Dell Precision 7540 and 7740 Owner's Thread

@Aaron44126 thanks for a terrific and insightful reply. People like you are the reason I love this forum site over other sources. Replies embedded below.

Based on your comment, I've done some further research on the GPU. The funny thing is, the result of my research is really, "Argh. Too bad I can't just buy this machine with an RTX 2080 instead of a Quadro." My main DCC apps are Unity and Blender, and neither of those is ever likely to require Quadro as any kind of "certified" platform.

You're absolutely right that there is not much difference in real-world performance between the RTX 4000 and the RTX 5000 in the context of the 110W versions. (Am I correct to infer that this 110W spec means the Precision 7740 is not sticking me with the Max Q version? This is not clearly stated on the Dell web site.)

For my specific situation, the question is whether I'll benefit from the 16 GB VRAM and the extra CUDAs for my lightmap bakes -- and if so, whether it's worth US$1200 to me. On the one hand, the lightmap bakes aren't something I do "constantly", but they are "frequent" and do impact my productivity when they happen. On the other hand, if it's $1200 to go from a 4 hour baking time to a 3.75 hour baking time, is it worth it? The extra VRAM makes a large difference for handling high-resolution textures (I commonly work with 2K and 4K textures, with 8K on the horizon), but the texture size is a runtime issue more than a build-time issue. In other words, if I build an app that only performs well on a Quadro GPU with 16 GB VRAM, most of my clients won't be able to use it. It's an absolute win for me to go with any tech that improves my development productivity, but there's a risk in developing in an environment whose runtime performance is too far ahead of my clients.

TL;DR: In terms of GPU, you've given me a great deal to consider, for which I thank you. I'm seriously considering going with the RTX 4000 and diverting part of that $1200 into getting 64 GB RAM right from the start.

Is there a compatible RTX 2080 GPU card that could plug into this machine? Perhaps I should consider a strategy similar to my SSD plan -- that is, buy the machine from Dell with minimal GPU and plan to second-source an RTX 2080 for it. I'll also ask HID Evolution about that when I discuss the order with them; perhaps they can slot one in for me.

Thanks. I had seen those but wasn't clear on just how conclusive the result was. I'm looking at sourcing this machine through HID Evolution (which is where I bought my current system, and they've provided fantastic pre- and post-sales consultation). So I'll have the option of repasting. I'm a degreed computer engineer, but tbh I really don't trust myself to repaste CPU and GPU. Could I do it? Yes. Do I want to learn by experimenting on my production system? No. Ergo, let the pros do it when I buy the machine.

You're right -- I can get faster RAM if I go non-ECC. I have a professional colleague (who also happens to be a close friend) who is a Dell-certified field service tech, and I asked her about this. She said she hasn't seen any instances of recurring problems in the field that would be of the type ECC would help. Bad RAM sticks do happen, but they go all the way bad, crash the system with or without ECC, and are replaced. My situation is that I do have long-running processes, but they are measured in hours and not days. More importantly, almost every such process is read-only with respect to its source data, and in case of failure the worst case is I delete the output data and some cached intermediate data and start over. Lost time, but not lost information. TL;DR: ECC memory is not really justified for me, so I'll revise my spec.

The Xeon has a slightly higher turbo speed (5.0 vs 4.8 GHz), but I can get an 8 core i9 CPU (I had only noticed the 6 core before). My workload is becoming more oriented toward core count than raw clock, so the i9 is a viable alternative for me which I'll consider. My original reason for choosing the Xeon was that I had seen some online articles suggesting the Xeon was more reliable under continuous high load. If that's not the case, then the i9 is the way to go.

My colleague agrees with this, too. She says Dell's high SSD prices (in her opinion) exist because of corporate buyers who insist on everything coming from just one line item on the PO (i.e., Dell charges more "because they can"). Given the same advice from two knowledgeable sources (you and her), I'm going to say this is a definite way to go for me.

Yes, they do. And AMD was getting smug for a while until someone found almost-equivalent flaws in their CPUs. I have nothing against AMD, but generally speaking the old advice, "Pride goeth before a fall", is quite apropos to security flaws.

 

Unfortunately, Dell has done away with MXM GPUs. These systems have “DGFF” GPUs — Dell graphics form factor. They are swappable, and it has even been demonstrated that cross-generation upgrades are possible (installing newer RTX cards in the prior-gen 7X30 systems works), but your only choices are Quadro cards or the AMD ones; no GeForce cards have been produced. Alienware Area-51m (or whatever that new giant thing is called) also has “DGFF” cards but the form factor is different so they will not fit in here. The only way to run with a GeForce card is to stick it in a Thunderbolt enclosure and run it externally.

