HP OMEN 15 and 17 owners lounge

Discussion in 'HP' started by tweake628, Aug 23, 2017.

  1. Alexrose1uk

    Alexrose1uk Music, Media, Game

    Reputations:
    616
    Messages:
    2,324
    Likes Received:
    11
    Trophy Points:
    56
    After reasonable success without too much CPU throttling with the above I am now trying this, slightly less aggressive reduction (-3 on 1-4 cores active, -2 on 5-8 cores):
    Turbo ratio limit:
    1 Core active - 45
    2 Cores active - 45
    3 cores active - 44
    4 cores active - 43
    5 cores active -43
    6 cores active -42
    7 cores active -41
    8 cores active - 39


    The all core doesn't really seem to be the problem, its those low core count high clock speed/high voltage spikes, so going to give this a shot :)

    I might tinker with this further, maybe push 1-2 core back down to 44, but then 3-6 cores back up to 44, 7 to 42 and leave 8 core as 40/39. All a balance of heat vs CPU, especially now with GPU load at 190W.

    This CPU is GOING to get hot, up to PROCHOT most likely, but ideally I want it to run as high as it can as much as it can :)

    Not much else I can do the machine apart from lowering ambient temperature, physical case modifications (which would screw warranty) or better thermal paste (its MX4 atm, so Kryonaut or LM, but both of these have potential longevity issues).
     
    Last edited: Mar 12, 2021
  2. Burrick

    Burrick Notebook Geek

    Reputations:
    11
    Messages:
    96
    Likes Received:
    8
    Trophy Points:
    16
    The 14nm architecture is exponentially inefficient to get those max turbo speeds. The -H as opposed to the desktop processors are basically the same thing, except the base frequency is much lower to keep the profile at the 45W package profile (2.3 on the i9). All we want is the same profile as the desktop chip which is burned in to it, and it does work fine in this machine provided it's properly repasted and not concurrent with full GPU load. So you haven't repasted yours yet? I couldn't wait.
    It always breaks my heart to see so many people disable turbo to get this under control, like amputating an arm for a gangrenous pinkie. I'm curious why you're looking for a one-size fits all throttlestop profile, does it PROCHOT during regular work or just when the GPU is loaded, and you're trying to get max clocks during gaming for FPS purposes? Since the 4K is 60Hz and gsync I'm not in need of these 144+ framerates so I'm not chasing that, and the 2080 will drop due to game complexity before CPU limits it. I've got Performance, which is the max for the chip and runs fine for benchmarking, windows work, etc, and the game profile. I've added yours and will give it a try next time I game.
    Your revision should work well, after dealing with a non-flat heatsink and working around it, the 2X4 die has proven to be extremely problematic. I honestly think they'd get better results with a IHS on these, soldered like the new ones and why our 50% more powerful Nvidia chips stay at 70 C. That pinprick of 20+ watts once it gets above 4 GHz needs someplace to go immediately. I've achieved the same functionality with a copper shim, floating in kryonaut to compensate for the uneven heatsink contact. It's lasted a year so far.
     
  3. Alexrose1uk

    Alexrose1uk Music, Media, Game

    Reputations:
    616
    Messages:
    2,324
    Likes Received:
    11
    Trophy Points:
    56
    I've gone to MX4, but not Kryonaut with its pump out issues, or Liquid Metal with its...many...issues, especially as I have a multi year extended warranty and I want the machine as simple as possible to return to stock if necessary.

    As you say, I really don't want to go as far as disabling turbo, or locking it down to a 3.6/3.8GHz all core, as you're right, that defeats the point of having such a powerful machine and hurts me a little inside, I don't mind docking a few bins just to allow extra leeway for the extra heat produced by 40W extra on the GPU though :D

    Typically my machine only protchots when its under extreme stress test, or heavy GPU and CPU load, for day to day general usage its absolutely fine, but I've always been the sort of person who prefers a one-size-fits-all, 24/7 overclock over going for 'race' runs that are unstable; but I also want to extract as much performance as possible from the machine, and reducing all the Turbo levels to say 4GHz seems unnecessary, I am trying to see where I can find a nice balance!

