SOLVED ASUS Crosshair V Formula Thunderbolt and AMD FX-8150 Overclock Questions

[Vic Velcro]

I'm a bit overwhelmed with all the BIOS settings for this beast. I do not have an understanding of memory timings and haven't been able to make any sense of info I have found on the web. The last overclock I did was 5 years ago, and served me well until late last year when I decided to get something newer. The old box still works flawlessly, but is nothing like the new gear.

I put this all together and ran it at stock settings for short periods of time, to 'break it in'. It's been solid this entire time.The HWMonitor temps in the image below are at 100% load folding for several days, nonstop.

I began the first bit of O/C'ing just a few days ago. That's when I started seeing options that I know absolutely zip, nada, and squat about.

On this machine booting Win7x64Premium, I will be rendering video via Sony Vegas Platinum 11, converting video formats with one of the 100's of converters out there (haven't picked one yet), editing image files via Paint Shop Pro x3, creating and editing audio files via Sound Forge and Acid.

I will also be gaming from a second bootable SSD (using whatever OS seems best - probably Win7x64Premium) with Starcraft II, Total Annihilation, Batman Arkham City, Dungeon Siege II (with mods), and quite a few other legacy games going as far back as Duke Nukem 3D and Lords of the Realm II. I may want to add some newer games to the list in the near future but not any of the WoW or CoD types or anything STEAM and Microsoft LIVE. Basically, if it can go on an XBOX360, it won't be on my computer, either.

1. What would be good memory timings to start with? Or is this really a big deal?
2. What is a good CPU multiplier setting to safely begin further overclocking?
3. I'm not interested in changing the FSB or HTT or whatever the equivalent is now called.
4. Is the onboard SATAIII x6 better or worse than the ASMedia SATAIII that is also onboard?
5. The manual isn't clear regarding which of the PCI slots are which. Can anyone help me clearly identify which is 16, which is 8, which is 4, and whether/which any are 2.0 and/or 2.1? Does it really matter that much?
6. Should I consider doing anything with the Northbridge? I have no clear understanding of what it is or why it might be changed.
7. Are there any questions I *should* be asking, but am too stupid to know I should be asking?

[THE GEAR]
AMD FX 8150 @ 4.013GHz (3.6 Ghz stock)
ASUS Crosshair V Formula Thunderbolt ROG BIOS 1301
16GB (4x4GB) Corsair Vengeance 1600 DDR3 9-11-9-27 (allegedly 9-9-9-24 stock)
HIS 6870 IceQ 1GB DDR5 O/C 1000/1200 Catalyst 12.4
Corsair H100 Cooler (water)
Samsung 830 SATAIII SSD 128 GB (Main Boot)
Patriot Pyro SATAIII SSD 120 Gb (Secondary Boot - for gaming)
WD7500AACS 750GB SATAII HDD (not Boot - for storage of media and user files)
Ultra X4 1200w Gold PSU
Corsair Carbide 400R Tower
Plextor PX-L890SA DVD
iHAS 524b DVD
Thermaltake 4 bay 2.5" SATAIII hotswap w/fans
OS on the main boot is Win7x64Premium
OS on the secondary boot is Win7x64Premium but can be XP32 or XP64
OS on additional drives can be anything appropriate, since I have a crap-ton of HDDs and SSDs just sitting here

 

[trents]

Vic, please redo the CPU-z "SPD" tab screenshot. It's void of data because you don't have a memory module in slot #1 of the motherboard. Use the drop arrow to change to another slot that has a stick. The SPD tab shows the memory manufacturer's recommendations for timings and voltages at different frequencies that your ram is capable of. The "Memory" tab shows your current ram settings.

The things you probably want to focus on in overclocking that CPU are:

1. CPU multiplier (maybe called something with the term "ratio" in your bios). Increase it in .5x increments and test for stability after each increase.
2. CPU core voltage (it will need to be increased in small increments when you encounter instability in the testing.
3. Core temps and CPU (socket) temps. Don't let the former exceed 55c or the latter exceed 65c.

