FX-8150 @4.63 Can I push it more?

[firefiber]

So I've been playing around with my rig. I managed to get it stable at 4.63GHz, with 1.512V. Multiplier at 21.5, and bus at 215. I ran Prime95 overnight (nine hours in total). No errors. My temps are great, never strays beyond 50C to 54C. My ram is running at 2011MHz, with 1.6V (Corsair Vengeance 1866 kit).

I'm just wondering if I can push it anymore? I have an H100, so temps are great, but whenever I up the multiplier any higher, Prime95 either freezes, and I have to restart, or it gives me an error. Any ideas?

 

[caddi daddi]

could you post shots of
cpuz, processor tab, memory tab and spd tab?
cpuid hardware monitor? shot while under load, this will help the folks in here help you.
we also need the specs of your rig, motherboard, memory cooling system, powersuply and such.

 

[firefiber]

http://valid.canardpc.com/show_oc.php?id=2609734

That's my CPUZ validation.

Hardware specs:

CPU: AMD FX-8150
GPU: Sapphire HD 7970 Dual-X OC
RAM: Vengeance 1866
PSU: Corsair HX1050
MOBO: Asus Crosshair V
HDD: Kingston HyperX 3K (x2)
CPU Cooling: Corsair H100.
CASE: Corsair Carbide 500R

I'll try to get a shot of HWmonitor under load in a bit.

 

[I Mateo I]

I'd be interested in knowing this as well. I signed up here to post a thread almost identical to this. I'm running 4.6 stable on my 8150 but I really want to get to 4.8 or higher while keeping it safe for my everyday use if possible. I also have a liquid cooling unit for my cpu and run 50-54 as well. Let's OC this thing!

 

[ssjwizard]

Short answer is yes infact you can probably get 4.8 on that chip, but its not really worth it. I am in the middle of a bunch of clockspeed comparisons with the BD vs PD chips and I can say with confidence that above 4.0Ghz on a BD chip its pretty pointless to clock it higher, as the clockspeed scaling gets choked to death by the cache/resource scheduler.

 

[I Mateo I]

So, in short, bumping up to 4.8 from 4.6 wouldn't give a noticeable increase due to a bottleneck elsewhere?

 

[ssjwizard]

Yes thats pretty much it. The problem is the way they designed the upper level memory management on the BD chips. To get it out to market in time they had to leave certain bits very unpolished. The chips tend to hold onto cache longer than needed and once its full you start loosing work time waiting for the cache to dump old data so you can fetch new data. This starts at around 3.8Ghz in my testing and pretty much flatlines CPU performance above ~4.2Ghz.

Gaming benchmarks at normal clockspeeds here
Also see my pushing the limits threaded testing here

 

[caddi daddi]

ssjwizard, you have confirmed what i foud out a while ago with 8120's, I have a massive chassis setup program, run time with the 8120 was measured in days, 4.2 was where the processor i called the pig completed the task the fastest with no errors.

 

[RGone]

@ ssjwizard, I had not heard the why; being cache release bottlenecking, but your results are *exactly* what I see cpu speed wise with my FX-8120. I don't have the newer FX-83xx to test against but the results with my FX-8120 are almost an exact mirror of your results with the BD.

I tested Cinebench 11.5, 3D06 cpu performance, 3DMarkVantage and some SuperPi and the performance after the jump from the 4.2ghz you speak of is not worth the heat nor the extra voltage to get Prime 95 stable to consider those greater cpu mhz as day in and day out usable.

I did not start my testing as low as you did since I began at 4.1Ghz and incremented to 4.8Ghz and yet I see pretty darn close to what you see when using my FX-8120. AND I actually hedged my bets by letting my memory speed, CPU_NB and HT frequency rise right along with the CPU Mhz and I still did n0t see a big jump in performnace as the cpu speed increased by 100Mhz from 4.1Ghz to 4.8Ghz.

The biggest muscular pump up was just about 4.2Ghz and after that the effort to pump up more CPU Mhz was literally just an exercise in Epen nearly.

Thanks for sharing your efforts. RGone...ster.

Oh well I did not see C_D when I began to write this post. Hehehe.

 

[firefiber]

So there's no point in trying to push it any further (other than bragging rights, hehe)? Dang, that sucks. Oh well. I should just keep it at 4.63GHz then?

But now I've got another problem. I ran Prime95 a bunch of times, and it was always stable. I also ran OCCT (linpack) and LinX. No problems. Then, overnight I ran Memtest. And when I checked today morning, there were a whole bunch of errors. Any idea why? My memory is running at 2013MHz, 9-10-9-24, with 1.6V. They're a Corsair Vengeance 1866 kit (2x4GB). I'm at work now, but I reduced my bus back to 200, so now it's running at 1866 again, and I started Memtest before I left home.

Was my overclock too aggressive? I've seen people push it much higher on the same board though, with the same kit, so what am I doing wrong?

EDIT: since this is now regarding memory, should I create a new thread in the Memory section, or can I continue here?

 

[BeepBeep2]

@firefiber
If you need 1.51v for 4.63 GHz then you won't be going much higher. The better clocking CPUs will do 4.6 with about 1.41v, not 1.51

How long did you run Prime95 for? When you run Prime95, try selecting "Blend", then "Custom" and input 80% of your total memory, run the test for 4 hours to ensure memory stability.

Also, on the label for your ram, what is the version number? If it is 5.1x (Hynix CFR/BFR), then try setting 9-11-9 instead and see if it helps stability. If it is 4.1x (Samsung), try 9-10-10.

 

[firefiber]

Alright, I'll try that the moment I get home from work (this is so frustrating).

A quick question though: I've always wondered, is it really worth increasing the frequency, when I have to loosen the timings? Wouldn't that do the same thing? I mean, it's a faster clock, but high latency, so overall it's going to be slow anyway, right?

