1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

The AVX Offset Problem

Discussion in 'Processors and motherboards Intel' started by krazymagic, Oct 8, 2019.

  1. krazymagic

    krazymagic Member

    Messages:
    10
    Likes Received:
    2
    GPU:
    MSI RTX 2080
    I recently decided to embark on tweaking my slightly older system the last few days. It is an i7 7700k, MSI Z270 Mobo, MSI RTX 2080. Back when I first built this system (I was one of those who bought in to a 7700k build right before 8700k showed up, so that was annoying) but I stopped my overclock on the 7700k at 4.8ghz entirely because I couldnt get Prime 95 stability due to AVX instructions causing failures beyond 4.8ghz. I did not want to run an AVX offset because I always noted in my testing that the cpu would toggle to the avx offset constantly in games, often ping ponging between my real frequency and the avx offset. I eventually just said screw it, set it at 4.8, dialed in the voltage and left it alone. That was over 1 year ago, and at that time 4.8ghz and a 1080p monitor it did all I needed honestly. Now I am running 3440x1440 ultrawide, an RTX 2080 that I added several months ago, and a brand new kit of 32GB samsung B-die memory arriving in the mail today (Before its fully discontinued everywhere) so Ive decided to revisit overclocking my system to try and pull every drop of performance out of it.

    At 3440x1440 resolution, with graphics settings on the high side in most games, the gap between my 7700k and say, the 9700k at the same clock speeds is actually pretty tight. My pixel count is over that of 1440p, but less than full 4K res, but its enough pixel density that my RTX 2080 (2025 core / 8100 memory) tends to be the bottleneck in most games, and when games bottleneck on the gpu as most of yow know, the difference between a 7700k , an 8700k, or a 9700k are pretty small in benchmarks given they are all running at the same speed, etc. I say all that to say that I just cant see myself investing in a new mobo and cpu at this time, but I wanted to get this 7700k to 5GHz.

    So to the main point here - I feel like Ive actually got a great 7700k, The avx issue is just really annoying me and Ill go into why here in a second. I have this 7700k right now running at 5ghz, and it is rock stable everywhere. I can game on it for hours, and have done so. I can run Intel Xtreme Tuning utility stress for hours, heavenbench for hours, 3d mark stress for hours, the old, non AVX prime 95 for hours, and it passes everything. My voltage in bios is set to 1.285 adaptive, so it swings from that 1.285 up to around 1.315v depending on load, but that voltage at 5ghz on this chip from what ive seen is pretty impressive. My problem is that as soon as I run the new versions of prime, that hammer avx instructions, 2 or 3 of the cores fail within seconds. So far I have ignored it because I am so stable in everything else as I mentioned. I also really hate the avx offset because I feel that it triggers so often that it almost negates my overclock to 5ghz, and here in lies my AVX Offset dilemma. Some games, like AC Origins will run 100% of the time at the avx offset if it is set. This annoys me, because at my full 5ghz clock it runs completely stable. I also find it odd that prime 95 will fail, as it also forces my cpu to declock due to current throttling. I can bump the amperage throttle up to prevent the throttling but this obviously puts the cpu in dangerous temperature ranges of right around 100c so then I hit thermal throttling. Logically, I would expect the throttle that is occuring to prevent the prime 95 stress from failing because the throttle is pretty hard, from 5ghz to down to 4.2 - 4.4ghz, yet several workers still fail. If I declock my cpu to 4.8ghz though I can run the same test all day long without fail, so I suspect the failure must be occurring just before the declock is occuring.

    So I step back and look at the situation I got here. Part of me almost says I should just ignore prime 95. It puts the most insane load on the cpu - 95c temps, hitting current limit, all due to the massive string of avx instructions. No software I will ever run will hammer the cpu this hard, so I tell myself there is no reason to hurt my performance just to pass this one test - but it still bothers me. The little intel extreme burn in utility which also hammers avx instructions is also a hard hitting stress test for me, it also throttles the cpu and ramps up the temperatues, but it never fails a pass, likely because the cpu ends up running below 4.5ghz, just as it does with prime 95. In all the other stress tests and bencharks I have run at 5ghz I pass with no issues. Im just curious how others have dealt with this issue. Im sure others have run into this exact scenario. I feel like I might could even get this chip to 5.1ghz. If I could do that, I wouldnt mind setting an avx offset of 1 in that case to maximize stability, but I really hate an avx offset that puts me at 4.8ghz, simply because of how often it seems to trigger in random games - almost as if there is some bug with it.
     
