Only 242x10 on my 3500+. Why won't it go higher?

[KatanaBob]

Hi! I have a AMD64 3500+ I'm trying to overclock (in windows). I only get it to 242x10 rock solid, but any higher, and Prime95 fails very quickly. It's a Newcastle 130nm (YES, I know it's not the best for overclocking, but it should still be maxing out at 2,6 right? At least others have gotten it that high)

Here's the things I've tried already:

1. setting cpu voltage to 1,55V (from 1,50... motherboard won't let me set it higher than 1,55)
2. setting ram voltage to 2,6V (from 2,5)
3. Increased latency from 2,5-3-3-7 to 3-4-4-8
4. set memory to half divider (100mhz, from 200mhz in bios)
5. HTT multiplier set to 3 (still way under max, which is 1000=fsb x multiplier)
6. temperature under heavy load is never over 60 degrees celcius on the CPU (temperature isn't the problem, because it ran at 67 degrees celcius under load before I removed some dust from the heatsink)
7. locked agp and pci frequencies
8. removed dust...
9. max fsb is over 300, so that's not the problem
10. moved sata disks to sata3&4 (from1&2, motherboard apparently doesn't lock 1&2)

Specs:
MSI K8N neo2 platinum 54 - Socket 939
AMD64 3500+ Newcastle @2.2stock, stock cooling
OCZ Premium 2,5-3-3-7... think it was pc-3200

And I'm using prime95, HWmonitor and Clockgen (specific for Nforce3-boards)


PS: I only get to ~2400mhz on the CPU on both 1,50V and 1,55V. Does that mean I've forgotten something? I mean, increasing voltage SHOULD make it go at least ALITTLE higher, right?

Thanks in advance.

 

[arges86]

I always recommend OCing from the BIOS not windows. I don't know why, but it seems to be more reliable that way.

Sometimes its just the luck of the draw in getting a CPU.
Sometimes a processor can be overclocked by a Ghz and others of the same line can only be OCed by 100 Mhz

 

[KatanaBob]

can't clock from bios, don't have safety catch, so I have to use cmos jumper everytime I clock too high. I made a thread about it, but I can't seem to find it now.

Anyway, maybe it's as you say, that I'm unlucky. Just kinda disappointed.

Thanks anyway

 

[AsAnAtheist]

You should understand how CPU's are manufacturers .

Let's take for example Phenom II processors (45nm). They make 100 chips (quad core chips), and due to mass production all of them are manufacturers differently.

Out of the 100 chips:

Only 10 make it to Phenom II 965 Quality:
Only 15 Make it to Phenom II 955 Quality:
Only 20 Make it to Phenom II 945 Quality:
Only 30 make it to Phenom II 720 (x3 core) being 1 core is faulty or cannot be clocked high enough:
Only 25 make it to Phenom II 550 (x2 core) Being 2 cores are faulty or cannot be clocked high enough:

Now each one of these processors can be "handicapped" or under clocked /cores locked to change them from a 945 to a 550. This is where supply and demand goes.

Out of the 100 Chips, only these many have sold.

Only 0 are sold as Phenom II 965 Leaving 10 chips to sell
Only 2 are sold as Phenom II 955 Leaving 13 chips to sell
Only 10 are sold as Phenom II 945 Leaving 10 chips to sell
Only 25 are sold as Phenom II 720 Leaving 5 chips to sell
All 25 Phenom II 550's are sold, and there is a demand of 40 more chips.


What will AMD do, well obviously the higher end processors arn't selling so perhaps they should sell them as phenom II 550s? So they take some from perhaps the 945 line, some from the 955 line and put them as Phenom II 550s.

This is how someone may get lucky and get 3.8 ghz out of the box/stock fan.... Some have even unlocked cores on the 550s and ran them near to the 965's clocks.

This is true for all lines of processors, AMD or Intel.

 

[OvrClkr]

have you tried raising the FSB in the bios? and if so how high does it go w/o crashing?

 

[KatanaBob]

Asanatheist, that's fascinating, I didn't know they made them like that. I always thought they made the lower end processors with fewer transistors or something like that. That they made them worse on purpose. So a certain percentage mid-range phII processors are actually "failed" higher end processors? Must be hard to predict earnings and sales with so big variables.

By the way, does that mean the Black Editions are not 110% certain to be good overclockers? I mean, that's kinda why you pay the extra money for BE.

