[nextpage title=”Introduction”]
Corsair TWINX2048-3500LLPRO memory kit uses two 1 GB CMX1024-3500LLPRO memory modules (219 MHz/DDR438, 2-3-2-6 timings) and is optimized to be used on ASUS A8N32-SLI motherboards series. CMX1024-3500LLPRO modules were the first to use CAS 2 latency, which is considered very aggressive in high-capacity modules (1 GB) with a 219 MHz clock. Let’s see how this kit goes on our review.
Figure 1: Corsair TWINX2048-3500LLPRO memory kit.
The module has the same look as the other XMS Pro series modules from Corsair. The modules have a big black aluminum heatspreader and on their top they have a series of 18 LEDs that shows the memory activity. For those that have a case with a transparent side window the effect is very interesting. On the other hand these modules are too high, what makes it difficult to use them with certain motherboards when bigger CPU coolers are used.
The reviewed modules use Infineon BE-5 chips and printed circuit board (PCB) designed by BrainPower.
Infineon BE-5 chips can run up to 260 MHz using a low operating voltage.
Before going to our benchmarks, let’s take a look at the main technical features of the reviewed modules.
[nextpage title=”Main Specifications”]
Dual channel Corsair TWINX2048-3500LLPRO memory kit main features are:
- Rating: DDR438/PC3500.
- Capacity: 2 x 1,024 MB.
- Type: Non-ECC, Unregistered and Unbuffered.
- Recommended voltage: 2,6 V.
- Programmed timings:
Clock (MHz) | 219 |
CAS Latency Control (TCL) | 2 |
RAS# to CAS# delay (TRCD) | 3 |
RAS# Precharge (TRP) | 2 |
RAS# Active Time (TRAS) | 6 |
- Extra Features: Heatspreader with 18 activity LEDs, dual channel-optimized and lifetime warranty.
- More Information: https://www.corsair.com
- Average Price in the US*: USD 264.00
* Researched at Shopping.com on the day we published this review. [nextpage title=”How We Tested”] All memory modules were tested under dual channel configuration (2x 1,024 MB). Since the goal of our review is to measure the memory performance, all benchmarks were conducted with the CPU as close as possible from its original clock speed of 2.2 GHz (Athlon 64 3500+). This insured that the results from our benchmarks were influenced only by the memory clock and by the memory latency.
We ran our benchmarks using the configurations listed below, always configuring the memories with the lower supported latencies with stability.
CPU Clock | Memory Clock(DDR) |
200 x 11 = 2.200 MHz | 200 MHz (DDR400) |
220 x 10 = 2.200 MHz | 220 MHz (DDR440) |
245 x 9 = 2.205 MHz | 245 MHz (DDR490) |
275 x 8 = 2.200 MHz | 275 MHz (DDR 550) |
If the memory was able to go beyond 275 MHz we controlled the CPU multiplier in order to keep its internal clock as close as possible from its original 2.2 GHz.
During our benchmarking sessions, we used the configuration listed below. Between our benchmarking sessions the only variable device were the memory modules being tested.
Hardware Configuration
- Motherboard: DFI LANParty UT nF4 SLI-D (nForce4 SLI).
- BIOS version: B14 – November 14th, 2005.
- PCB revision: A00.
- CPU: Athlon 64 3500+ (2.2 GHz) with 512 KB L2 memory cache and Venice core.
- Cooler: In-a-Box.
- Hard Disk Drive: Samsung SpinPoint SP0411N (7,200 rpm, 40 GB, ATA-133).
- Video Card: NVIDIA Geforce 6800 GT 256 MB PCI Express.
- Video Resolution: 1024x768x32@72Hz.
- Power Supply: Seventeam ST420BKV-03F.
Software Configuration
- Windows XP Professional, installed on NTFS.
- Service Pack 2.
- Direct X 9.0c.
Driver Versions
- NVIDIA video driver version: 81.98 WHQL.
- NVIDIA nForce driver version: 6.70 WHQL.
Used Software
We adopted a 3% error margin; thus, differences below 3% cannot be considered relevant. In other words, products with a performance difference below 3% should be considered as having similar performance.
[nextpage title=”Benchmarking at 200 MHz/DDR400″] We used four programs to measure the performance of the memory modules: Sandra Lite, 3DMark2001 SE, Quake III and Super Pi Mod 1.4.
The following memory modules were selected to be compared with Corsair TWINX2048-3500LLPRO 2 GB Dual Channel kit, all containing two 1 GB modules each:
- OCZ EL DDR PC-4000 Gold GX XTC
- Patriot PDC2G3500LLK
We separated our tests according to the memory clock we configured. Let’s see the results achieved by the reviewed modules running at 200 MHz/DDR400, i.e., at the system default configuration.
