win32asmguy
-
Posts
564 -
Joined
-
Last visited
-
Days Won
3
Content Type
Profiles
Forums
Events
Everything posted by win32asmguy
-
Yep. Did you mention if there was a specific way to test how leaky a chip is? I think I found that certain desktop boards have dedicated voltage regular sensors that more accurately report how much voltage and current is being supplied. So the more leaky chip would draw higher current at the same voltage and load, correct?
-
Got another chip on the way now. This one is a 12900ks, SP91 overall, P-core SP100, e-core SP75. Already delidded with the copper IHS glued down.
-
For the 7760 I ordered last year, it was confirmed on 7/12, went into production on 8/19, shipped 8/25, picked up by Fedex in Chicago 8/30, delivered in Colorado 9/2.
- 972 replies
-
- 1
-
-
- dell precision 7770
- dell precision 7670
- (and 6 more)
-
Ordered on the morning of 7/7, with the 2 day shipping option. So far it is only in the confirmed status.
- 972 replies
-
- 1
-
-
- dell precision 7770
- dell precision 7670
- (and 6 more)
-
I have 7/26 ESD and 7/29 EDD, for a 7770 with 12950HX/3080Ti/32GB/FHD. It may ship on time as I have already received a notification from Fedex that a shipping label was created.
- 972 replies
-
- 2
-
-
- dell precision 7770
- dell precision 7670
- (and 6 more)
-
I have seen bursts of 136W with the 12900k in the NH55. Not bad given Clevo designed it to target 65W LGA CPU's. The undervolting really helps boost performance and lower noise I am very glad its possible now. The BGA chips are ok but I think performance testing in the reviews are kind of useless and potentially misleading because we don't know if the results are because of good/poor silicon or good/poor thermal contact both of which can highly vary even with the exact same model.
-
Yeah those precision tools tend to be better than the generic stuff at hardware stores. How bad is the IHS warped by the stock mounting frame?
-
It was delidded but the IHS was in the same plastic packaging as the CPU without any tape to hold the packaging shut tight. So it looks like the IHS worked the packaging slightly open then started rubbing against the CPU PCB. The photo of it from the listing looks like those SMD's were intact and his rating was high. So it looks like with the cache the magic number is 40x where I can undervolt and it doesn't crash with 10 mins CBR23. It is still kind of strange though as 47x is stable with a reboot.
-
Yep, got a case opened with eBay. Shame though as I was really looking forward to trying it out. In other news, I finally seem to have an idea why P-cache undervolt was crashing. It appears that it is related to the cache speed which is 47x by default and may be too high at least with E-cores enabled. I tried 43x which did not crash immediately but instead after completing a CBR23 run. I think cache undervolting is very important on this system, because it seems to share that limited power budget between the cores and cache. Less power used by the cache means more that the cores can use, correct?
-
I got the golden 12900k from Fedex today. Sadly it appears to have been damaged in transit. The IHS was loose and broke off five SMD devices on the CPU. I was only able to find four of the broken off devices in the foam packaging.
-
It would always thermal throttle on one run, even with a custom fan profile with both set to maximum speed, and on a cooling pad with low room temperatures. I think it had a poor liquid metal application but I did not want to hassle with it given the issues with booting Linux and the gaudy chassis. It also would worry me because if the expensive BGA CPU+GPU motherboard failed they may try to blame the failure on a repaste during an RMA. So, the best way to purchase a Strix Scar 17 SE would be through a reseller so you can pay them to repaste if needed and they guarantee their modification work in the event of an RMA. Technically the same advice applies to the MSI laptops as well.
-
It would burst up to 175W, but settle around 140W due to thermal throttling. Try running it again but in the default "Smart AI Auto" profile. I think that one allows beyond 115W sustained.
-
How much power does the CPU report being consumed while it runs the test? In HWInfo64 it is listed under "CPU Package Power". In theory it should be able to go up to 180W, but may drop depending on thermal throttling or if the system decides to limit power consumption. For these chips Intel tests them at the factory, and sets a predefined curve of voltages at a given frequency, similar to how Nvidia sets up their GPU's (if you have ever messed with that in Afterburner). So one chip may have lower voltages from the factory depending on its quality, despite both being called "12800HX". So far we have observed this performance delta to be up to 10% with the 12900k.
