Hacker News Reader

Comment by jwr | Hacker News Reader

It is worth mentioning that many developers were left out in the cold even when the cylindrical Mac Pro was originally announced. I care about CPU speeds and storage, not so much about the GPU, so the form factor and dual-GPU architecture of the new Mac Pro was a set of bad tradeoffs. I never wanted to pay for an extra GPU, and the cost of attached Thunderbolt drive arrays is astronomical.

I hope Apple will release a "developer machine": something with tons of CPU power, easily expandable internal storage, and a mid-range single GPU. And for heaven's sake, don't try to make it small, thin, or round. I'd buy it in an instant.

View on HN

Unfortunately, I wouldn't hold your breath. It's sad that they are lagging so far behind on internal updates, but at this point I think they care more about how it looks then performance. I'm just hoping they don't ruin the Macbook Pro next.

They kinda already did that with the retina version. you can't upgrade the SSDs(edit: it was the ram that I was upset about originally because they soldered it on the other point still stands) or add an extra SSD and remove the cd drive like you could with the last gen mbp. I want a powerhouse desktop replacement as a developer that I can bring with me places not a more expensive MB air. The screen is nice but not at the value proposition.

I upgraded the SSD on my Retina MBP. It's easily accessible after unscrewing the back panel

my bad. it was the ram that they soldered on and you can't replace or upgrade unless you purchase it at checkout. you still don't have the option to have two drives though.

Accessible, but proprietary. SSD upgrades depend on third-parties producing the hardware -- that can mean waiting a while, if not forever.

I waited a week or so for my SSD to be delivered from America to the UK and fitted it myself. It was trivial.

You do realize that's not the "wait" in question, right?

It can take months or years for a 3rd-party to reverse engineer the hardware in question and produce a compatible replacement.

Remember, Apple's goal is to externalize as much of the "stuff you'd want to upgrade" from their devices as possible, so that all that's left is a little hub with CPU+RAM that you have no reason not to replace as a unit.

Their viewpoint makes a lot more sense when you look at their Enterprise leasing program, where you effectively are paying for hardware "by subscription" and your laptop just gets replaced every two years like a phone. In such a setup, you don't want to stick extra stuff in the chassis that will go away when you replace the thing. You want all your add-ons to be little boxes with Thunderbolt ports, that you can just plug your replacement laptop into and continue on.

Why do you need to many CPUs? I can only think of compiling but in my experience not many compilers can actually handle multiple cores well and you might wanna consider using other tooling if your compiler needs to recompile that 100k LOC again and again instead of just compiling the changed files again (and their dependencies).

As cool as this machine is, it is definitely targeted at video makers who spend more money on powerful hardware (because they need that) compared to developers.

Xcode scales to the number of CPUs that you have very well. Any kind of c-compiler usually scales the compile step very well to the number of CPUs you have. When I had a 12 core mac pro, my compile speeds were actually 3.4x faster than my 4-core macbook pro.

And when you're an iOS / OSX developer, the only practical kind of machine you can use is apple's hardware. You can't go buy a 4 socket compile server that easily like people on other platforms.

Not really, no. Compiling C or C++ is not very distributable by its nature, and the parallelization you are seeing is scaling out builds by building multiple compilation units in parallel. If you're working in one file in an edit-build-test loop, all of those cores are pretty much unused as long as you're using incremental builds. This applies to a lot of languages, actually, not just C/C++. Even beyond that, you're usually I/O bound when you're building anyway, and I've found disk investment to pay off much better than CPU investment.

But yes, for an all-up project build, it will be slightly to much faster because big projects can horizontally scale with the right tooling (make -j, etc).

Very often you modify some upstream file and a 100 files have to be recompiled, so in an edit-build-test loop, it's very relevant. If you only have to rebuild one file, then everything is already usually very fast. When you don't then that's when the 12 cores becomes very helpful.

And having a PCIe SSD is already built into the mac pro and macbook. RAID 0ing a whole bunch of SSDs didn't increase build speeds for me either when we did a test.

And yes, the parallelization is done as you said. The compilation unit step tends to dwarf the linking step.

edit: since HN prevents a reply to the downstream comment, we did test the ram disk case and turning off swap, and comparing to an HDD. The SSD from HDD was a marked improvement, otherwise everything else wasn't that much better unfortunately. This was a few years ago for a full rebuild on a C++ project, focused on the linking step.

If you want to test the effect of disk speed on build times, you should probably get a control case: one where the disk is taken out of the equation completely. Disable swap, check your git repo out into a tmpfs, and build there. If it isn't much faster, then you can just stop worrying about the impact of disk speed.