Sandra Lite 2005.SR3 10.69
We used the Memory Bandwidth Benchmark module from Sandra to measure the maximum transfer rate that the memory was able to achieve. The result published is an arithmetic average from the Buffered Integer Stream and Buffered Float Stream results.
You can check the results on the graph below.
Memories – 200MHz/DDR400 1T | Sandra – Int and Float Average | % |
Corsair TWINX2048-3500LLPRO (2-3-2-5) – 2.6 V | 5773 | |
Patriot PDC2G3500LLK (2-3-2-5) – 2.6 V | 5731 | -0.73 |
OCZ EL DDR PC-4000 Gold GX XTC (2.5-3-3-7) – 2.6 V | 5698 | -1.32 |
Even though it has a lower latency Corsair TWINX2048-3500LLPRO achieved a performance similar to the other memory modules we tested.
3DMark2001 SE Build 3.3.0
3DMark2001 SE simulates older games based on DirectX 8.1 and it is very sensitive to changes in the hardware configuration of the system.
You can check the results on the graph below.
Memories – 200MHz/DDR400 1T | 3DMark2001 SE (Build 3.3.0) | % |
Corsair TWINX2048-3500LLPRO (2-3-2-5) – 2.6 V | 22690 | |
Patriot PDC2G3500LLK (2-3-2-5) – 2.6 V | 22676 | -0.06 |
OCZ EL DDR PC-4000 Gold GX XTC (2.5-3-3-7) – 2.6 V | 22408 | -1.26 |
At 3DMark2001 SE all tested memories achieved the same performance level.
Quake III Arena 1.32
Even though this is an old game its importance comes from the fact its engine is used by several other games, like Jedi Knight II and Medal of Honor, and also because this game is very sensitive to any changes on the hardware configuration.
We used the demo four available on version 1.32 of Quake III to make our benchmarking with this game. We ran this demo three times at 1024x768x32 resolution and all image quality settings on their default configuration and we picked the middle value for our comparisons, i.e., we discarded the highest and the lowest values.
Check the results below.
Memories – 200MHz/DDR400 1T | Quake III Arena – FPS | % |
Patriot PDC2G3500LLK (2-3-2-5) – 2.6 V | 456 | +0.53 |
Corsair TWINX2048-3500LLPRO (2-3-2-5) – 2.6 V | 453.6 | |
OCZ EL DDR PC-4000 Gold GX XTC (2.5-3-3-7) – 2.6 V | 444.4 | -2.07 |
Corsair’s kit achieved a performance similar to Patriot’s, which is also based on Infineon chips, and was 2.07% faster than OCZ’s, which, according to our methodology, means that they have a similar performance, since the performance difference was below 3%.
Super Pi Mod 1.4
Super Pi is a benchmarking software created at Tokyo University which measures the time spent to calculate Pi up to 32 million decimal places. In our tests we calculated Pi with 1 million decimal places.
Check the results below.
Memories – 200MHz/DDR400 1T | Super Pi Mod 1.4 | % |
Corsair TWINX2048-3500LLPRO (2-3-2-5) – 2.6 V | 38.640 | |
Patriot PDC2G3500LLK (2-3-2-5) – 2.6 V | 38.750 | +0.28 |
OCZ EL DDR PC-4000 Gold GX XTC (2.5-3-3-7) – 2.6 V | 39.172 | +1.36 |
On Super Pi the lower the result, the better, since it means seconds. Corsair’s kit achieved a performance similar to Patriot’s and was 1.36% faster than OCZ’s, which, according to our methodology, means that they have a similar performance, since the performance difference was below 3%.
[nextpage title=”Benchmarking at 220 MHz/DDR440″] We lowered the clock multiplier of our Athlon 64 3500+ (2.2 GHz) from 11x to 10x and increased its memory bus from 200 MHz to 220 MHz, making the memories to run at DDR440 keeping the CPU running internally at 2.2 GHz. Let’s see how the reviewed kit performed under this configuration.
Sandra Lite 2005.SR3 10.69
We used the Memory Bandwidth Benchmark module from Sandra to measure the maximum transfer rate that the memory was able to achieve. The result published is an arithmetic average from the Buffered Integer Stream and Buffered Float Stream results.
You can check the results on the graph below.
Memories – 220MHz/DDR440 1T | Sandra – Int and Float Average | % |
OCZ EL DDR PC-4000 Gold GX XTC (2.5-3-3-7) – 2.6 V | 5899 | +0.08 |
Corsair TWINX2048-3500LLPRO (2-3-2-5) – 2.6 V | 5894 | |
Patriot PDC2G3500LLK (2-3-2-5) – 2.6 V | 5894 | 0 |
On this test all memories achieved the same performance level.