    At the moment I actually am trying out the 1-2 core set to 45, 3-6 core set to 44, 7 to 42 and 8 set to 40, I reckon this is somewhere around the sweet spot, although this may be a little aggressive. I'm working to find that balance between doesn't throttle or throttle often, and maximum power though, because once I am there; I will literally just leave it there; I really want to find a nice balance between throttling and maximum performance for games though; max on CPU AND GPU.

    Up until recently I'd never docked the CPU performance at all, I'd simply unlocked the power limits to 90-100w, upped the turbo time limits to infinite, and then undervolted to allow it to sit there more of the time, but with the extra heat generated by the 190W VBIOS (GPU now goes up to mid-late 70s and clocks higher), I had to take a more pragmatic look at the CPU.

    Realistically the difference between 4.7-4.8 being rarely triggered, and being locked down to 4.4/4,5 over 1-2 cores is fairly low, but the difference on locking down to say 3.6/3.8 for any number of cores...that's more substantial. And unnecessary unless I was chasing optimal temperatures.

    As it is, I want to find the best balance with my machine and GPU/CPU for maximum performance, and minimal throttling. My office gets quite warm in summer, so now the weather is starting to get warmer again (and my MX4 has had a few months to cure), its a good time to find a balance.


    Plus it gives me a chance to update my signature performance figures haha!
     
    Last edited: Mar 12, 2021
  4. Alexrose1uk

    Alexrose1uk Music, Media, Game

    Reputations:
    616
    Messages:
    2,324
    Likes Received:
    11
    Trophy Points:
    56
    Small additional point I forgot to mention - ENABLE Hardware Accelerated Graphics Scheduling in Windows 10.
    It doesn't make much of a difference performance wise, either good OR bad at the moment but in some circumstances this reduces CPU load considerably (5-20%) due to the amount of work Nvidia tends to do in the GPU driver; which in turn helps reduce heat output from the CPU, and helps with that heat/performance equilibrium :)

    For that reason alone, its worth enabling just to help with the heat balancing.

    If you watch the side beside here, you can see sometimes the CPU usage is all but equal, but other times its noticeably lower with HAGS on.
    Hardware Accelerated GPU Scheduling ON vs OFF Windows 10 2004 - RTX 2070 Super 10 Games 1080p 1440p - YouTube

    Hardware Accelerated GPU Scheduling OFF vs ON - RTX 3060 Ti + Ryzen 5 5600X 10 Games 1080p 1440p 4k - YouTube

    This all being said though, I found a new worst case scenario for heat generation for the machine, and it made me change my settings somewhat.

    FFXV Benchmark.

    Not only does this benchmark (and I assume the game) push very high GPU load, it also pushes (70-80% at some points!) high CPU load...literally loading all 8 cores and 16 threads, and pushing wattages higher than any I'd seen before in a 3d load or any othe game, 50-60W+ at points. Combined with the 190W VBIOS, this was too much heat load for the heatsink array to handle, even on MX4 and with the CPU undervolted, so I've tweaked the ratios down again (I'd rather prevent 100 C in a worst case real world load as then everything else will run better/cooler)

    This load still pushes the CPU into the mid-high 90s at points, but following the tweak it at least IS getting VERY hot, but not enough to throttle heavily below my settings.

    My settings are now
    1 cores - 45 (-3)
    2 core - 45 (-3)
    3 core - 44 (-3)
    4 core - 44 (-2)
    5 core - 44 (-2)
    6 core - 43 (-1)
    7 core - 40 (-3)
    8 core - 37 (-4)

    If I stepped back to the 180W VBIOS or back to the stock 150W VBIOS, I'd probably be able to push this back up again. I am going to give the 180W VBIOS a shot again with the same settings and seeing how much difference that 10W load on the GPU actually makes in terms of performance.
     
