Originally Posted by Cyro999
I did, it passed completely fine (well, i didn't reflash bios)
I primed for 14 hours with 7000mb RAM, error checking, rounding etc whatever enabled, on 1.34v 4.7, before i applied LLC extreme on VRIN (which made me need less volts for everything) but x264 failed in under a minute when i stopped priming at 14 hours. It's just like "oh, why did i bother doing that?"
Which version would you reccomend? I'l have to go borrow a pen drive or something from someone. You think it'd make a difference?
And x264 doesn't use the igpu - i disabled it anyhow, to see if it helped, it didn't seem to change anything
It's not really the volts that frustrates me, it's that i have to add a ton on top of an OC that seems completely rock solid in everything (i 14 hour primed with 7000mb RAM on 5 mins per fft lengh for a reason) and then just instantly fails x264 unless i add 0.02-0.05v on top of what i need to pass everything else. With my last CPU, x264 held up completely fine on volts that had no chance of passing prime. My temps will be a lot cooler than stress test, but i dont really wanna run 1.345v or 1.35v for 4.7, when i can run 1.19v and pass 12 hour prime on 4.5, it just doesnt seem right
Try reflash the BIOS, I am thinking perhaps your ME firmware is messed up. Also check online see if there is some kind of issue with x264 and haswell, can you try to run it at stock and see if it works?
Also try to reflash your BIOS, it might help. BTW i thought the iGPU was better for transcoding than the CPU?
Originally Posted by ABAD1DEA
i dont know but its really hot in my room
Sin i tried 1.48Vcore and 2.3 VRIN and still cinebench freezes.
maybe i have to wait for a better bios or my cpu is damaged or my UD4 mainboard is broken
I doubt it is the board or BIOS
, i know you want it to be because you delidded your CPU and it might be hard to get it replaced, but it isn't. I tested your board against all my other boards, I have an ASUS board sitting here as well and the CPU clocks are the same. Unless the board is defect i highly doubt it is the board. You can try to reflash the BIOS and see if it works, but otherwise i would say CPUs have limits. Sorry
Ther is 1 thing left to do(toss the CPU out the window J/K) you can cool the CPU down better and i promise you will see better results. Also CPus degrade and cold temperatures help. You can look into peltier blocks, they are cheaper than phase change and easy to work with, they just need PSu power, but modern PSus are so high on the rails that you shouldn't need an extra PSU. Your OC will increase with a 30C drop as well and a peltier block might cost you $20 for a 200W or 300W or 400W one, and then all you need to do is cool it with a water cooling rig and do a little insulation.
You can also start from scratch again, and try auto with everything and then just vcore, fresh BIOS flash to make sure everything is good, turn down the CPU Uncore and the memory multiplier an clock the CPU, then increase the memory and then the uncore. But staiblize one after another.
Originally Posted by Nuck
- 4.6 GHz @ 1.265V core, Vrin auto (I think 1.8-1.9V?)
- RAM is at 10-12-12-31 2400MHz 1.65V
- Uncore frequency 4.4 GHz, Vring 1.15V
Do you think it's worth it for me to try dropping my uncore and raising my multiplier to try to squeeze out ~4.8-4.9 with slightly more Vcore (but no more than 1.3V)? I know I would probably have to fiddle with the IMC voltage settings too, but I was wondering if I would see an overall performance boost.
I think your numbers are great, really better than mine for sure!
I don't think youll need to mess with IMC voltages, but yea you might see a performance boost from raise in CPU speed rather than Uncore, but for best performance keep uncore 300-500 below at least. Really best is 1:1 but i think you will see more rains in CPU speed increase.
Originally Posted by Terraformer
Wait a sec...
I did some benchmarking on different dividers of Core and Uncore.
With understanding that the ring is the connecting highway between the components, and by setting Uncore multi, you set the speed at which that highway operates. If I am wrong then correct me pls
But then look at this :
Here you see the multipliers, so 35 multi means 3.5 Ghz if we have a 100 bclk.
L1 L2 L3 Cache in GB/sec.
An OC of a core multi of 44 compared to a stock speed of 35 multi, puts more strain on the ring with the same multi of the uncore of 35. You can see that increase of strain on the ring by looking at the L3 Cache bandwidth that has increased from just upping the Core Multi only.
A difference of 226GB/sec -188GB/sec = 38GB/sec (almost 20% faster) that the ring has to deal with at that higher OC even with the Uncore multi left at 35.
That means that the ring with its Vring of still at default 1.050v has to perform 20% more at the same voltage.
It is like having to do 20% more work without getting a raise...
This means simply that even if you "up" your Core Multi, and leave your Uncore Multi.. you need to add more voltage to the ring, so up Vring too.
And this also means, that if you have a stable Uncore multi at some Core Multi, it won't mean that if you change the Core Multi, the same Uncore Multi remains stable... Which complicates things a bit
I hope you understand what i am saying, and maybe I'm completely wrong in this but it seems logical to me.
Hey an yesterday I was doing testing to test whether uncore above the CPU Core ratio would help performance, I can show you guys the results, I ran SuperPI 8M 36 times, twice on each run with different CPU speed and memory speeds.
Take a look, with SUPERPI LOWER IS BETTER!!!!:
The bes tis at CPU and Uncore 1:1 but the difference in time is so small before reaching the CPu core ratio, there is almost very little gain b/c of the heat and voltage you need to do 1:1 ratio of uncore to core
here are both sets of results together, but with the numbers in the graph:
Yes i have very little to no life, i know, no need to say it.