|
You could also unlock the 6750s into 6950s? IIRC, I think I have the model numbers wrong but it was more shader units. The Fury also unlocked into the Fury X. Both were silicon lotto affairs and had the same stability issues when they were new. Once the drivers matured it definitely worked better in both cases E: I don't think you're actually unlocking anything with that bios flash, it's just changing the clocks/voltages? New Zealand can eat me fucked around with this message at 18:31 on Feb 13, 2018 |
#
?
Feb 13, 2018 18:28
|
|
|
|
| # ? Apr 16, 2024 19:57 |
|
|
VelociBacon posted:For only 5-10% increase that’s going to be super unstable. I think the 9600->9800 involved desoldering a pin or something. Ah thanks! This seemed like the reasonable thought to me, glad to know I'm not being too cautious  I'm at home sick today so I'm playing around with undervolting and underclocking to get the card to maintain boost 100% of the time. Shifted from a State 7 Max of 1333 MHz/1150 mV (stock overclock from the manufacturer) to a State 7 Max of 1265 MHz/1020 mV and it's now holding boost 100% of the time. Firestrike overall score went from 10,396 to 10,580 by undervolting/underclocking from the manufacturer overclock. I guess I just need to find the point between 1333/1150 and 1265/1020 to get the max performance? I've also heard people on other overclocking forums talking about ASIC quality. GPU-Z reports mine as 86.9% but I have no idea what that means in terms of overclocking/undervolting stuff. Says its higher than 95.3% of the "similar" GPUs in its database. edit: Is there a better way to check if a card holds boost other than keeping WattMan up in another window and checking the GPU clock graph? axeil fucked around with this message at 18:38 on Feb 13, 2018 |
|
#
?
Feb 13, 2018 18:30
|
|
|
Do you mean your overall score for Firestrike? I only get ~7100 in combined with an 1800X and a Vega 64 (but 19,500 overall) Otherwise I'm buying some RX 580s lol
|
|
#
?
Feb 13, 2018 18:34
|
|
|
New Zealand can eat me posted:Do you mean your overall score for Firestrike? I only get ~7100 in combined with an 1800X and a Vega 64 (but 19,500 overall) Sorry, overall score not combined. Combined is 4,706 for the out-of-the-box settings with the power limit set to 50% and 4,638 for the undervolt.
|
|
#
?
Feb 13, 2018 18:38
|
|
|
9500s could unlock 4 ROPs to become 9500pros or 9700s, depending on what card you started with. 9700s could be driver modded to 9800s for better driver support because 9700 driver support got dropped quickly after the 9800 came out. X800s could unlock ROPs on a large number of cards to become a large number other cards. 6950s could unlock shader cores to become 6970s, same for a lot of AMD cards after that, 7950s, 290s, Furies. No soldering required for any of these mods. For Nvidia cards you could turn GTXes into Teslas or Quadros with soldering and bios flashing. The resistor network on the card and the bios had to agree what type of card it was so you flashed it, bricking it, then resoldered resistors to match what card the bios said it was and it worked again. No unlocking like secret GK200s or anything, the chips had to match from the consumer card you were coming from. For AMD cards it was usually an all or nothing thing. If the disabled parts work they work 100%, if the disabled parts are faulty nothing brings them back. It's like society was conspiring against me talking about modding old AMD cards. I was interrupted 5 times as I was writing this post.
|
|
#
?
Feb 13, 2018 20:04
|
|
|
Here's what I managed to pull off for my RX480. Putting State 7 at 1075 mV got a better result in Firestrike but it crashed Heaven so I classified it as unstable. Max power at load went from 193 watts  to 170 watts Firestrike results Initial (1330 MHz/1150 mV)  Final Stable Settings (1305 MHz/1080 mV)  Heaven Results (didn't take an initial as I used it for stability tests only)  Thoughts? I did a version with higher level fans (3200 rpm) to try and maintain state 7 longer but I only saw about a 1-2 fps difference with a lot more noise. This setting still doesn't maintain boost 100% of the time but the only setups I got that managed that were either unstable or much slower. axeil fucked around with this message at 22:00 on Feb 13, 2018 |
|
#
?
