260 1:1 best ram? OCZ EB?

[Clevor]

260 fsb. Grrrr. That's the problem with the 2.80-3.0Cs. 255-265 fsb is the no-man's land of ram. Too high to achieve 1:1, 2-2-2-5 with BH-5 even with massive amounts of VDIMM and too low to get PC4000 to stretch it's legs.

Maxvla, do you have BH-5? I'm not too sure 260, 1:1, and for sure 3-2-2-x (since you are pushing it) will blow away plain ole 260, 5:4, 2-2-2-5 with BH-5. I'd like to see either OCZ or an objective party show us some benches.

And were the EB sticks Anandtech tested off the shelf?

BTW, if you got a fast 875P board like the DFI or AOpen, I've found running PC450/4000 at 255-260, 1:1, 2.5-4-3-6, will give 3500+ unbuffered with DS sticks and 3DMark scores are on a par with running a 2.40C up around 300 fsb, 3:2, 2-2-2-5. You're talking about 3.55-3.6 gig both ways. And I'm getting around 3200 unbuffered on the 2.40C.

 

[Maxvla]

i have plenty of bh5. 260 1:1 2.5 3 2 5 with the EB ram should trounce 260 5:4 2225 with bh5.

the anandtech stuff as a review sample.

 

[Clevor]

And I wouldn't be surprised if dropping down to 250, 1:1, 2-2-2-5 with BH-5 (needs 3.2+ real VDIMM, not the powderpuff Abit stuff) won't beat 260, 1:1, 3-2-2-x too. You are only losing 150 mhz.

 

[GTX_SlotCar]

I doubt 250 at 2-2-2-5 would beat 260 at 3-2-2-x in an 875 or 865 motherboard. The two important timings (for 875/865 motherboards), the middle ones, are the same.

I had read on a couple major sites that the CAS latency means very little, even though all the memory manufacturers seem to make a big (marketing) deal about it. I'll probably end up with some memory rated at 3-4-4-8, so I tried it for myself with my Call of Duty timedemo and my current Geil PC3500 Platinum Ultra. In each case I ran the demo 5 times and averaged the score.

Abit AI7, P4 2.4C at 3243 (270 FSB) 5:4 mem ratio.
The memory timing will be given this way:
CAS latency / RAS# to CAS# delay / RAS# precharge / Cycle Time (Tras)
FPS is rounded to nearest 10th.
Call of Duty with 4xFSAA and 8xAF.
2-3-3-5 ..... 79.1 FPS
3-3-3-8 ..... 79.0 FPS
2-4-4-5 ..... 77.2 FPS
3-4-4-8 ..... 77.1 FPS
2-2-2-5 ..... 79.1 FPS (3:2 mem ratio)

With the results I got from jumping the CAS from 2 to 3 and the Cycle Time from 5 to 8, I saw no need to test the numbers in between them. If going from CL2 to CL3 along with Tras 5 to Tras 8 only made a 1/10th of 1% difference, then going only from CL2.5 to CL3 and Tras 5 to Tras 6 would make even less difference. I wouldn't even be able to measure the difference accurately.

As you can see, the articles were absolutely right, at least as far as games are concerned (and that's all I care about). Also, this test was for DDR memory on 875/865 boards and the results might not be true for other types.

To see what frame rate gains or losses I could expect from going to 3-4-4-8 memory but using a 1:1 ratio, I ran a test again, this time using a 3:2 ratio and 2-2-2-5 timings. The average frame rate was 79.1 again. It seems that the loss in mHz is made up by the tighter timings. This leads me to believe that going from my current 5:4 2-3-3-5 to my new memory at 1:1 3-4-4-8, I should not expect an increase in game frame rates.

If you're looking for bragging rights in Sandra '04 memory bandwidth, then the 1st & last timings, tCL and Tras, should give you a slight increase.

With 875/865 motherboards, the thing that's exciting about the new OCZ 3700EB is the tight tRCD and tRP timings at high frequencies. If this stuff is Double Sided as well, then I would expect an increase of up to 5 FPS in my Call of Duty timedemo.

Gary

 

[mrspec3]

And I wouldn't be surprised if dropping down to 250, 1:1, 2-2-2-5 with BH-5 (needs 3.2+ real VDIMM, not the powderpuff Abit stuff) won't beat 260, 1:1, 3-2-2-x too. You are only losing 150 mhz.

No way man. The EB @equal speed and timings with bh-5 sticks is upto 200MB/s faster in sandra unbuffered.

