I am toying with the idea of getting a laptop that would be good for scientific computation, with a very good CPU with a high core/thread count. I would want to install Debian Linux. I have no interest in playing games on it. So, ideal components might be something like an AMD Ryzen 9 9955HX CPU, 96 GB of DDR5-5600 RAM, and a $20 GPU because I don’t care about gaming. It seems there isn’t much out there that meets these criteria. Most laptops with this type of CPU are “gaming laptops,” and installing Debian is apparently hit-and-miss due to the various non-free driver requirements for the graphics components that I don’t need in the first place.
Are there laptops available with just a powerful CPU, a lot of RAM, and not much else? Or is it simply foolish to attempt long-running, high-thread computations on a laptop since it will just throttle anyway, and this work should simply be done on a server?
Your requirements are a bit puzzling to me. You’re “toying” with the idea… of getting a scientific compute workstation? So is it meant to be a toy, or are you planning to use it for serious work?
If you’re serious, then I suggest double checking that MT CPU perf is actually what you need, and not GPU compute. In the past 10 years or so, a lot of scientific compute applications moved to GPUs…
And if this is for real work and it turns out your workload is actually GPU compute optimized, then you should probably get something with a NVIDIA GPU, even if that won’t earn you any street cred with the open source enthusiasts. Likewise, idk about running Debian on new hardware if you’re serious about whatever it is you plan to use it for.
You can get something like a Framework Laptop 16 and configure it with a GPU, or not. Though that’s only getting you a 12 core CPU.
I would build a desktop workstation if that’s an option, but idk what your use case actually is.
Thank you for your suggestions. “Toying” was probably the wrong word to use in my original post; it would be more precise to say I’m beginning to investigate the possibility of buying a laptop of the sort I described. And yes, the computations I would be running are entirely CPU-based.
If you have that kind of money why not go Threadripper or EPYC? Actually 32 core used EPYC CPUs are like $100-150 if you can somehow find a reasonably priced motherboard, but then you’re still screwed on RAM.
Or Strix Halo just in case you ever find GPU acceleration with effectively unlimited VRAM in a 96 or 128GB model useful? The 395+ has the same core count, possibly with a bit more performance than the 9955HX if the iGPU isn’t being used. Probably the better option long term, but those are like $3000.
The best 9955HX PC is the Miniforum A2 at the moment IMO. Even leaves you a PCI-e Slot just in case if you can live with that.
You’re just going to have to accept non-free drivers though.
The 9955HX already has integrated graphics though.
For Linux laptops check out System 76, Lenovo and already mentioned Framework. I have personally daily driven Linux on Lenovo laptops for years. Before IBM sold off the division, suse Linux support was mandatory for most of the product lines. Actual firmware controls most of the function buttons on Lenovo machines making them much easier to live with Linux as the only OS.
Ryzen 395 is the way to go in my opinion also: it is a beast while still being able to manage power consumption and heat output reasonably well. It has the same amount of L3 cache as 9955HX (64MB). It has a powerful iGPU capable of running local AI and other massive-parallelism workloads with reasonable performance in case this becomes handy.
9xxxHX CPUs are repurposed variants of desktop CPUs, drain batteries within minutes and produce tremendous amounts of heat that no laptop fans will be able to dissipate, so they throttle down a lot. The only one of them that is useful for anything is 9955HX3D for portable gaming workstations (I wouldn’t call anything that uses it a laptop as they really cannot be used without AC power source for more than few minutes).
With an Nvidia GPU every Linux distro is hit-or-miss as they all use the same drivers provided by Nvidia that are the root cause of any such problems.
Any other GPU (ie Intel or AMD) work perfectly fine on any distro as well. Debian is not special here either in good nor bad way.