It's not a good approach to ask for plug and play values from others when overclocking. Every system is different. The best way is to feel your way through the overclocking in small increments of change, testing for temps and stability along the way. If you are interested, I could give you a pattern for such an approach that works well and is safe.

But first of all, go into bios and disable all the green, power saving, downclocking stuff: Cool N Quiet, C1E, C6 and Turbo. Then go into Windows Control Panel Power Options and configure it to High Performance.

 

[Vic Velcro]

Ok, understood, so far.

I would definately appreciate an approach pattern, as it didn't used to be this involved.

What I am looking for is a good set of 'starting' settings to begin the process with, not so much a 'plug n play' answer. An explanation of timings that I can wrap my head around would help. I've tried searching up an explanation, but none of what I found is very clear to me.

I'm going to have to learn the 'new' stuff simply by doing it. I just want to be as certain as possible that I don't start out completely wrong. Burning up a couple hundred is no big deal, but ruining this set of gear would hurt the pocket just a bit.

As far as the CPU-Z goes, I have a mem stick in every slot the board has and all my ram is available. I've never seen it blank on any of my other machines. I figured posting the screenshot as it was might mean something to somebody.

For now, I'll start by changing the green and power options in the BIOS, then check Windows power options. I am pretty sure I had it set for HighPer, already. But a 2nd look can't hurt.

Thanks for the reply.

 

[trents]

Here's a quick and dirty explanation of memory timing and frequency and their bearing on performance:

Frequency, like 1333, 1600 or 1866. This is the DDR3 "transfer rate". The higher the frequency number, the broader is the bandwidth. Moving from 1333 to 1600 or from 1600 to 1866 is like widening the lanes or adding more lanes on a freeway. At rush hour, it will relieve congestion and speed travel times. When there is no real congestion, it doesn't help much. Higher memory frequency helps computer performance when the CPU is capable of handling and is calling for lots of data, i.e., when there is peak demand on the system.

"Timing", also referred to as "latency": These are like traffic lights on a street that manage the flow of vehicles. They cause traffic to pause at regular intervals. Each timing number represents a traffic light. The bigger the number, the longer the traffic light stays red and the slower the traffic can cover a given distance.

So, memory performance is affected both by frequency and timing. There is also a relationship between frequency and timing in that as frequency increases, timings need to be "relaxed" to prevent "crashes". "Relaxing" the timings means to assign larger numbers. So there is a degree to which the performance gains of higher frequencies are offset by more relaxed timings but as a general rule it is still worth it to go for the higher frequency.

Are you with me so far?

 

[Vic Velcro]

Yes, I am. That explanation is working for me, so far.

So, for timings, smaller is better but not larger.

 

[trents]

Yes, smaller is better. But too small and you will be unstable. If you will run the free program CPU-z and look at these three tabs: "CPU", "Memory" and "SPD" it will give you loads of info about your system and your bios settings. The "Memory" tab will tell you what the current frequency and timings are for your ram as you have it configured in bios. The "SPD" tab will display the memory manufacturer's recommendations for voltages and timings for your ram at various frequencies. Compare the two to see if your timings are appropriate. If you take the main memory timing bios line item off of "Auto" and put it on "Manual" you will see many individual timing adjustments. Only the five or six that show in CPU-z need be fussed with. The others can be left on Auto.

 

[Vic Velcro]

I restored the original board settings to stock. No overclock at all and auto memory timings. All power management has been disabled, per your earlier post. No C1E, turbo, etc.

Memory timings per the manufacturer and CPUZ indicate 9-9-9-24-41 but the current detected setting is 9-11-9-27-41. Not sure if this is an error, correct regardless of manufacturer, or what.

Here's a screenshot with the CPUZ tabs, HWMonitor temps, running Prime95 with no visible errors.

