OCZ, Corsair, Mushkin, Geil?

[guillaume]

Even by reading a lot of threads, it's kind of hard to determine what is the best o/c memory around. The polls favor Corsair, but are they the best for overclocking? What would you suggest? I plan to use it on a AI7 w/ 2.4C or 2.6C

Thanks! :)

[^wargod]

i would say Mushkin, i love mine. i had problems with the OCZ PC3500 plat in the past, had to RMA it. my Mushkin black lvl2 PC3500 rocks on the other hand.

[Speed_Mechanic2]

There is no best for everything memory :)

What do you want your memory to do? Run high speeds at a 1:1 FSB:MEM ratio with high latencies. Or run 80% of the FSB in a 5:4 FSB:MEM ratio with tight timings?

For high speed (260+), high latency memory, you would want something with Hynix 'D43' chips. These are used in most PC4000 and higher. Which maker is best depends entirely on circumstance, and all can deliver. There isn't a best memory maker - only a favorite.

For semi-high speed (210-220), tight timing memory, you would ideally want something with Winbond BH-5 chips. But these are hard to find, and even harder to find at a good price. Memory using these chips can be found in the "BH-5 pricelist" sticky. Again, there is no best company - only favorites.

[guillaume]

That's very interesting. I guess i'm not that savvy when it comes to memory :) So, what would make my system run faster: high speed + high latency or semi speed + tight timing?:D

[DaRky4YoU]

if money isnt an issue id say mushkin

but i mgetting some nice kingston hyperx pc 3000 2x512 mb for 200 bux :)

[Code 4]

My Geil PC4000 Ultra 2x512mb will run up to 522mhz @ 2.85V on a P4C800-E. That is the highest volts you can go without Vdimm Mod. However, after Vdimm mod and 3.05Volts it is 534mhz @ 2.5 4 4 7 (current CPU is stuck @ 266FSB so I have not been able to find the max on this ram). I have had Corsair, Mushkin, Samsung and some others. My 2 cents is that for a pretty cheap price you can get some decent Geil sticks that will do all you need them to and more.

[Sirius]

The Cosair Xms4000 is a good Ocer. You can get high FSB at 1:1 but you can't get tight timings.

[bumbi]

Well I don't think your gonna go over FSB220 so I say Mushkin Black Level II PC3500. PC3700+ are more fore high level and extream overclockers.

BTW the PC3500 will go at least 1:1 @ FSB216 with SDP timeings

[Csybe]

GeIL ;)

[Speed_Mechanic2]

How much do you want to spend?

[Code 4]

Originally posted by bumbi
Well I don't think your gonna go over FSB220 so I say Mushkin Black Level II PC3500. PC3700+ are more fore high level and extream overclockers.

BTW the PC3500 will go at least 1:1 @ FSB216 with SDP timeings

A 2.4 or 2.6 at only 220FSB would be a waste of time. Try more like 260 at a minimum and closer to 300 as a max. Go for PC4000 a lil looser ram timings, but higher FSB = more mem bandwidth in the end.

[Speed_Mechanic2]

Originally posted by Code 4


A 2.4 or 2.6 at only 220FSB would be a waste of time. Try more like 260 at a minimum and closer to 300 as a max. Go for PC4000 a lil looser ram timings, but higher FSB = more mem bandwidth in the end.

That's what the 5:4 FSB:MEM ratio is for, which Intel implemented quite nicely (with regards to low performance drop off from 1:1) in their i875/i865 chipsets.

[Code 4]

Originally posted by Speed_Mechanic2


That's what the 5:4 FSB:MEM ratio is for, which Intel implemented quite nicely (with regards to low performance drop off from 1:1) in their i875/i865 chipsets.

1:1 @ 250FSB benches faster than 5:4 @ 250FSB. I see no drop in performance when I run 1:1 @ a higher FSB and it blows the 5:4 out of the water. Hence the purpose of memory that clocks higher than the default 400mhz of a 200FSB. Why is it that people are spending more money on faster ram if they could just use the ratio and slower ram yeilding the same performance?

[Speed_Mechanic2]

Originally posted by Code 4


1:1 @ 250FSB benches faster than 5:4 @ 250FSB. I see no drop in performance when I run 1:1 @ a higher FSB and it blows the 5:4 out of the water. Hence the purpose of memory that clocks higher than the default 400mhz of a 200FSB. Why is it that people are spending more money on faster ram if they could just use the ratio and slower ram yeilding the same performance?

Where, with what settings, and what benches does a 250:250 high latency bench faster then 250:200 with tight timings. They're many ways benches can go either way. And many settings that will do it also. Just saying "1:1 @ 250FSB benches faster than 5:4 @ 250FSB" is a blind statement.