    Last edited: Mar 13, 2021
  5. Alexrose1uk

    Alexrose1uk Music, Media, Game

    Reputations:
    616
    Messages:
    2,324
    Likes Received:
    11
    Trophy Points:
    56
    Ok, well I stepped back to the 180W vbios, and the difference in temperatures was small but noticeable (about 4 degrees on both CPU and GPU).
    With the same CPU settings above, this pushed my temperatures noticeably back towards the mid 80s and much more comfortably so...BUT...it also meant I could tweak the CPU settings a bit more :) GPU clock speeds did drop a bit, but still noticeably higher than stock.

    This allowed me to change my CPU settings to:
    1 cores - 45 (-3)
    2 core - 45 (-3)
    3 core - 44 (-3)
    4 core - 44 (-2)
    5 core - 44 (-2)
    6 core - 44 (0)
    7 core - 41 (-2)
    8 core - 38 (-3)

    At least in the FFXV benchmark this actually allowed me to net slightly better performance vs the previous higher GPU wattage, lower CPU performance (13444 vs 13706). Temperatures still extremely hot, I saw 98 degrees peak and Throttlestop registered 100, but crucially I believe this was for a split second, and I did not see any noticeable clock speed drops compared to my earlier more aggressive turbo settings, CPU was running 3.7-3.8 GHz (as expected given the core/thread load) throughout the entire test; in tandem, the higher benchmark score with this 180W VBIOS and higher CPU performance suggests that this balance was at least better for this benchmark which weights heavily on both, so ultimately is more balanced overall :)

    I will give this a run around the benchmark circuit, see how I do on others that are less CPU focused, and decide whether I need to go back to my prior settings which were more conservative on 6-8 cores, but run cooler, or whether I will go with this :)

    ****

    Based on what I am seeing, I think the 2019 Omen 17 CPU cooling was obviously designed around the 6 core variants/45W sustained loads, with slightly lower peak turbo frequencies, and lower core counts, and NOT the 60W+ requirements of the I9 under full load; especially when combined with the 150W+ 2080 TGP.
    It desperately wants JUST A BIT more surface area, non-shared heatpipes, or an extra heatpipe on the CPU to move that heat away from the small die area that bit quicker, and it sounds like they MIGHT have resolved that on the 2020 Omens onwards.

    All being said, much as I'm having to tweak this machine for better all around performance/better balance, I have to say I think given the money I paid for this machine, I genuinely couldn't have gotten anything better for anywhere near this amount of money. Now that I've applied a bit of love, tinkering and the 180/190W VBIOS, I am genuinely seeing desktop 2080 region performance from the thing, and better than 2070 Super desktop cards in many tests (albeit maybe not a super clocked one).

    I will say whilst I've gained performance from the 180W/190W VBIOS, its definitely NOT equivalent to the additional 20-30% extra power I am feeding then GPU over default VBIOS, so I suspect HP actually made a rather good choice there at 150W, it does seem like it was probably the optimal balance between performance and heat (and still WAY higher than any MaxQ nonsense), especially in trying to balance the i9 CPUs. It is possible maybe my reasonably high VRAM OC (to 15.6Gbps, desktop 2080 Super levels) is leeching a bit of power from the GPU core but at least in my testing the extra RAM performance is better than leaving the VRAM at stock. Maybe my GPU clocks aren't as high as they could be if I optimised that, but the extra 30W the 180W VBIOS gives should definitely make up for that, with my GPU clocks now not dropping below 1800MHz, and regularly at the high end, or in the 1900s or even 2000s.

    HP are very close to making some truly exceptional machines. I also wish they'd come set to performance mode vs standard out of the box or offer more user configurable options (such as TGP/boost/RAM speed), as I've noted some reviewers miss this performance toggle and then lampoon them for more restricted CPU performance compared to those with unlocked wattage (which is then fixed by setting the machine back to performance mode) and means they probably lose sales they rightly should have won.

    It feels like with a little extra focus, and not even much extra cost, HP will be amongst the best options in this field. They are very close to providing something that rivals any of the other gaming laptop manufacturers release in UK/Europe, at a noticeably lower cost; if I was Dell (Alienware) or MSI, I would be scared, as HP only needs to resolve a few minor issues (and improve the speaker configuration) to be on top of them, at a much lower cost.
     
