alittleteapot

I do love me a good spec sheet ^____^ 1. I had to watch techpotato's (Dr. Ian Cutress)'s youtube on DDR5 ECC to really understand what is going on with the memory here. 128GB non-ECC CAMM @3600Mhz or 64GB ECC 2xSODIMM @5200Mhz? Didn't I hear that all DDR5 is ECC? Basically, on-module DDR5 ECC is a marketing gimmick that enables manufacturers to sell more denser RAM chips, but only helps protects from static memory thermal bitflips and does not protect memory integrity when it gets sent to the CPU. It's a bit of an unfortunate tradeoff because I see 128GB as vital nowadays. 2. On the other hand, the 30B Alpaca LLAML model will run on 64GB (it might even run on 32GB but that's unlikely), and it turns my 12-core Ryzen 3900X into a space heater. It'd be really interesting to see how 32 threads of Intel's best mobile CPU would perform on this. If I ask a question on ChatGPT, I can assume it's going to be uploaded to the cloud and staying there... if I ask the equivalent that's running on my desktop, I know my questions will stay between me, my desktop and my creator 😉 But seriously, these home ChatGPT clones are getting very close to the point where they can assist with coding and debugging IP source code that cannot leave the workstation, if they haven't already gotten there already. This field is changing basically on a daily basis. 3. On the subject of NVIDIA RTX™ 5000 Ada Zoomer Generation 16 GB GDDR6, 16GB is apparently enough to run one of those 13B LLAML models real quick-like and stuff, if you're a total nerd and are able to get it to work.

I'm just another old programmer 😄 For some reason, I think there's a lot of us here. I'm happy you found a "Golden Sample" laptop though - you are very, very persistent to go through that process. I'm more restrained in my hardware adventures - I really just have to know enough to avoid deleting everything and keep a few small servers running. However, just because I haven't updated to this generation of Precisions doesn't mean I can't share endless, ill-researched opinions! That's half the fun!

Just to play Devil's Advocate here - those are the same settings that can overvolt the CPU and lead to permanent damage in unskillful hands, correct? It's possible they had some damaged units sent in due to customers bricking their own PCs playing with these things. The worst reason they could have to do this is simply... marketing. If two PCs with the exact same configuration / SKU# / etc. can perform similarly only if PC #2 is undervolted/repasted/reconfigured all to hell, that means the value proposition you get for the price of that SKU# is a complete lottery. It's very likely what we're seeing here with the performance numbers is that all of these laptops are pre-tuned to perform to the lowest common denominator - which is why some are cool cucumbers and some are dumpster fires, as long as they get similar performance numbers out of the box. So, some will have performance on tap, and some laptops will already be pushing the thresholds of their maximum performance.

The variances of this CPU architecture are just crazy. It's likely a well-kept industrial secret just what this variance is; and it's not exactly an easy thing to figure out and isolate, either. Maybe... Cinebench R23 raw numbers, out of the box, no mods or repastes, to establish some sort of baseline - if you have a record of doing that on all five laptops, that could establish some sort of metric for min and max spec of the same CPU doing the same thing. I think in order to get any sort of wafer yields, Intel has to broaden the criteria at which the individual cores are tagged as acceptable, and the lithography processes are getting so fine that minor atomic variances in the materials are greatly affecting the semiconductor efficiency at turning logic into heat. This makes me (and probably many others) really curious to see a Ryzen Pro-based professional mobile workstation, and see how those numbers vary. On the other hand, some secondhand stories and firsthand exposure with Lenovo Ryzen based laptops make it look like there's still a learning curve on executing just the basics with this architecture.

I vaguely remember having similar problems with using the BIOS to make RAID striped arrays from my drives, but I've had no reliability issues with either 2 TBx3 or 4TBx3 striped from Windows Storage Spaces, and that's running lots of Hyper-V VMs on them - I just remember that I had to use a few PowerShell commands to set it up properly. The fact that three of the NVME slots are on the same PCI channel on both the 7760 and 7770 make it feel very natural to put them in a single logical Windows Storage spaces pool.

