win32asmguy
-
Posts
582 -
Joined
-
Last visited
-
Days Won
3
Content Type
Profiles
Forums
Events
Everything posted by win32asmguy
-
Yeah I think its a bios bug. To clarify further if I set Core offset it does show the value as applied (and persists across warm reboots) but when running a benchmark there is no change in observed voltage, load core ratios, or score. If I set even -1mV to P Cache, it crashes, reboots and now shows lower voltage under load, higher ratios and higher score. Oh, and just to rule out software I tried uninstalling XTU, and also tried applying the undervolt via the QuickCPU utility where it also crashed only applying P Cache offset.
-
Glad to hear Prema is aware of this model. So I actually do not have the virtualization stuff enabled already in Windows 10 21H2. I have determined the crash is related to adjust CPU P Cache only - so adjusting CPU Core or CPU E Cache does not trigger the crash. Oddly enough, if you don't adjust P Cache to any offset value then any Core undervolting applied does not take effect either. Maybe a weird bug with Throttlestop + H670? I am curious what @electrosoft's good bin 12900k scores at 125W PL1/PL2, just as a baseline against what mine is doing.
-
Just the tools linked in the guide. You can see the setup menu dump in the earlier post: Keep in mind that the offsets may be different if you have a different bios version on your machine from PCS, best to just dump it from your own machine to ensure things will work. The one from Eurocom was already very recent, built 4/16/2022 which was just a few days before mine shipped. It also has some interesting stuff in it, like options to configure four e-core clusters instead of just two, which may mean Raptor Lake support is be baked into the feature set with a microcode update. It would be funny if we could put one of those in to this machine on release day.
-
It will persist through warm boots, but is lost if you cold boot (maybe standby/resume as well). So far I have not found a bios option that can store a known stable undervolt (as you might do using a real desktop H670 board). Here is the EFI commands that must be run to allow undervolting via Throttlestop: # based on Eurocom Raptor X15 BIOS 1.07.05RPCS1 / EC 1.07.04 dump setup_var CpuSetup 0x43 0x0 # disable Config Lock setup_var CpuSetup 0x10e 0x0 # disable Overclocking Lock setup_var CpuSetup 0x1d9 0x1 # enable Overclocking Feature Here is a good blog post explaining the whole process, but you can start at Step 5 to make a bootable USB to modify the variables. Note that if you ever get a black screen from modifying things you can always remove the main and backup battery, and hold the power button down for 30 seconds to reset to factory defaults. https://brendangreenley.com/undervolting-2020-dell-laptops-like-the-vostro-7500-and-more-tips-to-improve-thermals-battery-life-and-speed/#cpu-undervolt
-
It does appear to be a bug of some sort. I eventually realized that even when it crashes after applying an undervolt, the value is still set after reboot and the system is stable. So 23,200 is possible in that scenario with CBR23 with a -90mV on core and P-cache with P-cores around 4ghz and 125W draw and max fans active with the system on a laptop stand, and the 3070Ti active at idle clocks just to drive my external display. CCC is not very useful. It can set keyboard backlight colors, switch performance profiles, or adjust the fan curve. I think the better, aftermarket heatsink could be a real possibility. If Clevo had added an extra mDP wired to the iGPU I would be much happier with forced hybrid graphics. I want all connected displays driven by the same GPU to simply the software configuration on Windows and Linux. I doubt Clevo will make a 17 inch variant of this model. It would likely be an X170 refresh with its own different limitations.
-
Any offset causes causes an instant crash. I tried -5mV, also even +5mV causes a crash as well. It only crashes if e-cores are enabled. If I could have Clevo change anything on this chassis it would be to make a variant with no Nvidia, make the USB-C support Thunderbolt 4, and put an excellent 6-pipe to dual fan cooling solution so going to 200W could be done with a Z690 chipset and full overclocking support and a binned chip, and a larger battery. I know most people want a GPU, but I still wish there was a CPU-focused DTR like this.
-
Would using VF Offset Points help? I tried searching around for 12th gen undervolting guides and noticed a few mentioned that instead of using a static offset. The crash occurred at idle after applying the static offset with e-cores enabled.
-
Eurocom told me that there is only one heatsink for all models. They also told me that the MUX may have been disabled after the manual was created. I am still hoping the chip is present on the system board and maybe a future bios update will fix it. They gave me a Control Center version 3.70 on a thumb drive with the rest of the Windows 10 drivers so I will try it. So with the e-cores enabled any undervolt seems to be causing an immediate crash. Does Core/Pcache/Ecache need to be adjusted in unison? Or is there some other trick I am missing. I know you mentioned Load Line Calibration before as well, I have not looked into that.
-
- XMP: The bios has hidden options allowing a custom profile or XMP profiles to be selected. I did try setting a custom profile but I got the ratio/speed incorrect so it could not boot. XMP should be more forgiving. - Gpu overclocking: It has the GPU Performance Scaling option in the bios so maybe. - Clevo Control Center: I have not installed it - XTU: Yes it works but P-core undervolting was grayed out. TS was perfectly happy to apply an undervolt with the same bios configuration so maybe XTU sees the "H670" chipset and disables stuff in its UI - Battery Life: I did not specifically lock things down to lowest multipliers, but on a clean boot with the 3070Ti not in use I saw about 18W idle power draw. So maybe 2-3 hours. I think a good bin K series, delidded, is a must if you want the most out of what this chassis can give. I have not tried with the e-cores enabled yet.