 

[trents]

How long was Prime95 running when you took that screenshot? The max, min and current core temps are the same? Please have HWMonitor open before you start Prime95 so we can see what the idle temps were. Also, scroll down a little on the HWMonitor interface so we can see the CPUTIN/TMPINx lines. That's where the CPU socket temp is and it's good to look at that too to get an idea if the core temp sensor is calibrated poorly.

I would go into bios and manually adjust those timings to what the XMP-1600 column in the CPU-z SPD tab shows. You only need to adjust the few you see in CPU-z not the 20-25 you will see in bios. The terminology will likely vary a little between CPU-z and your bios but a couple of hints there: The first 5 timings in CPU-z SPD tab will likely line up with the first 5 in bios. The CR (Command Rate) in bios will be toward the bottom of the list and is the only one with a "T". When in doubt, make a small change in bios and then load Windows again and check the change against what you now see in the CPU-z memory tab. If you changed the wrong one then it won't show up where you expect. Small changes are not likely to keep you from booting into Windows. Also, check the size of the values. Some of the are big numbers like 24 or 40 and some are small, like 8 or 9. That will help you line them up right too.

 

[Vic Velcro]

At stock settings, HWMonitor temps never changed during 15 mins of prime95. Adding to the multiplier, temps do change.

The attached screenshot is after several gradual overclocking steps, based on your advice and info. I've set the mem timings as you suggest, disabled 'Spread Spectrum' on the CPU and PCIe, upped the CPU multiplier, had to up the CPU voltage 3 increments at 21.5 multiplier and current prime95 is what you see.

Core #7 worker stopped at 17mins. with 8 tests, 0 errors, and 100 warnings.

All other cores are still chugging along at 19mins. elapsed.

 

[trents]

A core dropping out but no blue screen usually means you are very close. Give it another small increment of vcore. I think your temps look like they will handle a bit more core voltage.

 

[Vic Velcro]

So far, your advice has been perfect.

I am now at 4.4Ghz with CPU voltage of 1.35V (peaks at 1.4V due to VRM), spread spectrum on, and VRM set very high and/or ultimate. Temps are a bit above what I was hoping for, but not enough to cause me any panic. I can fudge/tweak those down by morning.

I had to bump the H100 cooler speed from minimum to medium.

So, in prime95 (15 minute duration), it blends. I'll run it 24 hours, later.
But, does it fold? Let's find out.

Anyhow, thanks for the help so far.

Without pushing the CPU any further, would it be reasonable or wise to alter any other settings for more O/C? Or would you deem it best to call it 'GOOD' right here?

 

[trents]

Looks like in your current memory timings the CR is 1T (no value in that spot for the CPU-z tab usually means it's 1T). The SPD tab indicates it should be 2T. If you have no instability from the 1T setting leave it alone as it's a quicker timing but if you encounter instability you might want to change it.

The only other suggestion I have is to increase your CPUNB voltage a bit over stock and bump your ram voltage up from 1.5 to 1.55. These two measures usually help with stability, even when the ram itself isn't overclocked. The CPUNB is a component of the integrated memory controller (ICM). I don't know what to tell you to set it too because I don't know what stock voltage is on that component with the FX chips. If it were a PII Deneb or Thuban I would tell you to put it to 1.225. Maybe someone else can shed light on this.

 

[Vic Velcro]

The VRM setting appears to boost the NB voltage just a bit, when CPU load goes up. I'll leave it as is for the moment and put a post-it on my monitor regarding your NB voltage advice, in case needed. Stock is 1.16 and shows 1.16 in HWM at idle. Under load, it goes up to 1.18 due to the VRM I mentioned.

The 1T timing *was* 2T, but somehow (maybe me by mistake) it become 1T. I haven't seen any weirdness with it yet, but will change that back to 2T later.

I'll go up to 1.55 on the mem voltage and see where it all stands for the next 24 hours.