3DMark2001 SE Build 3.3.0
3DMark2001 SE simulates older games based on DirectX 8.1 and it is very sensitive to changes in the hardware configuration of the system.
You can check the results on the graph below.
Memories – 220MHz/DDR440 1T | 3DMark2001 SE (Build 3.3.0) | % |
Corsair TWINX2048-3500LLPRO (2-3-2-5) – 2.6 V | 22922 | |
Patriot PDC2G3500LLK (2-3-2-5) – 2.6 V | 22866 | -0.24 |
OCZ EL DDR PC-4000 Gold GX XTC (2.5-3-3-7) – 2.6 V | 22622 | -1.33 |
On 3DMark2001 SE all tested memories also have achieved the same performance level.
Quake III Arena 1.32
Even though this is an old game its importance comes from the fact its engine is used by several other games, like Jedi Knight II and Medal of Honor, and also because this game is very sensitive to any changes on the hardware configuration.
We used the demo four available on version 1.32 of Quake III to make our benchmarking with this game. We ran this demo three times at 1024x768x32 resolution and all image quality settings on their default configuration and we picked the middle value for our comparisons, i.e., we discarded the highest and the lowest values.
Check the results below.
Memories – 220MHz/DDR440 1T | Quake III Arena – FPS | % |
Patriot PDC2G3500LLK (2-3-2-5) – 2.6 V | 465.3 | +0.58 |
Corsair TWINX2048-3500LLPRO (2-3-2-5) – 2.6 V | 462.6 | |
OCZ EL DDR PC-4000 Gold GX XTC (2.5-3-3-7) – 2,6 V | 452.6 | -2.21 |
Corsair’s kit achieved a performance similar to Patriot’s and was 2.21% faster than OCZ’s, which, according to our methodology, means that they have a similar performance, since the performance difference was below 3%.
Super Pi Mod 1.4
Super Pi is a benchmarking software created at Tokyo University which measures the time spent to calculate Pi up to 32 million decimal places. In our tests we calculated Pi with 1 million decimal places.
Check the results below.
Memories – 220MHz/DDR440 1T | Super Pi Mod 1.4 | % |
Corsair TWINX2048-3500LLPRO (2-3-2-5) – 2.6 V | 38.423 | |
Patriot PDC2G3500LLK (2-3-2-5) – 2.6 V | 38.438 | +0.04 |
OCZ EL DDR PC-4000 Gold GX XTC (2.5-3-3-7) – 2.6 V | 38.781 | +0.92 |
On Super Pi the lower the result, the better, since it means seconds. Corsair TWINX2048-3500LLPRO achieved a performance similar to the other memory modules we tested.
[nextpage title=”Benchmarking at 245 MHz/DDR490″] We lowered the clock multiplier of our Athlon 64 3500+ (2.2 GHz) from 11x to 9x and increased its memory bus from 200 MHz to 245 MHz, making the memories to run at DDR490 keeping the CPU running internally at 2.2 GHz. Let’s see how the reviewed kit performed under this configuration.
Sandra Lite 2005.SR3 10.69
We used the Memory Bandwidth Benchmark module from Sandra to measure the maximum transfer rate that the memory was able to achieve. The result published is an arithmetic average from the Buffered Integer Stream and Buffered Float Stream results.
You can check the results on the graph below.
Memories – 245MHz/DDR490 1T | Sandra – Int and Float Average | % |
Patriot PDC2G3500LLK (2.5-3-2-5) – 2.7 V | 6202 | +0.19 |
Corsair TWINX2048-3500LLPRO (2.5-3-2-5) – 2.7 V | 6190 | |
OCZ EL DDR PC-4000 Gold GX XTC (3-4-3-7) – 2.7 V | 6131 | -0.96 |
On this test all memories achieved the same performance level.
3DMark2001 SE Build 3.3.0
3DMark2001 SE simulates older games based on DirectX 8.1 and it is very sensitive to changes in the hardware configuration of the system.
You can check the results on the graph below.
Memories – 245MHz/DDR490 1T | 3DMark2001 SE (Build 3.3.0) | % |
Patriot PDC2G3500LLK (2.5-3-2-5) – 2.7 V | 23245 | +0.28 |
Corsair TWINX2048-3500LLPRO (2.5-3-2-5) – 2.7 V | 23180 | |
OCZ EL DDR PC-4000 Gold GX XTC (3-4-3-7) – 2.7 V | 22726 | -2.00 |
Corsair’s kit achieved a performance similar to Patriot’s and was 2% faster than OCZ’s, which, according to our methodology, means that they have a similar performance, since the performance difference was below 3%.