    Last edited: Mar 13, 2021
  6. Alexrose1uk

    Alexrose1uk Music, Media, Game

    Reputations:
    616
    Messages:
    2,324
    Likes Received:
    11
    Trophy Points:
    56
    For what its worth, in the FFXV benchmark, at 1080p I get around 13500-13750 or so on standard, and up to around 10850 on high. By contrast every video of the 2070 Super DESKTOP card I am seeing on the net is getting around 9000, so this is a CLEAR 10-15% faster than a desktop 2070 Super, and definitely in desktop 2080 territory.

    I also just watched a video of an R7 3700X and RTX 2080 Gigabyte Turbo OC desktop card (basically a slightly OC stock model) getting 10555, a very equivalent system with a slightly OC desktop 2080 and 8c/16th CPU, almost IDENTICAL to my score; so we are DEFINATELY in RAW high end desktop territory, even if our CPU and GPU have now been succeeded in the desktop space. With the much more limited offerings in the laptop space, we should be good for another year or two, even on high settings.

    There are other videos with the likes of one of the Aorus Xtreme super high end 2080s getting about another 5-10% (11800), but bare in mind those were around £1000 at the time we bought this laptop...

    FFXV benchmarks original stats available put a 2080 stock at high as around 9750 points, a Titan V at 10664 (so about the speed of my laptop...yay) and a stock 2080Ti as 11135.

    We are getting THAT close. Ultimately if I was able to run this CPU at full speed, and get away with the 200W VBIOS, at the same time (for example maybe in a AC 21 degrees controller room over the 25 approx with hotspots of this room), this really would be all over an equivalent desktop.

    It also really highlights to me that given how cutdown the mobile 3080 is (its basically a slightly cut 3070 desktop due to Ampere's power requirements) that there really isn't a great upgrade for this laptop available yet...STILL, a year and a half almost later.

    AMD is pulling further and further ahead in the CPU market, so the 5900HX for example looks like a good step forward there from the 9880H, both in power consumption, more performance and also probably a little cooler, but GPU-wise, we've maybe only got ~10% more performance and more VRAM this gen on the 3080, which is quite disappointing given its essentially two revisions forward if you count the 2080 Super. The desktop 3070 is basically a 2080TI in performance, and the mobile 3080 isn't even quite to that level, so if I'm only about 10-15% off that 2080ti, then that comparison carries pretty neatly.

    We're are essentially waiting for Nvidia 4000 mobile series or for AMD to bring high end RDNA2/3 to mobile for anything really worthwhile to upgrade to; right now there is legitimately NOTHING worth throwing 2K+ at, there IS no real upgrade path, as we've barely moved forward in two Nvidia mobile GPU generations.
     
    Last edited: Mar 13, 2021
  7. Alexrose1uk

    Alexrose1uk Music, Media, Game

    Reputations:
    616
    Messages:
    2,324
    Likes Received:
    11
    Trophy Points:
    56
    Well just for the sheer lols, I decided to set fans to max, remove all the CPU turbo bin reductions, and put the 200W VBIOS on (with my pre-existing UVOC curve, +800 VRAM, and the -.120 I have on the CPU).
    CPU DEFINATELY throttled on FFXV bench! Daaaang. Shame the heat output is too great to run this setup normally hah
    BUT got an end score of 11402! That is HUGE for a laptop of this era, and not that far from that Aorus Xtreme desktop GPU OR a stock 2080Ti. Bearing in mind how much power and cooling those have, these laptop 2080 Max P chips are definitely decent bins!

