How necessary is good RAM to overclock?

[DanRydell]

Will I be able to overclock at all if I'm using value RAM with a 3000+ 90nm and a decent mobo? Would 2-300 Mhz be too much to expect?

 

[Andres3605]

if you plan to overclock using 1:1 you will have no change but you can still using the dividers, or getting some quality ram

 

[DanRydell]

What do you mean? Is a non 1:1 ratio not even worth doing?

 

[Super6]

Using name brand value PC3200 should work ok with relaxed timings. I'm beginning to believe that some of the 939 mobos are a little particular when it comes to RAM type so your results may vary.

Super6

 

[Duvie]

From what I think he meant is there is NO CHANCE not no change.... Timings can be loosened up to cas3 and the ratios can have you running pc2700 to pc3000 levels. I will find out this week with my new setup but I believe their is a divider that at 300fsb could run that ram at 200x2 for pc3200...Other settings could be factor but it makes logical sense...

 

[BonzaiDuck]

Duvie

Per your remark on another thread . . . "Benchmarks show Ghz more important than RAM speed."

I think this would be more true with a larger L2/L3(if any) cache size than with smaller.

 

[BonzaiDuck]

Anyway, it's probably easier to OC with higher quality RAM. The other side of it -- without OC'ing, you get a little bonus with higher-quality lower-latency RAM. But that bonus is more than offset by looser timings and faster processor, FSB and memory speeds.

 

[Duvie]

I saw this with 512kb l2 cache Northwoods with no l3 cache.....I had a whole thread circulating with a half a dozen or so apps ran with 3 different ratios for 3 different speeds then ran with highest clock but not highest memory speed versus lower clocked to obtain highest ram speeds...the percentage results shows my conclusion exactly....

 

[DanRydell]

I'm trying to decide whether I should try to OC with value RAM on a 939 Winchester or if I should get a cheaper 754 Newcastle. Is it worth the bother to try to OC with a non 1:1 ratio? Or should I just not bother trying?

 

[Duvie]

Yes it is worth it...If you can get the oevrall clock of the cpu high enough it will be better.... This same issue kept many ppl from pushing the p4's back about a year ago...They thought 1:1 was all that you should try for....mnay would stop their oc short by as much at 200-300mhz. I proved in testing that 300mhz faster even running 100+mhz slower ddr was still much faster.

If you can clock to 2.4ghz for the highest with 1:1 yet you can clock to 2.6ghz with a divider closer to 3:2 the higher clock of 2.6ghz should still be faster in most all things. Closer in memory bandwidth intensive apps but as far away as 200mhz would be in cpu intensive apps like encoding, scientific, distributed computing, rendering, etc....

Lets say you get a 3200+ winchester...10x multi, right...

Let say you leave it at 1:1 but set HTT lower and set timings to the most conservative cas 3 timings.....Chances are you can probably only get to 220 fsb or 2.22ghz but be running 440mhz cas3 ddr....

Now same chip could possibly run 250fsb for 2.5ghz with a 6:5 divider (166mhz setting pciked) and have the ram running at 420mhz ddr or even 3:2 divider (133mhz setting picked) running ram at 333mhz ddr.... IN both of those cases I bet the scores in benchmarks would be overall greater then the slower closk higher speed 1:1 setup....

 

[BonzaiDuck]

I can't pick your processor for you. Duvie thinks it is actually better to run a divider that deviates from 1:1.

My view, in principle, is that opening up bottlenecks in the "storage pyramid" (with slower speeds at each lower level from L2 cache to RAM to disk) makes for an overall faster system, but it may depend on what sort of applications you're running. If you do video processing, then -- I'm speculating -- there will be DMA transfers from disk storage to memory, and the faster memory would help. The purpose of cache is to buffer slower storage (RAM) to relieve the bottleneck to faster storage (CPU registers). So -- as I said "in principle" -- larger cache would allow for better performance with a faster CPU and slower memory.

So, personally, I would think that a run-of-the-mill Northwood with 512K cache would benefit from memory running at 1:1. All I know is that I get some pretty good benchmarks at 1:1 with the processor over-clocked and memory that is rated to run at a higher FSB speed than the stock 800 Mhz.

But it would be easier to over-clock "value-ram" with a divider. That way, you don't have to push it beyond the limits of its spec, or loosen the timings as much to get there.

 

[Mean MrMustard]

Originally posted by: Super6
Using name brand value PC3200 should work ok with relaxed timings. I'm beginning to believe that some of the 939 mobos are a little particular when it comes to RAM type so your results may vary.

Super6

Damn right they're particular! I've trying for the last month to get my Neo2 and Ballistix RAM to run stable PERIOD! Its pissing me off that I might have to get new RAM even though I and many others have paid good money for high performance RAM.

