Test Setup

Test System
CPU:	Intel E6300 Conroe 1.8 GHz, 2 MB Cache
Motherboard:	ASUS P5K3 Deluxe BIOS 0602
Video Card:	PowerColor X800XL Pro 16 PCI-E
Harddisk:	Samsung P80 80 GB
Power Supply:	Ultra V-Power 450W
Software:	Windows XP SP2, Catalyst 7.6

Performance & Overclocking

Even though the Winchip DDR3 is meant to run at 1600 MHz, CL-8-8-8-15 using 1.95 - 2.05V, the SPD timings on the memory only go up to 1422 MHz with CL 8-8-8-27. Winchip simply takes the memory ICs and sorts out the ones, which manage 1600 at CL8-8-8-15. To actually get the memory running at this speed, the mainboard FSB needs to be set at 400 MHz. This translates to 2.8 GHz or 940 MHz overclock on the Core 2 Duo E6300. Intel is currently in the process of releasing 1333 MHz based CPUs which run at 333 MHz FSB, these would not require such a high overclock to reach the magical 400 MHz needed for the 1:2 ratio to result in 1600 MHz. After running the memory at the rated speed, the timings were loosend to CL10-10-10-30, but the memory simply would not budge any further at 2.05V. The E6300 used in this test is easily capable of running well over 2.8 GHz, thus it is the memory that is the limiting factor here.

The second step was to find the minimum CL the DDR3 would run at. The ASUS P5K3 can be set all the way down to CL5-3-3, but Winchip's memory was not stable at this setting, no matter how much voltage was applied. Loosening the timings to CL5-4-4-12 resulted in a booting system even at the standard 1.5V and 800 MHz. The success did not pay off, as the memory would become unstable with slightly higher clocks and would not be stable beyond 842 MHz, no matter how much voltage was applied.

Then, the Winchip DDR3 modules were set to CL5-5-5-15, a very popular timing on high-end DDR2 memory these days. At this rating the memory actually managed a very respectable 1066 Mhz with 1.6V with no problems. Further raising the speed did not succeed beyond 1072 MHz, no matter how much voltage was applied. There is a pattern to be seen here. The memory works at tight timings but simply cannot get past a certain point.

The next step meant loosening the memory to CL6-6-6-15. This resulted in a much broader overclock and managed 1268 MHz at 1.7V. Once again, this was the end of the line for the memory at CL 6. CL7-7-7-15 raised that bar one more, just barely missing the 1400 MHz mark with 1394 MHz at, once again 1.7V. Raising the voltage further did result in a few extra MHz, but the system was not completely stable and would freeze or restart during benchmarking. The memory only manages the get beyond that with the advertised CL8-8-8-15. At this it runs all the way up to the advertised speed, but not even 6 MHz beyond that.

Winchip DDR3 1600 MHz 1 GB Kit
CPU Clock & Memory Ratio	Memory Speed	Memory Timings	Everest Read	Everest Write	Everest Latency	Quake 3 Timedemo	3DMark 2001SE	SuperPi Mod 1M
7 x 266 2:3	400 MHz	5-4-4-12 1.5V	6281 MB/s	4846 MB/s	80.1 ns	533.7 fps	24361	29.00 s
7 x 281 2:3	421 MHz	5-4-4-12 1.5V	6599 MB/s	5056 MB/s	75.4 ns	561.0 fps	25330	27.53 s
7 x 266 2:3	400 MHz	5-5-5-15 1.5V	6246 MB/s	4820 MB/s	81.3 ns	530.3 fps	24277	29.09 s
7 x 355 2:3	533 MHz	5-5-5-15 1.6V	8004 MB/s	5628 MB/s	63.7 ns	614.0 fps	26504	25.14 s
7 x 289 1:2	578 MHz	6-6-6-15 1.5V	7356 MB/s	5251 MB/s	70.3 ns	578.2 fps	25690	26.61 s
7 x 308 1:2	616 MHz	6-6-6-15 1.6V	7839 MB/s	5536 MB/s	66.0 ns	617.6 fps	26529	24.96 s
7 x 308 1:2	634 MHz	6-6-6-15 1.7V	7783 MB/s	5745 MB/s	64.2 ns	630.6 fps	26907	24.3 s
7 x 306 1:2	612 MHz	7-7-7-15 1.5V	7741 MB/s	5546 MB/s	64.2 ns	609.3 fps	26591	25.46 s
7 x 333 1:2	667 MHz	7-7-7-15 1.6V	8125 MB/s	6005 MB/s	64.8 ns	660.7 fps	27704	23.17 s
7 x 347 1:2	694 MHz	7-7-7-15 1.7V	8428 MB/s	6187 MB/s	62.4 ns	688.9 fps	28010	22.31 s
7 x 355 1:2	710 MHz	8-8-8-15 1.5V	8300 MB/s	6419 MB/s	64.0 ns	689.3 fps	28219	21.92 s
7 x 376 1:2	752 MHz	8-8-8-15 1.7V	8804 MB/s	6716 MB/s	60.4 ns	730.4 fps	28967	20.67 s
7 x 400 1:2	800 MHz	8-8-8-15 2.05V	9282 MB/s	7043 MB/s	63.0 ns	775.4 fps	29280	19.45 s

This graph shows you at a quick glance which clock frequencies you can expect at which voltage. Please note that each individual module overclocks different, the results here can only be seen as an indicator of performance.