heatspreaders on my DDR modules seem to do is give a larger flat surface area for the heat to be dissipated from after the heat has been conducted from the individual chips (the heatspreaders even with overvolted modules don't feel particularly warm), and I'm unconvinced if
(a) there is any significant surface area in contact between the chips and heatspreader, let alone whether any thermal compound was used
(b) that might mean you would get better heat dissipation *without* a heatspreader as the cool air would be blowing directly over the memory chips themselves instead of having a heatspreader in the way
Given that DDR modules at normal voltages (up to 3V anyway, I'm not talking about the OCZ modules that could take 3.5V or so -- more than old SDRAM even) without getting very warm, why would cooler running DDR2 need heatspreaders unless pushed up from 1.8V to at least 2.3V. Memory chips don't give off much heat even when over-volted so heat-spreaders seem unnecessary, especially on DDR2 modules. Once you combine that with the fact that at least half the surface area of a memory module without a heatspreader is taken up by the memory chips anyway, there doesn't really seem much point in adding something to "spread" the heat over what is only a slightly larger area.
I will concede that heatspreaders look good though, and the minimal cost of them is returned many times in every premium module they sell with them.