EPoX EP-9U1697 GLI Motherboard Review – SLI Killer?

Our tests attempt to determine the following three overclocking limits: motherboard, memory, and ideal.

 

To figure out the maximum overclock for the motherboard, we drop memory timings, frequencies, and the CPU multiplier, and throttle up the system bus until it can no longer run our suite of tests (the same tests used throughout this article, takes about 2 hours).

 

For memory overclocking, we set the memory multiplier to a 1:1 ratio with the system bus, set the CPU multiplier lower, and then bump the system bus until the memory can not maintain a stable system.

 

Finally, the ideal overclock test attempts to find a good combination of bus frequency, CPU and memory overclock settings.

 

CPU  Voltage	1.65V (+0.15 V)
Memory Voltage	2.7V (+ 0.2 V)
Chipset Voltage	1.20V (+0.1 V)
Memory Timings	3-2-2-2T

 

Motherboard Overclock

With a CPU multiplier of 6x, 800MHz HTT, and lax memory timings, we proceeded to pump up the system bus. I started at 230 MHz, and incremented by 10 MHz doing a 8MB SuperPi calculation to determine stability before proceeding.

CPU Multiplier	6x
HT Bus	600 MHz
Memory Speed	100 MHz (2:1 Bus:Mem)

Once past 290 MHz, I started to get a little nervous – high bus frequencies can sometimes have unpredictable effects if overclocking fails. Pushing forward… 300 … 310 … 315 …. 357 MHz!!!

 

At 357 MHz, the system successfully ran all our benchmarks without an issue. While 2 hours of testing isn’t enough to really confirm if it’s 100% stable at that frequency, it does give a good ballpark number.

CPU Multiplier	Overclock
6x	357 MHz
8x	320 MHz
10x	265 MHz

Increasing the CPU multiplier to 8x, the overclock drops to 320 MHz. At 10x, it’s 265 MHz.

 

All these numbers are very impressive.

 

CPU Multiplier	6x
HT Frequency	800 MHz, 600 MHz
Memory Bus	200 MHz (1:1 Bus:Mem)
Overclock @ 800 MHz HT	214 MHz
Overclock @ 600 MHz HT	215 MHz

Overclocking the memory on the EP-9U1697 fell short of expectations. In past reviews, we have seen this pair of Corsair XMS RAM capable of 230 – 235 MHz at 2.5 CAS settings. So we should expect similar on the EP-9U1697 right? I was only able to get roughly 214 – 215 MHz before SuperPi gave errors.

 

Changing the HTT, CPU, and memory multipliers didn’t change anything – 215 MHz was the absolute max. A bit of a let-down after the extraordinary overclocking of the motherboard. At bus speeds higher than 215 MHz, SuperPi would give errors.

At 600 MHz HT, the results were the same: memory overclock was limited to 215 MHz.

 

Other reviewers of the EP-9U1697 did not have problems with memory overclocking, so I’m left thinking it’s a problem with my configuration.

 

Ideal Overclock

The “ideal” overclock is a bit subjective. For me, the ideal overclock is the highest attainable CPU frequency with the RAM in a 1:1 ratio if possible. From the memory overclock results above, I knew that getting the system to overclock above 215 MHz was going to be a challenge.

CPU Multiplier	12x, 11x
HT Frequency	1 GHz, 800 MHz
Memory Bus	200 MHz, 166 MHz

The highest overclock at 12x was 215 MHz (as predicted) for a resulting CPU speed of 2580 MHz, and the memory at 215MHz. Anything higher and I would get errors while running SuperPi. Dropping the HT to 800MHz did not help achieve a higher overclock.

At 11x CPU, 1 GHz HT, and the memory at 166MHz, I managed to get 238 MHz before 3DMark 06 started to stutter and struggle. From 238 MHz to 245 MHz, the system still ran and completed majority of the benchmarks, but 3DMark 06 kept aborting.

Here are some results of the benchmarks I ran during this test:

	EPoX EP-9U1697 GLI					EPoX  EP-9NPA+ SLI
Overclock (MHz)	200	215	238	240		200
CPU Multiplier	12	12	11	11		12
CPU Frequency (MHz)	2400	2600	2621	2643		2400
Memory Bus Speed (MHz)	200	200	166	166		200
Memory Frequency (MHz)	200	215	198	199		200
HT Frequency (MHz)	1000	1000	1000	800		1000
SiSoft Sandra
CPU – ALU	10719	11470	9821	10458		10775
CPU – iSSE3	4871	5196	4458	4736		4872
Memory – Float	4918	5257	3871	4129		4814
Memory – Int	4919	5255	3869	4124		4815
PCMark 05	3769	3833	3803	3541		3846
3DMark 06 (Single / SLI)	1302 / 2299	1301 / 2330	1322 / 2332	1306 / 2282		1521 / 2724
SM 2.0 (Single / SLI)	491/ 894	470 / 906	487 / 904	470 / 855		591 / 1166
HDR / SM 3.0 (Single / SLI)	476 / 942	490 / 940	491 / 942	493 / 942		557 / 1079
CPU (Single / SLI)	930 / 928	996 / 984	989 / 994	1003 / 1004		923 / 934
Call of Duty 2
Min (Single / SLI)	24 / 35	23 / 36	23 / 34	23 / 34		28 / 37
Avg (Single / SLI)	34 / 45.4	34.7 / 54.3	33.9 / 43.9	32 / 56.4		42 / 46.7
Quake 4
Min (Single / SLI)	38 / 73	38 / 73	37 / 71	34 / 65		40 / 60
Avg (Single / SLI)	87.3 / 125	87.9 / 123.4	85.5 / 120.9	79.2 / 110.5		93 / 124.27

Clearly the best results were achieved when the CPU was using a 12x multiplier and the memory and bus were synchronous. A 10 FPS boost in Call of Duty when in SLI mode at 12×215 compared to 12×200 is definitely worth attempting for an overclocker.