By sheer coincidence, a technician should be visiting me for the exact same reason - to replace the heat sink assembly. About a week ago, one of the fans started to make a loud, annoying buzzing sound, to the point that it is annoying to have what is essentially my development server turned on. I actually repasted it a few days ago to try and clean out the fans, but that didn't solve the issue (and my repaste job actually resulted in a 200 point loss on Cinebench R23). All the diagnostics passed with flying colors, but playing some video of the buzzing sound and seeing the two CPU fan failures in the BIOS log made it a very quick and easy Tech Support call. Edit: Getting a technician was a bit delayed for the holiday season, but the results are good and my 7760 isn't shaking the desk anymore (likely some minor imbalance in the fan that developed). Dell ProSupport is batting pretty much 100% for me.

I think overall, there's a very strong argument for the Apple MacBook Way of Doing Things and just solder everything in except for the NVME drives. Something's gotta give here - Support, Performance, Upgradability - of these three, which can many businesses do without? It really looks like the engineering solutions to the problems on the table seem to succeed at satisfying pretty much no one except for Joe Average Business Consumer who looks pretty much at the logo and brand and that's it. My proposal - reduce the possible SKUs by dramatically simplifying the high-end Precision lineup. If you think about it, the Dell XPS line caters to every dimension of the consumer crowd, so ignore them - but only the Precision line is for the prosumer / heavy-duty power user crowd. So RAM? 64GB or 128GB. GPU? Integrated GPU, A4500, A5500. CPU? Maximum and a not-so maximum - two tiers. Two choices of display panels, keep the 4 NVME bays exactly as they are (or better yet, allocate a GEN5 PCI lane to each and multiplex it to GEN4 lane, just sayin'). We now have 24 total possible mobo SKUs that need to be manufactured, not counting overpriced NVME drives. Solder that stuff down, shorten the traces, vapor chamber the whole business, and profit.

I am getting a shipment of 3 x 4TB TeamGroup MP34 gumsticks this Monday. Probably the easiest thing to do is just make a 12TB stripe with them using Windows Storage Spaces, so that's my plan. The 970 EVO 2TBs that are in there now have a nice TBW endurance of 1.2 PB, and these drives scale with that linearly to 2.4 PB - which, at $349 apiece from various places, is a bit cheaper than what I paid for those drives in the first place.

I've staying abreast of the Precision 7770 thread - and reading through it, I get the impression that getting the best performance out of a Precision 7770 involves black magic 🙂 (the vapor chamber seems like a very good approach for this generation of CPUs) My HP 14 G2 laptop is still going quite strong, so I have some positive impressions from the line. One question - how is the HP technical support for these laptops, especially compared to the Precision higher tier of support (assuming you've had experiences with both)? My personal experience with Dell technical support for the Precision line has been very positive - I feel like I'm getting real engineers over the phone that get the job done. Does HP have the same level support for their top-tier mobile hardware? When you think about it, the performance and extensibility of these platforms is just a single (very large) slice of the pie, so it would be interesting to hear about anyone's experiences with HP support.

I'm sure most of you are aware of 3-2-1 backups, but just a reminder: 3: One primary backup and 2 copies of data (three total sources), and two the copies have to be on different media, one of them should be in the cloud (encrypted via BackBlaze is very useful, and can be done for one laptop for $75 a year). I actually have a nightly RoboCopy job from a Windows Striped/Mirror shared drive to a USB external HDD, which I then mirror that to the cloud. I'd probably get in trouble for the Backup Police for stretching the definition of "3 Copies" by having the first copy be the "mirror"! @MyPC8MyBrain, that has got to be one of the most original approaches to a rapid restore I've read in a long while. Personally, I treat OS installs as "semi disposable" - the laptop is basically to install Windows and Office, and everything else is in VMs. Base Windows installs inevitably need reinstallation - it's just a matter of time, except for VMs which can last for over a decade with no issue.

Thank you for the correction!