[Code 4]

Unfortunately I do not have any LL ram to prove my point with benches and I do not have screen shots from my old Corsair LL. In my case, which I am sure is the same for everyone; I had higher sandra/mbench/3DMark at the Same FSB @ 1:1 with looser timings (2336) then with 5:4 or 3:2 (2226). If this is or is not the case someone with Mushkin or Corsair run some tests and display your results. I would like to see the results and if they differ from my experience.

[Speed_Mechanic2]

Originally posted by Code 4
Unfortunately I do not have any LL ram to prove my point with benches and I do not have screen shots from my old Corsair LL. In my case, which I am sure is the same for everyone; I had higher sandra/mbench/3DMark at the Same FSB @ 1:1 with looser timings (2336) then with 5:4 or 3:2 (2226). If this is or is not the case someone with Mushkin or Corsair run some tests and display your results. I would like to see the results and if they differ from my experience.

It depends on a myriad of situations. There is no one best way to go. High speed memory with high latency has it's benefits and downfalls, as does lower-speed memory with tight timings. Certain applications will show it going both ways. For the most part, unless you go searching for cheap Hynix BT-D43 chip memory, PC4x00 is expansive. It's more costy then what Winbond BH-5 chip memory use to be. Now that the memory that use to containt BH-5 chips is going Out-of-Stock everywhere, the price differance is about the same.

250MHz with 2-3-3 timings is not something that the average memory will be able to do. It's not something that you can go out and buy new and hope to achieve without getting lucky or increasing vDIMM past 3V. Therefore it's a bit unfair to compare 250:250 (w/ 2-3-3 timings) vs. 250:200 (w/ 2-2-2 timings), as it's not something you will usually see. When tRCD is increased to 4, which is far more likely the case with PC4x00, then the performance differance shows in many benchmarks.

[am8soccer]

At first, I was very disappointed with my OCZ 3500 Platinums which would not run at rated speeds and timings so i RMA'd them. Now the new sticks are doing great. So for I was stable at 226 2-2-2-6 2.85v and 233 2-3-2-6 2.85v. I wish I could push them more but i havent gotten around to doing a vdimm mod. However I think these modules are sold out now.

[Code 4]

Originally posted by Speed_Mechanic2

250MHz with 2-3-3 timings is not something that the average memory will be able to do. It's not something that you can go out and buy new and hope to achieve without getting lucky or increasing vDIMM past 3V. Therefore it's a bit unfair to compare 250:250 (w/ 2-3-3 timings) vs. 250:200 (w/ 2-2-2 timings), as it's not something you will usually see. When tRCD is increased to 4, which is far more likely the case with PC4x00, then the performance differance shows in many benchmarks.

I am not talking about 250. I did not state but the Corsair I reffered to was PC3200. Thus I was talking more like 220FSB 1:1 and 5:4. No Vdimm just 2.85V for plain bandwidth and performance testing. I will guarantee that you will see faster ram @ 2336 1:1 than you would with 220 2226 5:4. :rolleyes:

[Speed_Mechanic2]

Originally posted by Code 4
I am not talking about 250. I did not state but the Corsair I reffered to was PC3200. Thus I was talking more like 220FSB 1:1 and 5:4. No Vdimm just 2.85V for plain bandwidth and performance testing. I will guarantee that you will see faster ram @ 2336 1:1 than you would with 220 2226 5:4. :rolleyes:

Of course. I never said otherwise. I was referring to your statement, "1:1 @ 250FSB benches faster than 5:4 @ 250FSB."

But 1:1 w/ 2-2-2 timings will be faster then 1:1 w/ 2-3-3 timings. And the speeds with which most people will achieve with 2-3-3 timings can, in most cases, be duplicated by BH-5 memory with 2-2-2 timings. That's not questioned here though. High speed, high latency memory that, in most cases, will fail to run 250+ w/ a tRCD set below 4, and that's what the 5:4 w/ tight-timings should be compared to.

250:250 (FSB:MEM) w/ a tRCD set to 4 is not faster (in general) to 250:200 (FSB:MEM) w/ 2-2-2 timings.

[Code 4]

IC what you are saying now. I guess I would need to see what the difference would be in my current setup with 5:4 and tight timings. OK, so back to the idea of what ram Guillaume should get. IMO if you are going to at least do 250FSB(a 2.4 or 2.6 will do 250 NP) or a little more I would go with PC4000. You can at least run 1:1 up to their within the stock voltage of your board, hopefully. I like Geil due to cost and perfromance. Mine will do 522mhz @ 2.85V (highest stock of P4C800-E) If you have big bucks go for something like OCZ Gold.

