OCZ PC3700 Gold rev2 [lots of screenies]

[eva2000]

What better to do on Boxing day morning than some benchies

Notes

1. OCZ PC3700 Gold rev2 was able to do 2-3-2-5/6 at DDR400 and 2.5-3-3-7 at DDR466 at low vdimm of 2.7v on AI7 and do 3-3-3-8 at DDR500 on AI7 in dimm slot 2+4 at 2.9v in bios (uGuru reports 2.85v vdimm)

2. This is 256MB single sided memory so bare that in mind when you see memory bandwidth compared to double sided rev1 OCZ PC3700 Golds

3. Memory was very picky about GAT settings didn't like Turbo or anything at all from what i tested so far.. need a bit more testing to see.

4. Since 2 people have commented about this, cpuz has bug reporting 1:1 on AI7 when NB Strap is set to 667fsb it reverses the 1:1 and 5:4 so 5:4 shown in cpuz is 1:1. The 5:4 in cpuz is in fact 1:1 at 233mhz



2x 256MB OCZ PC3700 Gold rev2

2.4C Q322A285
Abit AI7 14bios
128MB Gigacube 9600PRO stock
80GB WD 800JB
550W Antec TC psu

vcore = 1.525v
vdimm = 2.7v
vAGP = 1.65v

GA = AUTO
Refresh Cycle Time = Enhanced
Read Delay = AUTO
Read Delay Adjust = Disabled
Command Per Clock = Disabed

NB Strap = 667FSB

2x 256MB OCZ PC3700 Gold rev2
1:1 @ 233mhz 2.5-3-3-7

Sandra 2004 STD

 

[eva2000]

 

[eva2000]

 

[mrspec3]

Nice stuff Eva....you get an xmas present from someone

 

[waddy]

OCZ is damn fine memory, some nice memory bandwidth you have there.

Santa gave me some OCZ to, gave me 1000mb/s extra in Si Soft

 

[larva]

To be honest these scores serve to show just how important all those bios optimizations and tight timings you had to forsake for stability's sake are. My BH5's are performing at a higher level (2999 unbuffered sandra, 10,700 PCMark2002) even though they are only operating at 380MHz. And of course the real question is--What do they do in 5:4? They are seriously limiting your overclock in 1:1 mode.

Not a whole lot of people seem to be aware of this, but the simple fact is running the ram much above 400MHz on 865/875 systems nearly always entails compromising the very effective bios and chipset optimizations in order to do so. Since bandwidth is pretty abundant supply at 400MHz the degregation in the latency behavior that results generally overweighs the gains brought by the increase in bandwidth. This is not an iron rule, but a cleary evident trend.

The only systems I have seen clearly avoid this pitfall are again, BH5 equipped, and use voltage to preserve what is so right about running high levels of PAT and tight timings. If you can scale the clock rate up to 440+ MHz without de-tuning the bios settings you can actually put more area under the memory performance curve, not just push it around. Other ram types running in excess of 400MHz lose through latency what they gain through bandwidth.

 

[eva2000]

5:4 benches coming soon

The whole point is BH-5 is running out of supply. This rev2 memory is capable of DDR500 at 3-3-3-8 at ~2.8-2.85v depending on board and is a good alternative to the Hynix D43 2.5-4-3-8 to 3-4-4-8 timings at >= 250mhz. At the same time this rev2 can do relatively tight 2-3-2-5/6 timings at DDR400 at 2.7v

Unfortunately on this AI7 at > 233mhz i have to use NB Strap of 800FSB which disables PAT and the scores drop. This one limitation of the AI7 board. Will test on other boards.

vcore = 1.675v
vdimm = 2.9v (windows 2.85-2.87v)
vAGP = 1.65v

GA = AUTO
Refresh Cycle Time = Enhanced
Read Delay = AUTO
Read Delay Adjust = Disabled
Command Per Clock = Disabed

NB Strap = 800FSB

2x 256MB OCZ PC3700 Gold rev2
1:1 @ 250mhz 3-3-3-8

Sandra 2004 STD
Buffered and Unbuffered

 

[eva2000]

 