Feb 13, 2018 21:55
|
|
|
I had mentioned earlier I just put a basic OC on my 8600k and was impressed at 5.0 ghz @ 1.29 vcore. However I noticed issues I've never seen in prime95. Very quickly 2 of my cores will fail giving me a message about unexpected rounding error (0.5 expecting <0.4). Maybe 5-10 minutes in they will all fail the same way. To me this sounded like an outright stability failure, though one I've never seen before despite using prime95 extensively years ago. So I bumped my vcore up and up all the way to 1.3750 with no change. So I started dropping clockspeeds and it would actually affect when they fail, but I could never get it to stop failing. The only way to make it go away was revert to stock settings. I thought perhaps my ram was causing math errors so I started bumping voltage there to no effect. I run the XMP profile, but its a modest 2666 mhz. Throughout all this I never actually crashed my computer unless I tried to OC above 5.0. I am wondering if there is another voltage setting that isn't playing nice. I am so thoroughly unfamiliar with the other settings these days. Its astonishing how much terminology has changed since just Haswell, and I'm much less inclined to go down that path as well. The other thing here is despite mining with it all night and most days, and playing games and some encoding and encryption, I haven't seen it crash at all. This is bizarre to me. Either something is failing in a way that just doesn't noticeably affect day to day use (that I can tell anyway) or I'm making prime95 fail for some reason. Does anybody have a clue?
|
|
#
?
Feb 13, 2018 22:34
|
|
|
I think prime85 is not considered a good/safe testing tool these days. I’d use OCCT.
|
|
#
?
Feb 13, 2018 23:04
|
|
|
It's the most extreme unrealistic test you'll ever see. If you're cool in Prime you're cool in everything. Up to you what level of stability you want. It's at a point where being Prime stable isn't a baseline anymore though, you can be unstable in Prime and otherwise fine.
|
|
#
?
Feb 13, 2018 23:09
|
|
|
Good enough for me lol. Thanks. Its a weird one to me because I've never had a stable system fail prime95, but failures always coincide with crashing at some point as well. I have had prime95 work all night and crash the first time I load Battlefield 4 or whatever it was at the time, so I have it in my head the opposite way where prime95 passing doesnt indicate stability but prime95 failing definitely indicates instability. But I guess that was dumb! I used to use OCCT too. And XTU, and linpack straight, and some command line tester. I also used to run linpack (or variant) with Heaven, 3dmark, and that spinning doughnut one at the same time. Those were the days...
|
|
#
?
Feb 13, 2018 23:45
|
|
|
Is there such a thing as a big sheet of fan filter material I can buy & cut up as needed ? I have an old school Coolermaster Stacker 830 case and all the front bay blank plates have a grill that holds a small bit of filter material. The problem is the material is poo poo because the edges are not "sealed" and the stuff literally starts coming apart if I brush it/aircan it too hard. Picture to illustrate: The unraveling should be noticeable around all four edges. The material is 1 3/8 inches by 5 3/4 inches by guessing 1/10th of a mm in thickness (slightly thicker than a piece of paper) in size, have 7 to replace. Something I could wash in a sink would be a plus.
|
|
#
?
Feb 19, 2018 23:35
|
|
|
MREBoy posted:Is there such a thing as a big sheet of fan filter material I can buy & cut up as needed ? I have an old school Coolermaster Stacker 830 case and all the front bay blank plates have a grill that holds a small bit of filter material. The problem is the material is poo poo because the edges are not "sealed" and the stuff literally starts coming apart if I brush it/aircan it too hard. Picture to illustrate: Search amazon for fan mesh
|
|
#
?
Feb 19, 2018 23:38
|
|
|
Hello goons. I am new to this thread, but I think this is the right place for this question. I just put together a: Asus Z370-E lntel 8770K Noctua NH-D15 SSO2 32gb 3600 RAM Just 2 case fans right now I noticed last night it was running at 75c in Overwatch, which threw up a red flag for me because of that beastly cooler I installed. Is this expected with MCE on? Should I turn it off? I am getting mixed and confused messages from randomly googled threads. Does this indicate that my cooler is installed wrong? I've seen some people reporting ~60c under stress test with similar settings.
|
|
#
?
Feb 20, 2018 21:11
|
|
|
That seems fine and isn’t too bad a temp, what do you get with other games?
|
|
#
?
Feb 20, 2018 21:35
|
|
|
Multi-core enhancements tends to shove a metric fucktonne (this is a scientific term, I assure you) of voltage where said fucktonne of voltage is not necessarily required. If you force an all-core turbo clock speed yourself and lower the voltages to wherever it's stable at you'll get better temperatures at the expense of spending more time on it.
|
|
#
?