 

[Clevor]

GTX, your results are different from mine. Going by 3DMark scores, particularly 3DMark2000 and 2001 (2003 is mainly video card dependent) I get better scores at 270, 5:4, 2-2-2-5 versus 1:1, 3-4-4-7. I'm talking by 150-200 points. Results vary based on fsb.

You can't just look at sheer unbuffered bandwidth. 3DMark looks at CAS timings. I've even seen 3:2, 2-2-2-5 beat PC4000 stuff at 1:1, again, depending on fsb. It depends on how 3DMark takes to 3-2-2-.

On the other hand, at around 250-260, I've seen PC4000 stuff at 1:1 (2.5-4-3-6) beat BH-5 at 5:4, 2-2-2-5, on a fast 875P board. So life is not so simple.

 

[Edward2]

I have 2*256 OCZ PC3700 EL Gold memory running at 250 FSB 1:1 2-3-3-7 with 2.95 volts. I haven't tried running it at 260 FSB yet.

 

[Clevor]

How high the original Rev. 1.0 PC3700 Golds do is HIGHLY dependent on motherboard.

I have a set of sticks and they will not do 250, 1:1, 2-3-3-6 on a majority of the boards. They won't even do it on the Epox boards where I tried up to 3.2 VDIMM. In some instances I could do 250, 1:1, 2-3-3-6 Memtest clean for Sandra screenies, but they were not 3D stable.

Best I could do is 248, 1:1, 2-3-3-6 on the Asus boards with max 2.85 VDIMM. I imagine with the VDIMM mod, 250, 1:1 should be no problem. I don't think even OCZ claims 260, 1:1.

This ram will not run at 5:4 or 3:2 (sometimes not even 1:1) on the Abit, DFI, and AOpen boards, boards with low VDIMM and who could benefit from the ram the most! In fact, only boards my sticks will run well on is the Asus and Epox boards, which have pretty decent VDIMM anyway and where BH-5 runs equally well so naturally I opt for that and their 2-2-2-5 timings.

 

[Edward2]

I forgot to say that I am running the OCZ PC3700 memory on my new Asus P4C800-ED mobo. Obviously, I have done the Vdimm mod to get 2.95 volts. It appears to be 100% stable at 250 FSB. I have run Prime95, Super-PI, PiFast, and 3dmark01 without any problems.

I did a quick check at 260 FSB, and it was stable enough to do a Sandra unbuffered memory test, but it crashed while running the Super-PI 1M calculation. I'm not sure if it is the memory or my 3.0C CPU that generated the errors. I will try running Memtest at 260 tomorrow if I get a chance.

 

[Clevor]

For a quick-and-dirty test, run the Sandra Cache-and-Memory test. It's a fairly long test compared to the memory test, and will fail if your ram is borderline.

Nowadays, Memtest passing in it's entirety means nothing, as the 'C' CPUs may not even boot Windows.

 

[wannaoc]

I would go with the EB but don't waste the extra cash on getting the PC3700. The PC3500 is the same sticks but with less voltage. The PC3500 is rated at 2.6V while the PC3700 is rated at 2.8V, you can up the volts as easy as OCZ and save money. I would say they are your best bet though.

 

[GTX_SlotCar]

GTX, your results are different from mine.....I get better scores at 270, 5:4, 2-2-2-5 versus 1:1, 3-4-4-7. I'm talking by 150-200 points.

We're on the same page, Clevor. I agree with you. My test at 5:4 was with 2-3-3-5 timing, not 2-2-2-5. If my Geil PC3500 would run x-2-2-x at 270, 5:4 (216 mHz), I'd use it for sure. Especially since it would also let me use PAT. I can't seem to run 270 fsb, 1:1 with PAT enabled.
My Geil 3500 won't run 270, 1:1 at any timing. I did the 3:2 test at 2-2-2-5 to compare it to the 5:4 test at 2-3-3-5, and applied that difference to a hypothetical 1:1 at 3-4-4-8. My conclusion was that I'd get about the same frame rates either way. Tighter timings offset the lower ratios.

I did get my single stick of Adata Vitesta PC4000 from newegg. It runs up to 285, 1:1 (my 2.4C won't go any higher) at 2.8v, which is the only voltage I've tried. Not bad for 61 bucks.

Gary

 

[jackson vile]

When is NewEgg going to get the OCZ 3500 EB in is that a better choice than 3200 EB is you are looking to obtain 247 and up?