Pretty sure that NVIDIA has a deal with OEMs that requires that systems advertised as “workstations” use Quadro instead of GeForce GPUs. Cash grab.

Never had an issue with CPU “reliability” and I’ve never picked up a Xeon. I’ve been through a few Precision’s that have run with a Core i7 at max load for days at a time with no trouble. I wasn’t aware of the difference in turbo speed but you can’t hit the max turbo speed on a multithreaded load and power and thermal limits will be pushing you down anyway...

 

One can always count on @Aaron44126's replies and advice to be reliable, precise, accurate, and generally very helpful.

I'd like to address this bit: I run Arch Linux on the Precision 7530, the smaller and older sibling of the 7740. Nearly everything is working: the only culprits are NVIDIA Optimus and the fingerprint reader, which are two notorious pain points with notebooks and Linux, in general.

However, I get a feeling given your workload, you are going to run things exclusively out of the NVIDIA discrete GPU, so you will not have a problem there. Only two steps needed in this case: blacklist the nouveau module, and install nvidia from your distribution's package manager.

Furthermore, there are signs that things might be moving forwards w.r.t. Linux and NVIDIA GPU compatibility in 2020. NVIDIA has already added PRIME offloading support to its drivers, so most hacks like bbswitch, bumblebee, nvidia-xrun, and so on have become obsolete in the past few months.

Do they offer it? I don't see the Precisions in their list of systems. That said, repasting these notebooks is fairly easy; Dell provides a complete service manual with a parts list down to the last screw, open to the public. The Precisions are by far the most modular and most repairable systems I've seen, outside of the very large and thick notebooks. Given your expertise, you should be able to do it in a matter of hours

Furthermore, buying this machine directly from Dell gives you access to ProSupport and ProSupport Plus, which (in my opinion) beats most services provided by anyone else (though I know HIDEvolution has a rather legendary reputation here, and many people in the 'LGA + MXM notebooks only crowd' swear by it; at one point, I was about to buy a Clevo from them). ProSupport Plus means that for the period of your warranty, if you ever catastrophically damage your machine, and this was accidental, Dell replaces your notebook with a new one, no questions asked, and you get to take advantage of one such incident per year. In my case, I got a new notebook this past August (precisely a year into ownership) without even using the accidental damage coverage, because Dell wanted to capture my original system for some analysis with the battery power draw.

The Precisions were offered with the 5 GHz Core i9-9980HK, as shown in this post. That user, @Hopper82 has also done extensive research with respect to undervolting and extracting maximum performance from the 7740; you might want to trawl through their post history for more information.

The option to select it, however, is now gone, and I get a feeling this is a stocking issue; you might want to call Dell Sales to check, first.

As already mentioned, the Precisions use the DGFF nowadays, and it is impossible to find GeForce cards in the form factor that fits and is electrically compatible with these notebooks. If you are interested, you might consider eGPUs to plug into the Thunderbolt ports. I agree with @Aaron44126: besides the VRAM increase, there is little else that the Quadro RTX 5000 offers over the RTX 4000, given its neutered power limits. One can't change these easily with MSI Afterburner, either, because Quadro cards are firstly as locked down as they get; secondly, these are extremely new, and the VBIOS encryption hasn't been figured out yet, and the TDP Tweaker not appropriately patched for it yet. You are stuck with what you get: 110 W power limit.

 

Hi,

well i have a 7740 with i9-9980HK & RTX 4000 & 4k nontouch screen and will now try to give it back to Dell.
A month ago the WLAN-Card started to fail and was replaced by Dell support, and now the same happened to the replacement unit.
And I think it's because of the poor thermal design of that machine:
- when putting heavy load on the machine, CPU runs constantly around 100° and GPU around 90°
- in (only) that state, my sound card becomes unusable, - through constant high frequency squeals
- the nvme SSD placed more central on the motherboard reaches 65° without beeing used, and that's way too much
(all slots are used, the others are max at 42°)
- the failures of the wlan module appeard allways after 2 - 3 hours heavy load, so i assume that's also connected to the high temperatures.
- i posted it a while ago here: When running with heavy load, even the accu gets drained, so there seems to be a time limit for heavy loads imposed by the accu.
- Looking at the cpu: it throttles a lot, due to power limits, - even with this obviously too small power supply.
...what sense does it make to put an "HK"-Version of that cpu in that machine, when it throttles regulary even without overclocking...

So putting in a RTX 5000 cannot make sense in my eyes, when even a RTX4000 gets too hot in that case.
That cpu is also complete nonsense, for longtime loads, - a smaller CPU, that's capable of running without throttling would be much more (cost) efficient.

The 4k non touch display is a real highlight, thats what i really like on that machine, but the current keyboard is a big step back compared to the keyboard on my old M4800.