    Also ran the 3Dmark suite and got:
    Firestrike - 22907! - GPU score - 28595, CPU score - 20875, combined -9766
    https://www.3dmark.com/fs/25136128
    UPDATE: IF THIS WAS USING A VERIFIED DRIVER, I WOULD HAVE BEEN THE WORLD RECORD HOLDER FOR A 9880H/2080 Mobile GPU. I BEAT THE VERIFIED NUMBER 1 by ~100 points.
    Search (3dmark.com)

    Timespy - 10940. Gpu score - 11322, cpu score 9186
    Night Raid - 48078, GPU score 101480, CPU score 12074


    Goes without saying this is pretty insane levels of performance from a laptop, especially one that although big is by far NOT the biggest or chunkiest of the gaming laptop breed. I am genuinely wondering now if its worth upgrading the MX4 in the laptop to something else slightly better (that isn't LM).

    The benchmarker in me is genuinely thrilled as that Firestrike is possibly a record for this CPU/GPU combo as Firestrike's result screen told me the best score for my config was 22800~.
     
    Last edited: Mar 13, 2021
  8. AlleyKat

    AlleyKat Newbie

    Reputations:
    0
    Messages:
    8
    Likes Received:
    2
    Trophy Points:
    6
    Hey all! New member of the Omen club here :D Picked up a factory refurb 15" 2020 model (ek-0013) from Microcenter a few days ago; I was originally out to score a deal on an open-box Zephyrus G14 or M15 but Best Buy sux and I couldn't find one in-stock for a good price. I'm glad I ended up with this machine instead because I'm super enthused about it so far. Specs-
    10750H
    16GB
    RTX 2060
    512GB+32GB
    1080P 300hz

    I knew that being a refurb there would be some mechanical issues, and there were, but I mostly got them fixed. Trackpad was loose and rattley but I shimmed the retainer on the front (sadly there's no adjustment set screw like on Macbooks) and it's absolutely perfect now. The CPU side fan has an intermittent rattle at low RPM which I'm not sure what to think about... dunno whether it's just rattling against other components and needs to be dampened, or if the bearings are bad and it needs to be replaced. Leaning towards the former since it sounds different (more of a rattle than a grind) than when I've had confirmed fan failures on other machines and it goes away at other speeds. It's pretty annoying but fixable. The storage was also questionable... I hadn't used an Optane setup before, let alone the H10 c-c-c-c-ombo drive and was pretty disappointed. Some things maybe seemed snappier than a regular NVMe but overall there was a lot of random lags and stutters that I couldn't explain- sometimes it actually felt like using a spinning HDD where the system would zone out while loading a program or opening an Explorer window. And on top of that the Intel RST driver was using like 2-10% CPU constantly which was enough to do weird things to CPU Turbo. So I said heck that and put a fresh install of Windows on a proper NVMe in the second M.2 slot (an Inland 512GB with DRAM & E12 controller that I pulled from my old laptop). Performance is better, idle CPU use is lower. I separated out the Optane and NAND on the H10 in BIOS and I'm using the NAND portion as a secondary storage disk for things that don't mind being on a slower SSD. The Optane I guess I'll use for a scratch disk or something lol, idk.