Feb 20, 2018 21:36
|
|
|
Kazinsal posted:Multi-core enhancements tends to shove a metric fucktonne (this is a scientific term, I assure you) of voltage where said fucktonne of voltage is not necessarily required. If you force an all-core turbo clock speed yourself and lower the voltages to wherever it's stable at you'll get better temperatures at the expense of spending more time on it. Just to add to this, an easy way to do this while retaining c-state stuff is to just use a voltage offset. Just keep pushing it lower and play quick play or whatever in Overwatch to stress test it.
|
|
#
?
Feb 20, 2018 21:38
|
|
|
Other... games? Seriously though. That is the only piece of software I have installed so far, and I'm a terrible addict. Edit: Got it. I can't wait to get home and try it. Thanks for the replies. headcase fucked around with this message at 21:40 on Feb 20, 2018 |
|
#
?
Feb 20, 2018 21:38
|
|
|
Kazinsal posted:Multi-core enhancements tends to shove a metric fucktonne (this is a scientific term, I assure you) of voltage where said fucktonne of voltage is not necessarily required. If you force an all-core turbo clock speed yourself and lower the voltages to wherever it's stable at you'll get better temperatures at the expense of spending more time on it. Is there any reason to force an all core turbo vs letting Intel manage it?
|
|
#
?
Feb 20, 2018 21:42
|
|
|
VelociBacon posted:Just to add to this, an easy way to do this while retaining c-state stuff is to just use a voltage offset. Just keep pushing it lower and play quick play or whatever in Overwatch to stress test it. headcase posted:Is there any reason to force an all core turbo vs letting Intel manage it? Yeah sorry I meant "synchronize all turbo clocks at 4.7 GHz or whatever so it doesn't step down to 4.3 GHz when it's using all six cores", not like "never not 4.7 GHz". Bad phrasing on my part. Kazinsal fucked around with this message at 21:47 on Feb 20, 2018 |
|
#
?
Feb 20, 2018 21:43
|
|
|
headcase posted:Hello goons. I am new to this thread, but I think this is the right place for this question. I just put together a: I had a similar problem with a beastly cooler. Did you remember to take off the clear plastic sticker from the coolers coldplate? I know you are sure you did, but did you really?
|
|
#
?
Feb 20, 2018 23:28
|
|
|
heh... there was a hard plastic piece on it, but I don't remember a sticker. Maybe I should check. edit: Oh haha. Found the problem. Overwatch has a monitor for GPU temp not CPU. My CPU is sitting at 34 in overwatch according to Asus' monitor so all is well. That is stock speeds with MCE off. headcase fucked around with this message at 01:10 on Feb 21, 2018 |
|
#
?
Feb 21, 2018 00:52
|
|
|
I was having excessive temperatures on my stock 8700K on the Asus Prime A. The fans would spin up doing mundane stuff like downloading through Steam. Idle temperatures were fine so I looked into it and it seemed like the automatic voltages were cranking the Vcore way too high, peaking in the mid 1.4 and usually past 1.35. After screwing with the settings I set a manual of 1.25 and the difference is enormous. It's shaved about 30% off the temperatures. IntelBurnTest would hit 96c on the high setting, now it hits 71c. It's sitting at 1.248 Vcore. Only problem is I don't know how to do this in a way that drops Vcore when not under load. I think I want adaptive but when I tried that it was still way too high. This motherboard has VID and LLC options, I think it's ignoring VID in manual, I put LLC on level 2 but I'm not sure if it's doing anything, lowest Vcore on Manual recorded is 1.136 so I think I might want to bump it up? No stability issues yet. Carecat fucked around with this message at 02:08 on Feb 26, 2018 |
|
#
?
Feb 26, 2018 00:24
|
|
|
Carecat posted:I was having excessive temperatures on my stock 8700K on the Asus Prime A. The fans would spin up doing mundane stuff like downloading through Steam. Idle temperatures were fine so I looked into it and it seemed like the automatic voltages were cranking the Vcore way too high, peaking in the mid 1.4 and usually past 1.35. After screwing with the settings I set a manual of 1.25 and the difference is enormous. It's shaved about 30% off the temperatures. IntelBurnTest would hit 96c on the high setting, now it hits 71c. It's sitting at 1.248 Vcore. I had this problem and ended up using offset setting. Make sure c states are enabled also.
|
|
#
?