I put the memory in single channel mode last night. With the memory running at 200 Mhz, HTT 240 x 10, I was stable for over 2 hours (I stopped Prime). In dual channel mode I have never gotten past ONE minute. Like I said in another thread, dropped some Corsiar Value PC3200 in there and ran it in DC, and it ran stable. I put the Ballistix in my NF7-S and it worked fine (no DC though).

Before anybody asks:

Athlon 64 3200+, HTT 240x10 ~ 2400 w/ a Zalman AlCu HSF 1.575v -0.1v(Neo2)
Neo2 Plat. MB
HT 3x
2x512 Ballistix PC3200 @200 Mhz, 2.5-3-3-9 2T @2.85v, currently in single channel mode :|
Antec TruePower 480W PSU
BFG 6800GT

 

[Steeplerot]

Gigabyte GA-K8NS Ultra-939 geil 3200 value ram no problems at all here HTT 267 mem actually running at 218 stock timings no vdimm increase timings by SPD and in dual channel mode.
Solid as a rock all day every day. Possibly those MSI boards?
I wouldnt touch a MSI mobo with a ten foot pole after being burned quite a few times in the past with amd+msi.

 

[DAPUNISHER]

Originally posted by: Steeplerot
I wouldnt touch a MSI mobo with a ten foot pole after being burned quite a few times in the past with amd+msi.

I have said that for years now. I am one of the worst MSI bashers around in fact. I lost my a$$ on the K7T series and the bad caps. Even the first couple RMA replacements died!!! After that I just ate the cost of exchanging the boards for the clients. Then there was all that time and effort of the service calls.

Know what, I'm posting from a SMP system not only powered by MSI, but a VIA chipset to boot! I hadn't used VIA for a single build in years either. SiS for ultra-budget systems and my first P4, and nV and Intel for everything else.

The horror of horrors is this thing is a rock for the 2 weeks it has been running 24/7 BOINC SETI, gaming, DVD back-ups, ect. 2x242@1.725ghz, 6800GT@431/1.1ghz, 2 HDD, 2 ATAPI, 2x512mb, 120mm+2x80mm fans, 2 MSI coolers@2600rpm, LED case readout, and all on a True430! :shocked: This thing manages to pass the longevity test and I may have to stop bashing MSI....but then again, with their crappy QC I doubt it


To address the OP, I use inexpensive CAS3 Apacer 3200DDR Samsunbg TCC4 chips I have had for 2yrs+ now and it has proven compatible and reliable in every system it has been in. I've been running my old C0 3000+@2.4ghz almost a year now and it's running 400DDR 2.5-3-3-10 right now using a divider and that makes it basically a stock 3400+ system. I'm certain I could get a few more FPS and definitely better benchmark numbers by running 2-2-2 and higher frequency but it isn't worth the price premium to me for the small performance gain, YMMV.

 

[Zebo]

Read the link in my sig. First thing you need to do is set mem setting to 100 to elimiate that as a varible and MAX your chip out using P95. Then test mem separatly by lowering multi and mem setting of 200 while upping HTT/FSB.,

EIT LOL@DP

 

[Zebo]

I wouldnt touch a MSI mobo with a ten foot pole after being burned quite a few times in the past with amd+msi.
----------------------

Makes it hard right now to resist when every review site shows the neo2 as superior to all boards. I also have had issues with MSI both a KT333 board and Neo 754 which both got RMA'd after wasting hours w/ them trying even stock speeds. Even now the Neo2 has Sata issues like slots 1 and 2 not having a lock which 3 and 4 do. But it seems if you get one 300Mhz capable it's the best mobo. I have the same board as you do and have serious problems getting my TCCD ram running above 240mhz which it does 265 all day in chaintech VNF3.

 

[Duvie]

Originally posted by: Zebo
Read the link in my sig. First thing you need to do is set mem setting to 100 to elimiate that as a varible and MAX your chip out using P95. Then test mem separatly by lowering multi and mem setting of 200 while upping HTT/FSB.,

EIT LOL@DP

Exactly why do ppl forget this and then scratch their head when they can't get far. I also set to ram to most conservative timings and lower divider initailly until I find the cpus limit. then I test, Then I start raising ram speed and lower clock to find rams limit. Test there...Ultimately I see what ram speed versus cpu speed dopes for me and make my choice from there interms of the apps I use...

 

[Andres3605]

i can just say i have excellent luck with msi products (mobos and video cards) not a single serius problem and rock stable, lets say 8*275 and 240*10 without even reaching teh limits at 1.45V(cpu) and 2.7V(RAM) may be just one of the lot but i was the lucky guy..