    The biggest issue though was terrible throttling on the CPU. Initially I couldn't go above ~3.2Ghz single-core or ~2.8Ghz all-core because temps on individual cores would usually jump right up to 97* PROCHOT and even if I could keep them in the mid-90s with max fan and an (unstable) -100mV undervolt the PL1 would creep down to around 27W and limit clocks. This was especially baffling because the GPU was running perfectly, like around 60* even in 100W "Performance" mode. I know that this is basically the worst CPU to have to cool combined with the easiest GPU to cool but something seemed off with just how poorly the CPU was running so I decided to jump right to repasting. I wish I'd taken a photo of the next part because uh wow. I pulled off the heatsink to find... someone else's questionable Liquid Metal job! I was completely baffled for a moment while my brain processed what I was seeing. To this mystery previous owner's credit, they did take all the precautions to discourage metal creep with tape "moats" around the CPU & GPU. But on the other hand, this is what did them in because they made a terrible mistake. Again, wish I'd taken photos but I was just so gobsmacked all I could think about was cleaning up and avoiding getting droplets of gallium or whatever under a BGA. So picture an Intel mobile CPU. There's that metal frame around the edge, right? That spreads the mechanical load so the heatsink isn't only pressing on the die? The well-intentioned doink who did the liquid metal covered that frame with with electrical tape to keep the LM in. o_O The end result being that the f**king CPU wasn't even making direct contact with the the heatsink. The only way it was getting any cooling at all was because the LM was sort of filling in the electrical tape-width gap. I cleaned and repasted both CPU and GPU and got much, much better thermal results. The GPU got a few degrees warmer which I knew would happen going from LM to MX-4 but it's still staying real cool. The CPU difference though is night-and-day and I'm no longer cursing Intel's name. With a bit o' Throttlestop magic I can get the 10750 to stay at its max 4.3Ghz all-core indefinitely, pulling about 70W and staying below 90*C which is going to be great for content creation stuff where the GPU isn't heavily loaded and I can put all my power & thermal budget into the CPU. At these settings, CPUz bench got 532 ST / 3802 MT... like holy crap wow. That's DOUBLE what I get on the OC'd FX-8320 in my desktop! I'm still figuring out tuning for gaming but early results are promising; I think I'll be able to get 3.6-4Ghz with the GPU loaded but I have noticed that if I go too hard on the CPU & GPU at the same time the system kicks the GPU from 100W down to 80W which I can't have (gotta go fast).

    Speaking of the GPU, it's amazing. I have less to say about it other than it's running real nice. In 100W mode stock clocks were around 1650-1700Mhz and with an OC in Afterburner I'm getting ~1790-1845Mhz which seems phenomenal to me for a laptop. Memory seems ok OC'd to 13GT/s, might push it further might not. Haven't done much actual benchmarking yet, just stress testing and playing Cyberpunk (which counts as stress testing, I promise!) but I have a feeling I'm gonna get real close to the overclocked Vega 64 in my desktop :eek: I just really, really, really wish there was a way to control the GPU TBP manually with Afterburner or something. It would be nice to have, for example, a 100W profile for xtreme gaming, a 60W profile for less intense games, and like a 30W profile for things that need more GPU than the Intel 630 but not the full powah of a TU106. Nice chip though and runs cool (~70*ish) thanks to the suuuppper low thermal density one gets when running a 445mm^2 die at 100W.

    So yep. Pleased with my new machine despite the tweaking needed out-of-box (or is it because the out-of-box tweaking has yielded such good results? :p) I'll post some benchmarks when I have the CPU & GPU perf really dialed in.

    Oh, and the display is fantastic! Nice contrast and color reproduction and I'm finding that 300hz is almost as good as having Adaptive Sync for ability to run at low framerates (30-50fps) with Vsync but without the extreme lag that happens when running Vsync @ low FPS on a 60hz panel. Extremely good times.

    Team Omen w00t
     
    Last edited: Mar 16, 2021
  9. Burrick

    Burrick Notebook Geek

    Reputations:
    11
    Messages:
    96
    Likes Received:
    8
    Trophy Points:
    16
    I came to the conclusion that the H10 has no business in a laptop. The optane module itself gets much hotter than the QLC half, and it just doesn't run right without a heatsink. I got a M01 i3-9100 for $120 refurbished to drive my TV, obviously not getting a video card for it anytime soon, but even in that well ventilated case the H10 removed from my Omen didn't fare well until I got a beefy heatsink on it. It still spikes up to 60C during backups and such. Such a weird thing they pushed out the door with no use-case engineering. The Omen 17 has the most awful heatsink bracket I've ever seen, since the modules don't lie flat on the H10 I don't think it ever made good contact.
     
  10. AlleyKat

    AlleyKat Newbie

    Reputations:
    0
    Messages:
    8
    Likes Received:
    2
    Trophy Points:
    6
    I noticed the heat too... also got over 60* temps on the Optane running CrystalDiskMark. I read about the heatsink issues and I think mine are making good contact cause the NAND portion of the H10 and the SSD I added are staying reasonably cool. All those IOPS on the Optane just drive the controller real hard I guess
     
Loading...

Share This Page