OvrClkr, I can raise the FSB to 242, just like in windows. But like I said, it's extremely frustrating to experiment through the bios, because I lack the warning that I mentioned in the previous post. Found the thread explaining it: http://www.tomshardware.co.uk/forum/page-254407_11_0.html

 

[OvrClkr]

What is your clock speed at 242 (FSB)?

 

[KatanaBob]

What is your clock speed at 242 (FSB)?

I've set the multiplier to 10, so that would be 10x242 = ~2420 MHz. This is where the CPU maxes out it seems. Just a few MHz more and it would fail within 2-5 min in prime95.

Also, in BIOS DDR-RAM is listed at 333 MHz, even though it should be around 400. I've set the divider to (166/200), so 166/200*242 = 200,86 for one stick, and 401,72 MHz in dual channel, right? So why doesn't it say 401,72MHz instead of 333? I suppose it's a motherboard thing.

Edit: The RAM doesn't actually have the name "Gold" in it, but it was premium... fixed.

 

[OvrClkr]

Assuming that your multi is @ 10 have you tried lowering it to 9 and cranking up the FSB a bit higher?

 

[KatanaBob]

Assuming that your multi is @ 10 have you tried lowering it to 9 and cranking up the FSB a bit higher?

I'll try it right away, but I thought the CPU's max limit was decided by MHz and not by a specific combination of multiplier and fsb. I mean, if the limit is 2420, it's 2420, no matter how I factor it or...?

 

[ill b ben]

Its not so much that, its to see if your hitting a fsb wall, in other words it could b the mobo maxing out rather then the chip.

Altho imo software overclocking simply is not as good as using the bios, its handy for some easy oc's at slight increases but to go ''all the way'' bios is definatly recommended.

But then again if ur saying most chips max out at 2.6, 2,4 sounds rather resonable.

 

[OvrClkr]

Not in all cases, you can try lowering the multi and raising the FSB above 242. If your CPU maxes out at 2.42Ghz no matter what then it could be the mobo as well since it has limited values in the bios. Have you considered upgrading the CPU/mobo? You can get a much faster CPU and better mobo for less than 120.00$ US....

 

[KatanaBob]

Thank you for replies.

Originally, I could get up to 300 with the fsb without freezing or anything, but I forgot to test it in prime95. Still, I think the fsb hasn't hit the wall.

OvrClkr, prime 95 would fail within 1 minute at 273*9 = 2457 MHz, and within 2 minutes at 272*9 = 2448, so I don't think lowering the multiplier helped. Was worth a try though.

I'd love to upgrade of course, but it's just that once I go from socket 939 to a new socket, I'll not only have to buy only the CPU and MB, but new RAM, new GPU (I have a 7800gs AGP... ), new power supply, new case, and when I've bought that much, I might as well go all the way, you know? But then there's the money.

When I set up hypothetical builds I always get around the equivalent of 500-700 dollars, and I ask myself "Do I really need this?". Answer is no of course.

 

[AsAnAtheist]

Asanatheist, that's fascinating, I didn't know they made them like that. I always thought they made the lower end processors with fewer transistors or something like that. That they made them worse on purpose. So a certain percentage mid-range phII processors are actually "failed" higher end processors? Must be hard to predict earnings and sales with so big variables.

By the way, does that mean the Black Editions are not 110% certain to be good overclockers? I mean, that's kinda why you pay the extra money for BE.

OvrClkr, I can raise the FSB to 242, just like in windows. But like I said, it's extremely frustrating to experiment through the bios, because I lack the warning that I mentioned in the previous post. Found the thread explaining it: http://www.tomshardware.co.uk/forum/page-254407_11_0.html

Black Edition processors are taken from the higher quality chips which may have failed 1 or 2 cores but leaving 2 cores in excellent quality then being left unlocked (multipliers) (Sometimes all cores working, just getting locked to be sold as black editions). Please understand that my demonstration was just an example.

It isn't too hard to predict variables/sales. Any of the higher end processors can be underclocked/locked as easy as they were set to higher clocks. All they have to do is release small batches to the market, and see the sales numbers. From there, they edit they're next batches to meet demand (using previous batches to fill in the gaps meanwhile. Which could be why the first batch of Phenom II 550's all had 100% success rate, I cannot recall where this information was taken from so it may not be valid (first batch=100%).