Quake III Arena 1.32
Even though this is an old game its importance comes from the fact its engine is used by several other games, like Jedi Knight II and Medal of Honor, and also because this game is very sensitive to any changes on the hardware configuration.
We used the demo four available on version 1.32 of Quake III to make our benchmarking with this game. We ran this demo three times at 1024x768x32 resolution and all image quality settings on their default configuration and we picked the middle value for our comparisons, i.e., we discarded the highest and the lowest values.
Check the results below.
Memories – 245MHz/DDR490 1T | Quake III Arena – FPS | % |
Patriot PDC2G3500LLK (2.5-3-2-5) – 2.7 V | 474.8 | +0.08 |
Corsair TWINX2048-3500LLPRO (2.5-3-2-5) – 2.7 V | 474.4 | |
OCZ EL DDR PC-4000 Gold GX XTC (3-4-3-7) – 2.7 V | 460.7 | -2.97 |
Once again Corsair’s kit achieved a performance similar to Patriot’s and was 2.97% faster than OCZ’s, which, according to our methodology, means that they have a similar performance, since the performance difference was below 3%.
Super Pi Mod 1.4
Super Pi is a benchmarking software created at Tokyo University which measures the time spent to calculate Pi up to 32 million decimal places. In our tests we calculated Pi with 1 million decimal places.
Check the results below.
Memories – 245MHz/DDR490 1T | Super Pi Mod 1.4 | % |
Patriot PDC2G3500LLK (2.5-3-2-5) – 2.7 V | 37.860 | -0.25 |
Corsair TWINX2048-3500LLPRO (2.5-3-2-5) – 2,7 V | 37.953 | |
OCZ EL DDR PC-4000 Gold GX XTC (3-4-3-7) – 2.7 V | 38.609 | +1.70 |
On Super Pi the lower the result, the better, since it means seconds. Corsair’s kit achieved a performance similar to Patriot’s and was 1.70% faster than OCZ’s, which, according to our methodology, means that they have a similar performance, since the performance difference was below 3%.
[nextpage title=”Benchmarking at The Maximum Clock Rate”] We lowered the clock multiplier of our Athlon 64 3500+ (2.2 GHz) from 11x to 8x and increased its memory bus from 200 MHz to 275 MHz, making the memories to run at DDR550 keeping the CPU running internally at 2.2 GHz. The reviewed memory modules from OCZ were the only ones that worked under this configuration. So we included the results achieved under this configuration together with our maximum clock rate results.
The maximum clock rate test is exactly what it says: the higher clock rate we could achieve with the module being tested. On the results we included memory voltage, memory clock, CPU clock multiplier and CPU internal clock that were used to achieve the posted results.
Sandra Lite 2005.SR3 10.69
We used the Memory Bandwidth Benchmark module from Sandra to measure the maximum transfer rate that the memory was able to achieve. The result published is an arithmetic average from the Buffered Integer Stream and Buffered Float Stream results.
You can check the results on the graph below.
Memories – Maximum Clock Rate | Sandra – média – Int e Float | % |
Patriot PDC2G3500LLK (3-4-3-7) – 2.8 V – 270 MHz x 9 – 2.43 GHz | 6734 | +3.98 |
Corsair TWINX2048-3500LLPRO (3-4-3-8) – 2.8 V – 260 MHz x 9 – 2.34 GHz | 6281 | |
OCZ EL DDR PC-4000 Gold GX XTC (3-4-4-8) – 2.8 V – 280 MHz x 8 – 2.24 GHz | 6281 | -3.10 |
OCZ EL DDR PC-4000 Gold GX XTC (3-4-3-8) – 2.8 V – 275 MHz x 8 – 2.2 GHz | 6160 | -5.13 |
Since under this test the CPU clock varied according to the memory module being reviewed, it isn’t fair to make direct comparisons using the published results. Anyway, you can have an idea of the maximum clock rate you can achieve with each memory module included in our tests and also it is a good opportunity to see how a higher clock does not necessarily translates into a higher performance because of the latencies used.
Not taking the performance data into account, the reviewed memory modules by Corsair achieved the lowest clock rate among the memory modules we included in our review – including the Patriot ones, which also use memory chips from Infineon.
3DMark2001 SE Build 3.3.0
3DMark2001 SE simulates older games based on DirectX 8.1 and it is very sensitive to changes in the hardware configuration of the system.
You can check the results on the graph below.