I work on a large C++ application. A full rebuild is 35 minutes of our main configuration - that's with 2x 20 core xeons. We also have a "gaming" pc, which is an overclocked 6700k (think it's running at about 4.5GHz) - I believe a full build of the same configuration is about an hour.

Think things like test VMs for development when you're targeting disparate OSes and versions. Instead of needing to deploy things to VMs in sequence, you can deploy to them all at once and still get decent performance.

Depends what sort of developing you do if your working as technical programmer there are plenty of areas where you want as much CPU and memory as possible.

Look at the average node in a big data cluster

Clojure code scales to multiple cores very nicely and I can make very good use of them in my work. That said, I do care about single-core performance too, the number of cores is secondary.

I always suspect Apple built Mac Pro for Pixar and Disney. It would be nice if there is a mid range (higher than Mac mini lower than Mac Pro) Mac that is not iMac and shape in a perfect cube. The outdated spec and high price makes Mac Pro all looks with no substance.

Why would they even use a apple machine specifically for pixar, disney or some other animation house? They have a few cross platform apps they use and the option of creating huge beefy workstations that live at an office desk.

Pixar and Disney mostly use Linux (as do most of the VFX / CG studios).

Artists can use Windows / OS X in some cases, but given the main apps they use (Maya, Mari, Modo, Katana, Nuke) all run on Linux, there's no point really, and using Linux everywhere makes admin much easier and more consistent.

There are some apps that are used which are Windows-only (ZBrush), and so certain artists/modellers who need to use it will use that, but these are generally the exception.

Thanks for the info. Any idea what GUI lib/framework are these Linux apps (Maya, Mari etc) using?

Maya, Mari, Modo, Nuke use C++ Qt, Katana uses PyQt, other apps like Houdini use their own solution.

And before someone says "that means they're not native" - well obviously, and anyway, they have their own theme which is quite dark, as artists don't like bright colours in an interface distracting them from the content or what they're doing.

Ditto. I still use the old Mac Pro as my main workstation, and really wish Apple would just release a new pro tower. For the love of god do not merge it with the iMac.

Not to mention the reliability of Thunderbolt drive connections under OS X. I've got an iMac with a 2-drive Thunderbolt enclosure with a homemade fusion drive. At least 1 time in 10, it disconnects 'unsafely' on sleep, and then I cannot get the drives to show up (even diskutil doesn't show them) despite unplugging and replugging thunderbolt, powering off the drives, etc.

The GPUs are important for accelerating video rendering (which is a big Apple market), see https://www.apple.com/final-cut-pro/all-features/

And today if you work with Deep Learning you'll want a fast GPU

Correction, you'll want a fast 'Nvidia' GPU. AMD is pretty much not a part of accelerated deep learning. So this makes buying a Mac even more problematic.

Yeah - it's not just "a GPU". The inability to get an Nvidia GPU is a problem for me as well, as I use tons of CUDA-specific code.

> don't try to make it small, thin, or round

Maybe they'll go back to the cube. Developers seemed to like that shape.

Hardware designers as well ;)

Pretty much this. As my primary focus is scientific computing tasks, which means CPU and CUDA, I basically walked away from the Mac Pro, after using (and loving) the old metal 2012-era version.

My next workstation? Not from Apple.

why did you walk away from the dual gpu machine if you use cuda. did the dual gpu machine not use nvidia gpus?

The current pros use non-replaceable Radeons. More significantly, they use Radeons that are 3 generations old. In those 3 generations, AMD literally doubled the stream processors, and performance, they can get from a single chip. The GPU's lag is in a lot of ways more significant than the processor, because we've seen such sweeping changes in the past couple years.

They don't use Nvidia GPUs, nor is there an easy way to replace them.

What does a modern developer do now? Bear with this or an MBP? Cross compile?

From what I've seen it's a mixture of Macbook Pro or running a Linux desktop - probably largely depends on what you're actually making.

I use a Mac Mini from the times it was still upgradable. Put in 2 SSDs and maxed the RAM — cheap and convenient.

Develop on a PC?

By a dual xenon pc and run windows or Linux

I am primarily a Mac/iOS developer, but occasionally have to do CPU or GPU intensive stuff. I use a macbook for most of my work but have a windows desktop machine for when I need it. (Ironically, the beefy windows machine was cheaper). For example: lightmap baking.

Switching between the two is annoying.

>Ironically, the beefy windows machine was cheaper

That's logical, since Apple focuses on all other aspects of a machine than raw specs first (construction, size, weight, battery life, trackpad, screen, etc).

Late 2012 i7 Mac Mini with 10gb RAM and manually replaced SSD is a very powerful machine.

I hope Apple will release a better one this year because 2014 model lacks i7 option.

Reply on news.ycombinator.com