Feb 26, 2018 01:12
|
|
|
As far as I know they are on. Didn't touch any of those settings, wherever they are hidden. I tried Adaptive which did drop the Vcore in idle but it kept cranking it up to 1.4 even if the Turbo target was 1.2 and offset was 0.001, which for some reason is the minimum outside the mysterious "Auto". I don't know if that was related to VID, maybe I need to change VID from auto to disabled. Offset sounds like it's slightly inferior to Adaptive but I might have to take it if I can't figure out how to get adaptive to do anything useful.
|
|
#
?
Feb 26, 2018 01:57
|
|
|
Carecat posted:As far as I know they are on. Didn't touch any of those settings, wherever they are hidden. Honestly offset solved my problems 100%. I wasn't hoping to use it either but it works great. As far as I can tell it's only affecting the max vcore, it's super stable, I'm happy. e: Just set the offset negative to whatever amount to drop the vcore to 1.25 under load if that's what you found works well.
|
|
#
?
Feb 26, 2018 02:30
|
|
|
Mobo makers should just let us set custom CPU vcore/freq curves like for Pascal GPUs than the still half-assed vcore methods in loving 2018.
|
|
#
?
Feb 26, 2018 04:19
|
|
|
VelociBacon posted:I think prime85 is not considered a good/safe testing tool these days. I’d use OCCT. Buildzoid was also not so hot on P95 in one of his videos, he uses an hour of Intel Burn Test. If it’s good enough for that guy, it’s good enough for me. But holy poo poo do some people defend P95 to the death. Like, okay, enjoy possibly degrading components during your stability test.
|
|
#
?
Feb 26, 2018 05:43
|
|
|
Craptacular! posted:Buildzoid was also not so hot on P95 in one of his videos, he uses an hour of Intel Burn Test. If it’s good enough for that guy, it’s good enough for me. I'm also not someone who is willing to run the extra voltage if the options are: a) stable in every real world situation b) stable in every real world situation and also stable in synthetic benchmarks If it means I'm running more than 0.2v more I'm not really keen on it. Give me the lower temps instead pls.
|
|
#
?
Feb 26, 2018 07:35
|
|
|
VelociBacon posted:I'm also not someone who is willing to run the extra voltage if the options are: I've never overclocked poo poo and shied away from OCing this 3770K for five years, but one thing that annoys the hell out of me is that my Asus motherboard apparently ups the voltage on it's own in response to your changing the multiplier, and people five years ago thought this was a good thing. And nothing written in the time when this thing was new and anybody cared even bothered to mention VRMs, which I don't know if that means the heatsinks were better then or what.
|
|
#
?
Feb 26, 2018 08:38
|
|
|
Craptacular! posted:I've never overclocked poo poo and shied away from OCing this 3770K for five years, but one thing that annoys the hell out of me is that my Asus motherboard apparently ups the voltage on it's own in response to your changing the multiplier, and people five years ago thought this was a good thing. And nothing written in the time when this thing was new and anybody cared even bothered to mention VRMs, which I don't know if that means the heatsinks were better then or what. They overvolt poo poo like VCCIO to the loving moon if you set high memory clocks sometimes.
|
|
#
?
Feb 26, 2018 10:29
|
|
|
Craptacular! posted:I've never overclocked poo poo and shied away from OCing this 3770K for five years, but one thing that annoys the hell out of me is that my Asus motherboard apparently ups the voltage on it's own in response to your changing the multiplier, and people five years ago thought this was a good thing. And nothing written in the time when this thing was new and anybody cared even bothered to mention VRMs, which I don't know if that means the heatsinks were better then or what. Intel stock VIDs scales exponentially with clocks as part of the design. The stupid thing is the mobo makers add even more volts with default auto mode on top of that to sell the "OCing is so easy" meme to the masses who wouldn't know they are running hilariously overvolted CPUs for 10% more performance at 100% more power draw. Palladium fucked around with this message at 11:36 on Feb 26, 2018 |
|
#
?