  2. Kolt

    Kolt Ancient Guru

    Messages:
    1,601
    Likes Received:
    439
    GPU:
    RTX 2080 OC
    Tough one. I've been exactly where you are at, and in the end I ultimately decided to get the system as stable as possible while ignoring Prime95 AVX. It was just too intensive and added way more heat than anything in the "real world", including errors that I couldn't reproduce elsewhere. My opinion is to do just that. Remove the AVX offset (for the reasons you mentioned above) and focus on stability from other sources (Intel xtreme utility for example: https://downloadcenter.intel.com/download/24075/Intel-Extreme-Tuning-Utility-Intel-XTU- ) or even the non AVX Prime95 (v226)

    The reality is.. if everything is completely stable and the only thing that ever gives you a problem is the newer AVX Prime95s then just ignore it. Go play a few hours of BF5 and then check for whea errors in your event viewer. That honestly did a better job of identifying CPU OC issues than messing with AVX p95 (Kind of a joke but true at the same time. BF5 was an amazing game to use as far as stability testing on my system. That game seems to be very sensitive to unstable OCs and will let you know real quick.)

    I'll just cover my butt here and point out that I am indeed telling you to ignore a potential instability, but it's my opinion that the instability you are facing is something that probably won't be a real world problem or cause instability in anything you would probably use. I may be wrong here, but this is my experience with the same problem and ultimately how I decided to ignore it. My current system has been extremely stable for absolutely everything for a long time now, except I am pretty sure if I run AVX P95 it will give errors like you are having (currently 4.9ghz, 1.3v, 8700k. Can also do 5ghz @ 1.34~ but I don't like the jump in voltage for just 100mhz).

    Worst case, you start to crash in games and need to add your AVX or mess with other settings. Best case, you don't have any problems and don't need to worry about it. At least you'll have a good idea of where to look.

    EDIT: I am currently AVX 1, but your post is making me reconsider removing the AVX 1. I always did it for heat reasons, but I've noticed some non AVX games bugging out and dropping to AVX speeds so might as well.
     
    Last edited: Oct 8, 2019
  3. krazymagic

    krazymagic Member

    Messages:
    10
    Likes Received:
    2
    GPU:
    MSI RTX 2080
    Enjoyed reading the well reasoned reply kolt. BF5 is indeed one of the games I played when testing my overclock. I played the new map that just released for nearly 2 hours so far and never had any crash or issues. I have also played cod bo4 for several hours. Those are the only two games so far I have put substantial time into with the 5ghz overclock, as I just did it this past weekend. Ive had no crashes. I did have some stress failures in the intel extreme tuning utilities stress test which i suspected to be a low voltage issue, so I bumped voltage from 1.280 to 1.285 and have not had another error in that stress test. I noticed that at 1.280 with variable set in bios my voltage would occasionaly go below 1.280v. Bumping to 1.285 fixed that. The voltage required for 5ghz is what has really amazed me with my 7700k so far. Ive often seen where folks needed 1.375v to hit 5ghz on the 7700k. I have never ran this chip over 1.35v. I literally have it set in the bios at 1.285v and variable. In cod bo4 last night for example I watched my voltage as I played and it tended to swing up and sit at 1.31v under that load, when idle it goes back to the 1.285v or so. I did try only briefly to go to 5.1ghz on this same voltage and got into windows, but as soon as I fired up prime 95 v226 (non avs p95) my system BSOD within seconds. Very likely a voltage issue. I plan to continue to my tweaking to find the top end of this 7700k. Ive not read of very many going over 5.1ghz on it though so i know i gotta be right at its threshold. Like you, I would not want to run really high voltages on the cpu just for an extra 100mhz. If I could get 5.1ghz stable but had to run near 1.4vcore like Ive seen from some,I wouldnt do it. I should also add that last night I began increasing the ring ratio as well. through all my testing so far it has sat at 42, so last night I bumped it to 45 for a 4.5ghz cache frequency. It seems around 4.7ghz is the cieling for most 7700k's on the ring ratio frequency, and often yields miniscule performance increases at the cost of big instability if you get too high.