-
Are you able to comment on power limits allowed by the 12950HX, when the 7770 is in its highest performance mode, and the dedicated graphics is idle? Thanks!
- 972 replies
-
- 1
-
-
- dell precision 7770
- dell precision 7670
- (and 6 more)
-
Yes. Here is some interesting food for thought in the meantime: 12800HX VF Curve(AC/DC LL @ 0.01): 23x 1.087v 28x 1.094v 35x 1.111v 42x 1.171v 48x 1.263v
-
The 12950HX is about 5% faster in 1-2 P-core loads at stock 5.0ghz, 10% faster if you set the favored cores to 52x multiplier but it may require changing EFI setup variables to unlock it. For multicore loads the silicon quality has the most effect and the i9 is generally better. A poor quality 12950HX should always be better than a poor quality 12850HX. With the right tools it may be possible to read out the factory V/F points intel set for a given CPU so we can build up some empirical data. Recent XTU versions seem to have this capability but it wants to put the CPU under load to generate the graph so thermal conditions (and power limits) can skew the result. A4500 will be able to go up to 140W max: https://nvdam.widen.net/s/97whpwwqqb/ampere-mobile-line-card-us-web Dell may also opt to ship it with lower speed GDDR6 so certain memory intensive workloads may benefit from the A5500/3080Ti.
- 972 replies
-
- 2
-
-
- dell precision 7770
- dell precision 7670
- (and 6 more)
-
Yep I hope the numbers from the SP rating and VF curve mean higher performance without an undervolt. I don't have a desktop board yet. I notice the Z690 Strix D4 is on sale for $320 though.
-
Yep. It is already delidded so I can use the copper IHS from the Rockit kit. The final bid was less than I expected. 540usd seems like a great price for what it is.
-
Well it looks my delid is postponed a bit. Ended up picking up a high binned SP104 P-core, SP81 e-core from eBay in hopes it is a bit better than the my 12900k from Newegg and can close the gap with performance I saw on the Strix Scar 17 SE with 12950HX.
-
The improved memory can slightly increase Timespy scores. You could try setting an offset voltage to all cores in Throttlestop and see if it helps. Undervolting P-cache also greatly reduces power consumption and heat generation, although it would cause a crash for me with Throttlestop when applied. You may need to set "Overclocking Lock" to disabled for Throttlestop to work. It is in "Advanced -> Power & Performance -> CPU - Power Management Control -> CPU Lock Configuration". To get the advanced menus it is Left Alt + Right Control + Right Shift + F2 all at the same time. Depending on your Function Lock setting you may also need to press the Fn key so F2 is sent instead of the modifier. The memory scores are not too bad. My Clevo NH55JNNQ gets 48363MB/s and 69.8ns latency with DDR4-3200 CL20 memory. I wonder if DDR5-5200 CL34 could match it in latency, provided the memory and GE77HX could run that stable.
-
Interesting. That kit appears to have better timings than others I have seen. AIDA64 has a memory benchmark if you are looking to compare before/after results.
-
I think the gaming DTR niche effectively collapsed itself when we had the Alienware 51m, MSI GT76, and X170SM all competing against each other for the same customers that year. I would not be surprised if Clevo lost many sales to those competitors, and their R&D budget was reduced as a result. I know monopolies are generally frowned upon but in this case it is probably best to have less competition for this type of laptop.
-
Yeah, no harm in waiting, especially if you don't already have something good to repaste with. Would you swap the K5 Pro with regular thermal pads if you repaste? GentechPC may be able to provide pad height measurements for it as they should have already documented it for their own repaste/repad service. I have heard K5 Pro can be better though because it does not impede heatsink pressure on the CPU/GPU die. Definitely can be messy to work with, but it is not electrically conductive so even if it looks bad it should be ok. For the metal cover on the memory I would use a plastic spudger to pop it off. Look at a video for a Legion 5 Pro 2021 memory swap as it has the same kind of cover.
-
Great information! The K5 Pro they use on the GPU memory and VRM's is fine. The core temp differences are not too bad, 86-96. I am unsure if some of that is just the odd positioned mounting posts, or if they used paste or a phase change pad on this model.