[Shadco]

Originally posted by Speed_Mechanic2


Where, with what settings, and what benches does a 250:250 high latency bench faster then 250:200 with tight timings. They're many ways benches can go either way. And many settings that will do it also. Just saying "1:1 @ 250FSB benches faster than 5:4 @ 250FSB" is a blind statement.

You might wanna look at these.

The first one is using Twinmos 3200 w/BH5's 255 fsb 5:4 ratio 2 2 2 5

The second one is OCZ PC3700 gold rev 2 250 fsb 1:1 ratio 2.5 3 3 6

PCmark mem bench and Super Pi show pretty much the same diff.

http://www.pbase.com/image/25423443/original.jpg

http://www.pbase.com/image/25420932/original.jpg

any questions?

Shad

[Code 4]

Thanks for the benches Shad :clap: Wow 1:1 with looser timings faster? I never would have guessed :rolleyes:

[Sirius]

Nice benches. Those scores make me would to get the OCZ rev.2 3700 with tight timing and drop this xms 4000.

[Speed_Mechanic2]

Originally posted by Shadco


You might wanna look at these.

The first one is using Twinmos 3200 w/BH5's 255 fsb 5:4 ratio 2 2 2 5

The second one is OCZ PC3700 gold rev 2 250 fsb 1:1 ratio 2.5 3 3 6

PCmark mem bench and Super Pi show pretty much the same diff.

http://www.pbase.com/image/25423443/original.jpg

http://www.pbase.com/image/25420932/original.jpg

any questions?

Shad

250MHz w/ 2.5-3-3 timings is very good :)

OCZ's PC3700 Gold Rev. 2 is a good product - especially suited for someone with an Intel 2.8C or 3.0C. If guillaume was getting one of these processors and didn't plan to go extreme on the cooling, then I would recommend it for him. For a 2.4C/2.6C, there are other, possibly better, options.

:)

edit: Bandwidth is but isn't affected by latency. However latency does have an affect on bandwidth benchmarks. But the effect on bandwidth benchmarks by latency is less then it is in what are called "real-world" tests. These are benchmarks (or non-benchmarks) based on applications, using memory the way according to how the application uses memory. Differant applications will be affected by bandwidth in differant ways. This and the fact that buffered bandwidth tests are misleading, lead to a comparison between memory using just SANDRA unconclusive at the best.

Unbuffered SANDRA tests, and 3DMark01's Lobby (High) and Car (High) Tests are good basic telling points for memory performance. But they don't answer which is best.

[Shadco]

Originally posted by Speed_Mechanic2


250MHz w/ 2.5-3-3 timings is very good :)

OCZ's PC3700 Gold Rev. 2 is a good product - especially suited for someone with an Intel 2.8C or 3.0C. If guillaume was getting one of these processors and didn't plan to go extreme on the cooling, then I would recommend it for him. For a 2.4C/2.6C, there are other, possibly better, options.

:)

edit: Bandwidth is but isn't affected by latency. However latency does have an affect on bandwidth benchmarks. But the effect on bandwidth benchmarks by latency is less then it is in what are called "real-world" tests. These are benchmarks (or non-benchmarks) based on applications, using memory the way according to how the application uses memory. Differant applications will be affected by bandwidth in differant ways. This and the fact that buffered bandwidth tests are misleading, lead to a comparison between memory using just SANDRA unconclusive at the best.

Unbuffered SANDRA tests, and 3DMark01's Lobby (High) and Car (High) Tests are good basic telling points for memory performance. But they don't answer which is best.

Well I'm not gonna bother sticking the Twinmos back in. This is a Photoshop system and I focused on building a well balanced system where Disk I/O is just about as important as CPU/Memory.

but heres an unbuffered run from the OCZ which I have paired with a 2.6C btw.


http://www.pbase.com/image/25420933/original.jpg

Overall I'm quite pleased with it. I found the trade off to run 1:1 to be worth it.

Shad

[Speed_Mechanic2]

Originally posted by Shadco


Well I'm not gonna bother sticking the Twinmos back in. This is a Photoshop system and I focused on building a well balanced system where Disk I/O is just about as important as CPU/Memory.

but heres an unbuffered run from the OCZ which I have paired with a 2.6C btw.

Overall I'm quite pleased with it. I found the trade off to run 1:1 to be worth it.

Shad

Nice scores, what CPU you got going with it?

[Shadco]

Originally posted by Speed_Mechanic2


Nice scores, what CPU you got going with it?

P4 2.6C Malay sl6ws batch L342A534 packdate 11/15/03




http://www.pbase.com/image/25418712/large.jpg

Shad