 

[trents]

VRM? Wonder if that's the same as LLC (Load Line Calibration) in your bios from the way you speak of it. I'm used to hearing "VRM" in terms of the Voltage Regulation Modules on the motherboard - the cube-shaped black things next to the CPU socket.

 

[Vic Velcro]

Yep. Without looking, going from recent memory, DigiVRM is the main BIOS category, and Load Line Calibration is available for CPU and CPUNB. I set CPU for max and CPUNB for ultimate.

 

[trents]

Be careful with LLC that is too aggressive. It will supplement your bios voltages and under load and could put them in dangerous range. Check HWMonitor CPU voltage when under load. I would not feel comfortable with core voltages exceeding maybe 1.525 under load and 1.5 at idle.

 

[Vic Velcro]

Currently running prime95 hopefully for 24 hours. CPU VCORE is 1.38V up from 1.33V at idle. CPU temp (can be ignored on FX chips) is 66C and Cores are at 52C.

So, it looks like Load Line Calibration raises the VCORE by .05V and the NB by .02V, unless I'm getting a flawed reading.

I could turn off the LLC, but then I'd have to up the base VCORE in the BIOS to 1.38 all the time, I think.

If my understanding is incorrect, I wouldn't mind being 'set straight'.

 

[trents]

No, I think your understanding is correct. The only thing that puzzles me is your statement about ignoring the CPU temp with the FX CPUs. CPU temp for any CPU is rather a misnomer as it is actually the CPU socket temp, being measured as it were from the motherboard sensor in or around the socket. In my experience this temp sensor is usually more reliably calibrated than the core temp sensor, especially on certain AMD CPU model lines, the FX being one of them. Thubans and the old Athlon 64 X2 Brisbane core being others. I always like to check one against the other. With good aftermarket air cooling the CPU temps typically are about 10c higher than the core temps. With good water there is a little more differential I think. When I see huge differentials between core temp and CPU temp I suspect the core temp sensor is poorly calibrated.

 

[Vic Velcro]

According to somebody very official at AMD, Core temp is what matters on the FX series because the CPU temp reading is absolutely not accurate.

Apparently, the actual physical temperature of the cores and/or the CPU case (the can around the chip, not the tower that gear is mounted to) does not matter. The junction temperatures are the important thing. Junction temps are calculated somehow, and are considered by AMD to be the final answer for the FX series.

http://www.overclock.net/t/1128821/amd-temp-information-and-guide#post_15113829

 

[Vic Velcro]

Updated info after trying some different settings in BIOS.

Without LLC, system freezes, screen freaks out, then goes BSOD immediately upon attempting prime95. Manually upping VCORE does not help at all. I tried manual VCOREs as high as 1.44V, but the voltage always reads 1.308V 'actual' in the BIOS, in CPUZ, and in HWMonitor - regardless of what is set and locked in the BIOS.

When I enable LLC again, I can set VCORE in BIOS to 1.33v and prime runs fine up to 38 minutes (still counting). This gets me 50C Core temps on all 8 cores. CPU temp shows 66C in HWMonitor, but my IR gun and a physical probe show the temperature of the surface of the processor chip to be 52C (AMD guy appears to know what he's talking about). Monitoring software shows the BIOS selected voltage, not 1.308v as before.

So, with this combination of hardware (at least my specific components), Load Line Calibration is a 'must-have'. Also, it appears that Core temps and not CPU temps are indeed the accurate way to see exactly what's what. My guess is that the motherboard socket sensor does not align to an actual spot on the cpu can where a core makes internal contact and also that the Tjunctions where molecular temperature exists happen to not be the colloquial 'temperature' we think of as hot or cold. Checking definition of 'temperature' defines it as the rate at which atoms vibrate. Thus, thermometers and sensors don't apply here, but the algorithm AMD incorporates appears to be proper.

One more observation, whatever it's worth - if I choose anything other than stock settings for the CPU, the memory timing '2T' disappears from CPUZ, but BIOS shows '2T' for the timings when I go back in and check.