-
Ok I purchased a 12900k and just swapped it out. Undervolting is now possible via Throttlestop. Out of the box this 12900k scores 17000 instead of 16000 (just p-cores active again). Close to 18000 with a -75mv undervolt, 4.4 to 4.2 on all cores still 120W draw. Now I have to decide if I am going to try and have a awkward conversation with Eurocom to see if they will exchange for a different CPU as the one I purchased cannot undervolt.
-
I am starting to think the non k 12900 cannot undervolt. I tried configuring an all core and per-core undervolt via the hidden bios options and nothing would take effect. I have not seen total system power to exceed 235W. The GPU just stays at 115W and the CPU at 120W. I have not really seen the GPU go up to 140W either. RJTech will only offer it if Sager does as well. Eluktronics might pick it up as well.
-
I have seen as high as 4.1 at the start of a multicore test. Max power consumption observed is 130W, and it always has EDP OTHER in limit reasons. THERMAL appears as soon as it hits 100C. It can sustain 5.1 single just fine. Still no luck on getting undervolting configurable in Throttlestop even with OcLock/CfgLock disabled and "OverClocking Feature" enabled. Maybe I can try manually specify a voltage offset via those options. No idea if the EC allows it to draw more than the rated 11.8A at 19.5v for the 230W brick. But I would certainly consider a 280W brick if it does not heat up as much as the 230W under load. I have not tested the GPU much so no idea if the EC imposes a lower PL1 if the GPU is under load.
-
Note that it is the single pin barrel connector, not the 4-pin that was used on models like the P775TM and Tornado F5. So, when I look at HWInfo64 it says the 12900 has a max 7-8c ratio limit of 47x, so there is certainly room for more. I want to get the overclocking lock disabled so I can try undervolting as well. Temps were actually higher with the e-cores disabled. 3.8ghz was with the e-cores disabled.
-
Actually I was just going to reply, yes I did rerun CBR23, multi score was 16100 after disabling the E-cores, at the same 125W draw. It appears their is some limit imposed at 125W as I have yet to see power draw beyond that even with using Throttlestop, and the Clevo default to 150W PL2. Throttlestop reports it as "EDP OTHER" in the limit reasons panel. The score with E-cores disabled seems low to me, I would expect it to surpass the 11900K much more than that. Although it was only at 125W / 3.8ghz all core clock speeds, so maybe it is accurate. Oh and the power plug is a barrel connector not the the rectangular plug. It is used on quite a few clevo laptops.
-
Yes there is but it is hidden by Clevo. I modified the hidden option and now the cores are not visible in the OS. I tried modifying the hidden MUX option as well but it resulted in a black screen. File_DXE_driver_DriverSampleDxe_SetupUtility IFR.txt
-
I did not run the single core test. I noticed rockitcool offers a delid service including their copper IHS for $70, does that seem decent? Is there some way I can tell if my 12900 is a good bin other than installing it in a desktop board with no power or thermal constraints?
-
Here is with throttlestop with PL1 set to 120W. It is higher than I have seen with any 12900hk atleast.
-
Performance mode has PL1 set to 70W and PL2 set to 150W. CBR23 17800 with the stock paste job at those limits. I am using some generic DDR4-3200 CL22 memory for now, any suggestions for something better (has to be 32GB kit). Mine is the 140W 3070Ti. Fan noise is better than the X170KM. The weight is fine, great for traveling and lap use despite having a modular CPU. The keyboard does not flex, but mine did arrive with what I think is the UK English layout despite ordering with the US English option.
-
Yeah the user manual had photos of the bios showing the mux option. Unless this motherboard variant does not have the mux chip. As far as the heatsink goes it is strange because there is clearly a channel for a heatpipe going to the other fan. Maybe the 3060 model ships with a 3 CPU heatpipe solution? Oh and yes it has a 2.5 inch drive bay.
-
The heatsink is different than the photos on Eurocom's site. It does have 4 screws for the CPU, but only two heatpipes going to just one fan. For Entertainment mode, PL1 is 65W, PL2 is 125W. HWInfo64 says 240A for its IccMax. The bios has no MUX option and the display is connected to the iGPU. It will not be great if it cannot be switched to the Nvidia GPU.
-
I wish they made a similar but less featured product that just combines USB-A and DP 1.4 and optionally power delivery into a single USB-C for use with gaming monitors.
-
Technically Alder Lake S Desktop chips have 8 lanes of Gen4 equivalent bandwidth for the DMI link, 15.6GBps total. That should carry over to the mobile WM690 chipset as well. It should have a single x4 M.2 Gen4 slot connected to the CPU, just like the 7560/7760. It was actually a very nice upgrade for the storage situation with Tiger Lake in general, had there not been the buggy disconnect switch issue.
- 972 replies
-
- 2
-
-
- dell precision 7770
- dell precision 7670
- (and 6 more)
-
@Dell-Mano_GAny chance we might see a 32-64GB CAMM XMP module as the tech matures? Is the palm rest and keyboard temperature still relatively uniform with the new cooling design? Is the AX211 Wifi a modular card or soldered to the system board? Will discrete graphics direct output mode for the TB4 ports still be an option? Any chance these models can support traditional S3 Standby in addition to Modern Standby? Thanks!
- 972 replies
-
- 1
-
-
- dell precision 7770
- dell precision 7670
- (and 6 more)
-
Nice! Mine with FHD did ship late last week so I should get it this week. Maybe if I am lucky the 12900 will be a decent bin. If not then I will just use it at lower power limits until I can either bin a 12900k myself or purchase one from someone moving to a 12900ks or raptor lake.
-
Yeah, I kind of gave up on the X170KM. How is the fan noise on the X170SM with the binned CPU? Does it jump up under short bursts like loading a web page? Ideally for me the fans could be running at a constant low RPM, and smoothly adjust upward under sustained load.