Feb 26, 2018 11:34
|
|
|
GRINDCORE MEGGIDO posted:They overvolt poo poo like VCCIO to the loving moon if you set high memory clocks sometimes. I checked and it looked like mine was quite high at 1.3 something. Turned VCCIO, SA and standby down to 1.0. Current, Min, Max, Average with a few runs of IBT. 
|
|
#
?
Feb 26, 2018 12:08
|
|
|
Palladium posted:Intel stock VIDs scales exponentially with clocks as part of the design. The stupid thing is the mobo makers add even more volts with default auto mode on top of that to sell the "OCing is so easy" meme to the masses who wouldn't know they are running hilariously overvolted CPUs for 10% more performance at 100% more power draw. As I do want an easy OC that is better conservative than sorry, wondering if I shouldn’t use Intel XTU instead of toying with Asus’s bullshit.
|
|
#
?
Feb 26, 2018 15:16
|
|
|
So I am running an 8700K on a Asus Z370-E. It runs really cool: 29 idle and less than 50 at FPS load. I decided to push the multiplier up a bit so I did: *MCE off *sync all cores *48 multiplier *1.25v manual static voltage My game froze within 5 minutes, but the temp never exceeded 50c. Should I just assume that is a voltage supply issue? Does that mean I should: Just stick with 47 because EH what's a few more cycles anyway? Go back to adaptive voltage and let it have it's 1.4v sometimes? Try manual 1.3v? Look for some other setting that must be wrong? It's been rock solid with default automatic settings (MCE off).
|
|
#
?
Feb 26, 2018 16:06
|
|
|
Sync all cores at 4.7 and try again. 4.8 is starting to push a bit but most CPUs should be able to handle that. You might have a poor OCer. Alternatively, start at 1.35v and work down. You really can't hurt your CPU, thermal throttling will keep it well within parameters. Trying the "inch up" approach just means you'll have to deal with more crashes.
|
|
#
?
Feb 26, 2018 16:21
|
|
|
I didn't get into this earlier when it was mentioned to me but since it seems to be common knowledge now, whats wrong with P95? Especially relative to IBT? Its funny because in the past IBT was looked at with suspicion rather than P95. Afaik small FFT P95 and linpack are virtually identical workloads on memory and CPU, though they are solving different kinds of math, with IBT being clearly more stressful. Whats changed?
|
|
#
?
Feb 26, 2018 16:28
|
|
|
1gnoirents posted:I didn't get into this earlier when it was mentioned to me but since it seems to be common knowledge now, whats wrong with P95? Especially relative to IBT? Its funny because in the past IBT was looked at with suspicion rather than P95. Afaik small FFT P95 and linpack are virtually identical workloads on memory and CPU, though they are solving different kinds of math, with IBT being clearly more stressful. Whats changed? I don't know about relative to IBT, but P95's AVX stresses ago well beyond real-word situations (today). Since AVX is a specific outlier that runs up heat I think its warranted that a special caveat is given when running tests on it. It's much more likely with the newer architecture that you can have a system that runs cool and stable on literally everything but running AVX full bore for 20 minutes straight. If you are trying to test for workstation-level stability, that isn't super realistic. Basically, you want a stability test that is outside normal working conditions, but the AVX testing is now so far outside the regular performance curve it makes sense to work around it.
|
|
#
?
Feb 26, 2018 17:00
|
|
|
Also AVX Offset can be used to address instability if you do crash in any AVX applications.
|
|
#
?
Feb 26, 2018 22:12
|
|
|
Carecat posted:I checked and it looked like mine was quite high at 1.3 something. Looking good.
|
|
#
?
Feb 26, 2018 22:33
|
|
|
|
| # ? Apr 16, 2024 19:57 |
|
|
Does anybody else have problems with the AVX offset kicking in non-AVX programs such as games? If run my 8700K at 5.0 GHz with a -3 offset my average frequency even in old titles like Starcraft 2 is something around 4.85 GHz. It's definitely not VRM or power limit throttling. My workaround is to run with zero offset which is stable in everything including Realbench and Blender but not Prime95 AVX with Small FFTs or LinX. (which are both fairly unrealistic anyway). By the way here's a nice graph of 8700K OC scaling. German but the pictures are self-explanatory, sensible OC targets should be 4.7 or 4.8. https://www.hardwareluxx.de/index.php/artikel/hardware/prozessoren/44780-coffee-lake-overclocking-check.html?start=7
|
|
#
?
Feb 26, 2018 23:22
|
|