[Psykoikonov]

Some nice RAM you got there. I know mine works very well (check sig)

Psyko

 

[larva]

5:4 benches coming soon

The whole point is BH-5 is running out of supply. This rev2 memory is capable of DDR500 at 3-3-3-8 at ~2.8-2.85v depending on board and is a good alternative to the Hynix D43 2.5-4-3-8 to 3-4-4-8 timings at >= 250mhz. At the same time this rev2 can do relatively tight 2-3-2-5/6 timings at DDR400 at 2.7v

BH5 is indeed being made again, and can be bought in countless ways, including Mushkin PC3200 or 3500 level 2, OCZ 3500 Plat. LE, Kingston HyperX3200 or 3000, or Corsair XMS3500. The new production is not quite as good as the stuff made in 2002, but it's still the best memory available for this application. And if 2-3-2-6 at 2.7V appeals to you, you can achieve it a whole lot more cheaply with CH5.

Unfortunately on this AI7 at > 233mhz i have to use NB Strap of 800FSB which disables PAT and the scores drop. This one limitation of the AI7 board. Will test on other boards.

vcore = 1.675v
vdimm = 2.9v (windows 2.85-2.87v)
vAGP = 1.65v

GA = AUTO
Refresh Cycle Time = Enhanced
Read Delay = AUTO
Read Delay Adjust = Disabled
Command Per Clock = Disabed

NB Strap = 800FSB

2x 256MB OCZ PC3700 Gold rev2
1:1 @ 250mhz 3-3-3-8

As the AI7 is a 865pe product, it has no PAT. GAT is the only feature at work here, and it is disabled when you set the first option to auto. Turbo, SR, or F1 enable varrying degrees of GAT, but as you know your ram is not stable when you enable it. This, along with the loose timings, is responsible for the poor performance, and as such makes it difficult to recommend this 3700 over the OCZ 3500 Plat. LE or other BH5 modules.

The bug here is why the performance mode reports "enabled" in your first configuration, not that it reports "disabled" in the second. CPU-z is entirely confused by late model Abit 865 bioses though, so these factors need to be experimentally determined though performance testing to know if CPU-z is at all correct on these points.

 

[eva2000]

Larva, production of BH-5 stopped in Sept 2003, whatever is available now was from Sept 2003 and prior.

I have 6x 512MB Corsair XMS3500C2

batch numbers x quantity = max oc at 2-2-2-6 and @ vdimm 3.2 - 3.3v

024117 x 2 = 240mhz
0331125 x 1 = 245mhz
0331125 x 1 = 242mhz
0331125 x 1 = 240mhz
0337061 x 1 = 245mhz

2x 512MB OCZ PC3500 Platinum Limited edition max is 240mhz 2-2-2-6 at 3.2v

while nice and all, all BH-5 need 3.2-3.3v vdimm to run at high frequencies

here in Australia, alot of places are out of stock of XMS3500C2 now, and OCZ PC3500 Plat Limited Edition is already EOL


When i mean PAT i should of said PAT' like performance is gone/disabled. And remember above benchies are with single sided 256MB sticks... with double sided sticks performance is alot better

According to ctiaw PAT is disabled @ 250mhz 3-3-3-8 with 800FSB NB Strap and

667fsb NB Strap enabled at 233mhz 2.5-3-3-7

maybe ctiaw AND cpuz are buggy or incorrect ?

The whole point is to look beyond these particular benchies as it's board, and bios dependent, and look at the rev2's achieved memory speed, timings and use of LOW vdimm at up to 250mhz - put 512MB sticks in on a P4C800-E unmodded at 2.85v and benchies will be alot better

 

[larva]

The big point is, you don't have to run BH5 at 240+MHz to outperform the stuff you've shown, and as such don't have to put 3.2V to them. They will run as much as 225MHz at 2-2-2-5 with PAT/GAT on 2.85V, eclispsing the benchmark and application performance you've demonstrated. And if you want to pump up the voltage, the gap just widens.

You show the screen shot of cpu-z mis-reporting the ratio and memory clock and then act like it would be shocking for it and/or ctlaw to be wrong on these points?