I think this video is really great. Using highly scientific methods, I actually logged into my 7760 and jammed the Windows button repeatedly, and was not able to get the CPU to rise above 55 C, with HWINO64 running. Voltage maxed out at 1.2 V during this operation, but normally is .746 V, but zero throttling, everything is exactly as you'd expect from a base, non-repasted 7760. The Dell Engineering team had successive generations of small, iterative architectural updates and were able to hone the thermal design to a sharp, sharp edge with the 7760. I think no one questions that this previous generation of laptops will last for half a decade at least. But for the Alder Lake architecture, it looks like the Dell Engineering Team tried to swing for the fences and fumbled the ball. If we head on over to https://nanoreview.net/en/laptop-compare/hp-zbook-power-g9-vs-hp-zbook-fury-16-g9?m=-and-c.5 and look at the Cinebench R23 scores for the I9-12950HX on a 16" laptop it shows... 26072? (I wish I had further confirmation of this score, but apparently this laptop started appearing in the wild only recently...) Of course, that laptop has half the available NVME drive bays as the 7770, and is completely chonky, but in this instance, there's no question that HP pulled a real win on this. Oh... I think they use real vapor chambers, too.

I've been following this thread with interest - I actually was inspired by this thread to run Cinebench R23 on my 7760. I'm actually amazed by the sheer amount of energy everyone is dedicating to max these systems out, and it's such a huge difference in heat transfer from the 7760 to 7770. One thing that was pointed out in the Product Q & A / Review page is that the Precision 7770 with 3080 Ti dGPU is that does not have switchable graphics - does that mean it lacks Optimus support? Or, is it just really buggy in an early BIOS version? https://www.dell.com/en-au/work/shop/business-laptops-ultrabooks-and-tablets/new-precision-7770-workstation/spd/precision-17-7770-laptop/on7770wm03au_vp#ratings_section Yes, I have one with the 3080 Ti (and a i9-12950HX) and DO NOT RECOMMEND. There is no option for switchable graphics with this machine. If you turn off the UHD Graphics, the laptop screen will not power. Further, you cannot undo the Bios change without being able to power the laptop screen. Being that everything flows through the UHD card, operation is laden with graphics hangs. When benchmarked with 3Dmark, scores are horrible. The only way to test it is without a secondary monitor, as the graphics hang during testing. My sons Alienware M7R4 scores better with a 3060 and 10th gen i7 (and he loves to rub it in). One, and a funny sidenote in these comments - Dell says putting the laptop on an actual lap could cause permanent damage to you, and putting the laptop on a blanket on your lap will cause permanent damage to the laptop 🙂

My desktop actually runs Manjaro, and it runs a Windows Server VM to be a fileserver and domain server, and also "Work" and "Play" VMs. "Play" is a fully isolated VM, whereas "Work" has lots of premade RDP shortcuts to the many VMs I have running on other servers (and also Fortnite, because, well, Fortnite). The GPU is fully dedicated to one of those two VMs while they run, and completely switches back to Manjaro when they shut down. so I pretty much consider this my "Holy Grail" configuration. It is rock-solid stable doing these things, running 2 of the 3 VMs at a time. The only tradeoffs is that I have three VMs that are .qcow2 files that I need to back up now and then, instead of the usual VHDXs that I use with Hyper-V on my servers. In an ideal world, I could figure out how to do this in Windows Server, thus mitigating the need for Linux at all, but my feeling is that this is an area that Linux has far more advanced technology than Windows does. If I could get all the stuff above to work on my Precision 7760, even with Linux, then that would be really cool. But I feel it would take a lot of trial and error, and the few times I attempted it, I didn't get too far. Thinking about it makes me really want to try again, though...

This is completely bonkers. The only last ditch thing I could think of is to update your Intel ME firmware, in fact any firmware you can find on your laptop, because something in there just isn't working correctly. But I think your plan is really the best. If the warranty support call gets a little wonky..."Your problem only occurs once a month? Did you reinstall windows?" , Email them this whole thread as support, and let them sort through the details 🙂