Would you elaborate? Debian currently ships newest stable kernels within just a few days since they appear on kernel.org on Sid and within ~week on stable-backports (currently it’s on 7.0.9). Except of Nvidia drivers integration and gaming optimizations (2 things the OP explicitly stated they don’t care about), it is among the top versatile distros that exist and for this purpose, most server solutions are based on this. Desktop-wise, it provides the same common DEs as any other modern distro: Gnome, KDE, XFCE, MATE…
I mean, you kind of addressed it. I just don’t see the point of running Debian unstable. I would (and do) run stable Debian on a server or VM where the hardware support requirements are well known and stable, and where software not seeing big changes is a desirable feature. Where I want optimal performance on cutting edge hardware, I would use a distro that’s meant for it, like Fedora or whatever.
Like, why over-complicate your life doing something weird? If it’s a hobby then sure I guess, whatever makes one happy.
Perhaps you should re-read my post I would never ever advise anyone to use Sid (Debian unstable) as it’s not even a release: rather a loose bunch of packages waiting to be promoted to testing.
Debian (stable release) definitely is designed to also support the most recent hardware if you need it: as I’ve already stated, all you need to do is to enable stable-backports and use a kernel from there (about 4 mouse clicks via GUI).
I’ll say it again: for anything other than gaming, you cannot go wrong with Debian (Nvidia drivers are not integrated well OOB, but it requires downloading and clicking just one file with an apt repo config from Nvidia to have the latest&greatest driver directly from them).
Idk, maybe you’re right. My concerns would be more things like, how reliable is stable-backports in terms of CI, relative to stable? What are the differences in testing processes between the two? What about maintainer and other actively engaged volunteer testing capacity? How big are the user bases for each? And so on…
The site also mentions this:
Backports cannot be tested as extensively as Debian stable, and backports are provided on an as-is basis, with risk of incompatibilities with other components in Debian stable. Use with care!
It is therefore recommended to only select single backported packages that fit your needs, and not use all available backports.
So you’d use the kernel from backports and hope everything else is compatible?
I just don’t see why I would want to bother with this, but admittedly I’m not experienced in this.
I’m no expert, but my take is, once you start back-porting, you create your own distro. Debian is very well curated and all parts are selected to work together. And all updates also assume this stays this way (updates are security and bug fix only). So, Debian Stable, stays stable (=doesn’t change).
But once you bring in new things, the above isn’t guaranteed anymore. Maybe most the time it doesn’t matter. But if it breaks anything, you have to fix it yourself.
As for Sid and Testing, those are not official releases. They are what the name implies and for developers. People use them (and distros are based on them), but there is no guarantee of anythign being fixed in time. Testing also has a freeze a few months before the new stable release. If you have a bug, you won’t get a fix for a long time.
Anyone is free to do as they wish. but i realize I’m not qualified and just rely on the distro maintainers to do a good job with an official release version.
If this is an option it will probably be more cost efficient and upgradable.
You might even be able to get a humble workstation and add a cheap Laptop on the side for day to day work while the workstation is churning.
Which software are you going to use? Oftentimes Intel can still be a better choice for scientific software. Also for many scientific workloads memory bandwidth is the bigger constraint, in which case strix halo (laptop or desktop) could be an excellent choice thanks to its memory bandwidth.
Laptops are not great at running high power situations at longer intervals. Most are capped at power levels comparable to mid-range desktop CPU’s and their cooling is usually woefully inadequate for dealing with such workloads. So I’d just build a server and remote into it if that is at all feasible.
I’ve enjoyed the Dell 7750 and 7780. They’ve been replaced with the Dell Pro Max. You can order without a GPU. They have both Intel and AMD CPUs (more cores available with Intel though), have official Ubuntu support, and can be had without the GPU.
Do note a lot of Dell laptops are coming with CAMM for memory, which is only upgradable by replacing the whole card, so double check when ordering.
Is this a laptop you’re gonna need for university? And what kind of budget are you willing to spend?
Have you considered offloading your computing needs to a server and using a more portable laptop as your daily driver? You’d ssh into your computing workhorse (also I think you’re probably going to need an Nvidia GPU for a lot of scientific computing applications, which you could always buy later and add to the server more flexibly) with Tailscale, wherever you are, and you’d relieve yourself from the burden of transporting a heavy and bulky workstation
I would never ever advise anyone to use Sid (Debian