It's really a shame, my now 6 years old M4800 runs rock solid and you can put it under load as long as you want. It gets louder than the 7740, but doesn't throttle or melt away anything on the mainboard. Before that, i had two thinkpads in a row, that broke with damaged GPUs because of heat under heavy load, so i thought with the M4800 in mind, that Dell is capable of generating a real mobile workstation, - this 7740 looks, as if i was wrong.

In my eyes it looks as if top equipment doesn't make sense on that machine with it's current thermal design. Better go for some smaller cpu and smaller gpu, that are more likely to run flawless with these fans and heat sinks.

Anyone around that has a suggestion for a current mobile workstation model, that simply can cope with longtime heavy loads on CPU & GPU without damaging itself? - highly appreciated!

By the way: the current linux nvidia drivers seem to run together with the intel gpu drivers. I used the RPMFusion Nvidia driver for fedora and that worked without any extra configuration.

cheers

 

My notes.

Do you have PCIe link state power management set to "maximum power savings" in Windows advanced power settings? That helps alleviate the heat a little bit. Might not do much under heavy load on the GPU.

I'm a bit surprised at the GPU temperature (most people don't complain about NVIDIA GPUs getting too hot in these systems (but I guess I haven't heard much on the RTX 4000/5000). RTX 5000 would probably run cooler since the load would be more spread out on the chip; however, it is not much faster because of the power limit.

Power throttling issues with both the CPU and GPU are well-known and documented. You shouldn't buy a system like this under the impression that it will run at max turbo indefinitely, only low-spec laptop CPUs (in any system) would be able to do that at this point; this used to be possible in the M4800 days but not since Intel has raised the upper turbo speeds well past 4 GHz.

The system is not "damaging itself" running at over 90C; it is safe to run the chips at this level. (The throttling is there specifically to prevent damage.)

 

Well i know the difference between turbo and system defined core frequency,
and since i've had two laptops that really damaged themselves in simply running at heavy load, - and that weren't even 100° they reached -, i don't share your point of view.
One had an ATI- and one a Nvidia-Quadro GPU, and both melted away their connections to the motherboard.
In this case it looks to be the wlan card to be the first component to fail, and i cannot believe, that running that system at around 100° for long time periods, will not damage anything, - see the high temperatures at the nvmeSSD.

But anyhow, just assumptions upon my experiences with two other laptobs/mobile workstations from lenovo, and a failing wlan card.

 

Well, opposite experience here, I've had the Precision M6700, 7510, and 7530 each run at max CPU (high 90's) for days on end doing video or simulation jobs and I've never had any heat-related issues result from that. I don't have any system which can get the GPU that hot, though; the Quadro M5000M is the nicest GPU card that I have and it maxes out in the low 80's. In any case, I've never heard stories of permanent damage from heat in the Precision line specifically. A system that can't tolerate "full load" for an extended period of time simply wasn't designed properly.

 

Did you put your notebooks in an oven?

Solder melts at nearly 200 °C, well in excess of the 105 °C, at which the notebook halts, and instantly cuts power.

 

I'm using 7540 with i9980HK for software development, very CPU intensive. No GPU, running Gentoo Linux.
Thermal throttling is very real and I was easily hitting 100C with stock cooling. It can be fixed by undervoling and repasting, but took me a lot of effort. I can still hit 100C for a short while before fans fully spin up, but it mostly stays in lower 90s under full load.
Power throttling kicks in now. All-core load is limited to 4.2GHz by default, and it consumes over 85W. Going over 85W for 28s. or longer gets power throttled to 75W which means 3.8GHz on all-core load, and that's a substantial drop in performance for me. No way around it as far as I know.

Spectre and friends are still there, but I didn't find any performance changes when disabling mitigations in the kernel, at least for my use cases.

Linux compatibility is OK. Stock Ubuntu works fine but I spent lots of time figuring out which modules are needed. Besides, 4.x kernels don't support Intel Wifi (AX200), and <5.5.0 kernels have buggy i915 driver. My current solution is either using ubuntu-eoan kernel or 5.5.0-rc2 (rc3 has broken alsa). BTW support for WD19DC dock is also somewhat broken, ethernet driver crashes, rest works fine.

HiDPI screen with a low DPI external monitor is a PITA on Linux. Might work in Wayland tho, didn't try it

One outstanding bug is frequency scaling, sometimes clocks get locked at 800MHz or (mostly) 4.2GHz after resume, regardless of the kernel. Looks like multiplier gets stuck, can't find solution.

Hope this answers some of the questions.

I'm curious myself now if going with Lenovo P53 would be a better idea.

 

Hi
Anyone re above please?