Memories – Maximum Clock Rate | 3DMark2001 SE (Build3.3.0) | % |
Corsair TWINX2048-3500LLPRO (3-4-3-8) – 2.8 V – 260 MHz x 9 – 2.34 GHz | 23390 | |
OCZ EL DDR PC-4000 Gold GX XTC (3-4-4-8) – 2.8 V – 280 MHz x 8 – 2.24 GHz | 23209 | -0.78 |
Patriot PDC2G3500LLK (3-4-3-7) – 2.8 V – 270 MHz x 9 – 2.43 GHz | 23200 | -0.82 |
OCZ EL DDR PC-4000 Gold GX XTC (3-4-3-8) – 2.8 V – 275 MHz x 8 – 2.2 GHz | 23077 | -1.36 |
On 3DMark2001 SE all tested memories achieved the same performance level, even though different CPU clocks were used.
Quake III Arena 1.32
Even though this is an old game its importance comes from the fact its engine is used by several other games, like Jedi Knight II and Medal of Honor, and also because this game is very sensitive to any changes on the hardware configuration.
We used the demo four available on version 1.32 of Quake III to make our benchmarking with this game. We ran this demo three times at 1024x768x32 resolution and all image quality settings on their default configuration and we picked the middle value for our comparisons, i.e., we discarded the highest and the lowest values.
Check the results below.
Memories – Maximum Clock Rate | Quake III Arena – FPS | % |
Patriot PDC2G3500LLK (3-4-3-7) – 2.8 V – 270 MHz x 9 – 2.43 GHz | 505.8 | +4.70 |
Corsair TWINX2048-3500LLPRO (3-4-3-8) – 2.8 V – 260 MHz x 9 – 2.34 GHz | 483.1 | |
OCZ EL DDR PC-4000 Gold GX XTC (3-4-4-8) – 2.8 V – 280 MHz x 8 – 2.24 GHz | 471.9 | -2.37 |
OCZ EL DDR PC-4000 Gold GX XTC (3-4-3-8) – 2.8 V – 275 MHz x 8 – 2.2 GHz | 467.8 | -3.27 |
Since under this test the CPU clock varied according to the memory module being reviewed, it isn’t fair to make direct comparisons using the published results. Anyway, you can have an idea of the maximum clock rate you can achieve with each memory module included in our tests. The reviewed memory kit from Corsair presented the lowest overclocking potential among the memory modules included in our review.
Super Pi Mod 1.4
Super Pi is a benchmarking software created at Tokyo University which measures the time spent to calculate Pi up to 32 million decimal places. In our tests we calculated Pi with 1 million decimal places.
Check the results below.
Memories – Maximum Clock Rate | Super Pi Mod 1.4 | % |
Patriot PDC2G3500LLK (3-4-3-7) – 2.8 V – 270 MHz x 9 – 2.43 GHz | 35 | -4.36 |
Corsair TWINX2048-3500LLPRO (3-4-3-8) – 2,8 V – 260 MHz x 9 – 2.34 GHz | 36.594 | |
OCZ EL DDR PC-4000 Gold GX XTC (3-4-4-8) – 2.8 V – 280 MHz x 8 – 2.24 GHz | 37.672 | +2.86 |
OCZ EL DDR PC-4000 Gold GX XTC (3-4-3-8) – 2.8 V – 275 MHz x 8 – 2.2 GHz | 38.172 | +4.13 |
Since under this test the CPU clock varied according to the memory module being reviewed, it isn’t fair to make direct comparisons using the published results. Anyway, you can have an idea of the maximum clock rate you can achieve with each memory module included in our tests and how the increase on the clock rate reflects on the system performance.
[nextpage title=”Conclusions”]
TWINX2048-3500LLPRO memory kit was one of the first 2 GB kits to use very aggressive CAS latency (CAS 2) with a 219 MHz (DDR438) clock. Corsair scored a point here.
Infineon BE-5 chips used on Corsair module support a stable overclocking up to 260 MHz. We were able to configure over 260 MHz, but the system wasn’t stable for 3D applications, corrupting textures.
Patriot modules also use Infineon chips but from CE-5 revision, which have a higher overclocking capability. The good news is that newer kits from Corsair are also coming with Infineon CE-5 chips.
This kit from Corsair is perfect for the users that like low latency memories, being able to go up to 220 MHz with very low latencies (2-3-2-5) with an operating voltage of only 2.6 V.
With 18 LEDs blinking while it is working, this kit is perfect for users that have a case with transparent side window. But it is time for Corsair to make a facelift on their modules. It’s been more than two years with the same look.
Leave a Reply