 

[Anonaru]

Alright, I'll try that the moment I get home from work (this is so frustrating).

A quick question though: I've always wondered, is it really worth increasing the frequency, when I have to loosen the timings? Wouldn't that do the same thing? I mean, it's a faster clock, but high latency, so overall it's going to be slow anyway, right?

You have to bench your ram to find the answer to that question / run Cinebench. The aim is to get as high of a frequency and as tight of timings as possible, most RAM has a "Best" point for both values, where going too tight ends your frequency bumps early, and going too high of a frequency requires timinigs that actually gimp your ram.

1866 with 8's is faster than 1600 with 7's for most applications. Oh, for reference, the PD chips hit a softer (But similar) performance ceiling too. I can get noticable gains in overall benchmarking until the 4.4 - 4.6~ range, where going any higher really is redundant

 

[ch0b1ts]

You can try lowering the Multiplier and Raising the Bus Speed.
I wasn't able to get my CPU over 4.2Ghz stable by raising the multiplier regardless of the voltage supplied.
I dropped the multiplier back down to 16.0 and raised the bus speed to 280. It seems stable at 17.0x280 but I dropped it down a notch for 24/7 running just to be safe. Haven't had a single freeze, blue screen, or crash in 8 months running this rig.

 

[firefiber]

Okay, so after hours and hours of Prime95, voltage altering, BSOD's, I'm just about close to screaming at this thing and going Linux. =[

I've currently set my base clock to 240. Here's a pic of my cpuZ:

It's stable now, but I'm scared to push it more. Specially 'cause the more I read up about the different voltages and naming schemes the more confused I get.

I've increased my CPU/NB (which is the IMC that's ON the CPU right?) to 2.6GHz, with 1.45V on CPU/NB. I get that. But, what exactly is the HT? HyperTransport which is the direct connection between the CPU and the what exactly? PCIe? What does the Northbridge do then? So isn't the HT a different name for FSB? Or is HT an actual chip on the board?

I have a Crosshair V Formula. On the BIOS I don't see any place that mentions the Northbridge frequency. Or is it written under some other obscure term? There's the CPU/NB frequency, and the HT Link speed (which raises the question, what the hell is the HT Link? What's the difference between the HT Link and the HT?). And there's a NB HT voltage and a separate NB voltage. What are these for?

Now I've also read that AMD doesn't have an FSB (which, as far as I know, is the connection between the CPU and the NorthBridge; and the NB is the chip that handles the memory and PCIe). Since AMD has an integrated memory controller on the CPU, the NB only handles the PCIe right?

This is just so bloody confusing. Please don't tell me to go Google, 'cause that's what I've been doing for like a week. I've read everything on AnandTech, these forums, TomsHardware, the ROG forums, etc. It's so frustrating! ARGH!

 

[RGone]

This link is as described and in the first page and 1/2 the very bus speeds are described as the author moves down the bios menu screen where overclocking is "done" on the Crosshair V motherboard. After he hits those first few menu choices and gives description, he gets into the DRAM section. Good but you need to know Ram somewhat for the dram write-up to mean a thing or generally so.

ASUS Crosshair V Formula BIOS Guide - Overclocking

 

[firefiber]

I've read that before, I also have a printed copy of it in front of me.

Anyway, I increased my bus to 240. My multiplier is at x18, my CPU/NB is at 2.6GHz, and because of my bus, my memory is at 1926MHz (9-10-9-27). I've given 1.34V to my VCORE, 1.45V to the CPU/NB and 1.52V to my DRAM. I ran Prime95 overnight, and one of the cores had failed in an hour, while the others went on. The core that failed, gave me the rounding error (FATAL ERROR Rounding was 0.5, expected 0.4).

I was told that this usually means I need to increase my voltage, but which voltage? Surely not the VCORE? I'm only running at 4.3GHz.

Please help!

 

[Anonaru]

I've read that before, I also have a printed copy of it in front of me.

Anyway, I increased my bus to 240. My multiplier is at x18, my CPU/NB is at 2.6GHz, and because of my bus, my memory is at 1926MHz (9-10-9-27). I've given 1.34V to my VCORE, 1.45V to the CPU/NB and 1.52V to my DRAM. I ran Prime95 overnight, and one of the cores had failed in an hour, while the others went on. The core that failed, gave me the rounding error (FATAL ERROR Rounding was 0.5, expected 0.4).

I was told that this usually means I need to increase my voltage, but which voltage? Surely not the VCORE? I'm only running at 4.3GHz.

Please help!

Drop your NB back a tick. I almost 100% assure you that you will never get a fully stable OC on a BD chip with a 2.6+GHz Northbridge frequency

 

[Ajay57]

Just as a small heads up i have read many times that on these chips when OCing ODD Numbers work better. For some reason which i can not remember now why i.e. 15,17,19, and so on! AJ.

 

[firefiber]

Okay, so I set the CPU/NB to Auto, and decided not to mess with that until I know more about it.

Instead, I upped my core clock to 242, and the multiplier to 19. It's stable at 4.6GHz with the memory running at 1942MHz! I'm giving 1.5V to VCore and 1.52 to DRAM. I ran LinX, Prime95 (for about four hours), OCCT. No errors.

The thing is, I feel I can push it a wee bit more, but the only thing stopping me now is the heat. I tried upping the bus a little (244), the temps go over 60 degrees, and that's not something I want on this chip. Currently, they over around the late 50's. I'm using an H100, so is this the limiting factor?

I'm thinking of maybe reading up and trying out custom water-cooling, but I wanted to know, does it offer substantially lower temps as compared to an pre-made water-cooling kit (such as an H100)?