    I am kind of trending toward just ignoring the p95 avx instability. My theory is that under MOST real world scenarios that could ever hit the cpu as hard as p95 avx does, the cpu would end up being declocked due to current throttling, as I notice in the intel extreme burn in test utility (Not to be confused with the official intel extreme tuning utility) The Intel extreme burn in test utility uses avx aswell, and matches the load of p95 avx. As soon as I run it however, my cpu hits current throttling immediatly and drops to like 4.4ghz - the test runs and passes. As I mentioned I could increase the current limit to prevent this throttling but its almost a sort of shield for me against a potential avx instability I think. My theory is that any level of avx demand that would reach the level of causing errors should hopefully also hit the current throttle limit, as these 2 stress tests do, and once the cpu is throttled, any instability should be averted. The minor execution of some avx instructions, as we see in some games, like AC Origins which I have also ran, the cpu handles without any issue.It seems the only instability is in these Massive, high power load avx hammer tests that just slam the CPU with full on avx workloads like p95. If the avx instructions are mingled amongst other instructions, say in the course of gaming, then the cpu has no issue. I dont know of any software I would ever use that would match p95 avx workloads. I've never seen any game or legitimate real world software actually that hits the current throttle of the cpu either - you're talking about a lot of amperage there to hit that threshold. I have to admit though that the perfectionist in me does hate that i still fail this test. Ive been building and overclocking systems for over 20 years now, and Ive always been fairly meticulous in my appraoch. Ive always believed strongly in any overclock meeting a 100% stability mark, and It just nags me that im not able to hit that mark here. I just feel p95 is almost unreasonable in the amount of load it hits a cpu with, and does it with the power hungry avx instruction set in a way that no legimitate software would even be designed to do. That said however, I wouldn't change it either, because that is what a stress test SHOULD do, as that is the whole point of stress testing. I would still welcome other replies on the topic.I like just reading how other people approach this issue with avx offsets.
     
  4. Caesar

    Caesar Master Guru

    Messages:
    726
    Likes Received:
    259
    GPU:
    GTX 1070Ti Titanium

  5. Kolt

    Kolt Ancient Guru

    Messages:
    1,601
    Likes Received:
    439
    GPU:
    RTX 2080 OC
    No problem. I know it's not perfect at all. I'm certainly not the best OCer on here so definitely leave the door open for more suggestions. I just like to be real and realistically P95 AVX is the last thing I would care about if everything else is working fine.

    Aside from the obvious (more/less voltage and more/less ratio) you could look into vdrop and your LLC settings if you haven't already. Going by what you said "I did have some stress failures in the intel extreme tuning utilities stress test which i suspected to be a low voltage issue, so I bumped voltage from 1.280 to 1.285 and have not had another error in that stress test." you could have an issue with vdroop where you aren't getting enough voltage under that load of work. I'm not the best in this area so I'll let other Gurus chime in here.
     
  6. cleverjaja

    cleverjaja New Member

    Messages:
    1
    Likes Received:
    0
    GPU:
    1080 TI
    Your fine, reason why its unstable is because AXV is causing your CPU to use more watts than normal and your CPU cooler cant keep up. If you can run the new prime95 with avx unchecked for a few hours then your system is solid. Also run realbench too
     
  7. krazymagic

    krazymagic Member

    Messages:
    10
    Likes Received:
    2
    GPU:
    MSI RTX 2080
    Well in the few days since I made this post, I did something I rarely do.....I made a knee jerk decision and upgraded to an MSI MEG z390 ACE mobo and i7 9700k. I got it all put together yesterday and verified it all worked. This weekend I will begin overclocking this new system and see how it handles AVX. Im curious to see if it also has the same behaviour as my 7700k, where it ping pongs back and forth between avx clock and normal clock speed on the cpu in the middle of games and the like. I suspect it will.
     

Share This Page