Linus Torvalds Switches To AMD Ryzen Threadripper After 15 Years Of Intel Systems

[sandy8925]

You want a picture of Linus giving the finger to AMD?

He should have waited until the Intel Xe dGPU. I look forward to Linus ranting about AMD drivers constantly breaking. Maybe they'll stop doing that? Or maybe he'll stop merging in patches that break things.

[sandy8925]

At least as far as building a Gentoo system goes, I've found that 2gigs per core seems to be the sweet spot.

It depends on what you're doing in the first place. Even for Android development, 8 GB is not enough.

Android Studio + Gradle daemon + emulator + multiple browser tabs and windows exceed 8 GB + ZRAM.

[coder]

Gates most probably never said that famous 640k quote.

It gets taken out of context. At the time, he was talking about a CPU with only 20-bit addressing. And it wasn't even linear addressing -- using more than 64 kB, at a time, was annoying and inneficient.

I think he probably assumed that when CPUs with larger address spaces came along, the OS and its memory map would evolve, correspondingly. Probably his biggest error was really underestimating the inertia of DOS. Not that he could do much about that, even if he'd predicted it.

[coder]

AMD has lost their minds if this leak is true. This means they just copied Intel on every single thing and just compete on price. So now the G parts just have a weak igpu just for the UI and pixel art gaming, just like Intel was doing all along. What is the point of getting an 8core with such a weak igpu? Boring, trash product, only a tiny niche might be interested in it.

Nice troll, there.

You've obviously not followed the CPU market nearly enough to make this kind of commentary. AMD has made CPUs with integrated graphics (which they coined the term "APU" to describe) since 2011.

https://en.wikipedia.org/wiki/List_o...ocessing_units

Sadly, ignorance knows not its own bounds.

[coder]

My understanding was that AM4 motherboards include graphical outputs,

Not all of them do. A few are designed for users planning to use a dGPU.

There are also a few AM4 server boards which include a BMC. So, even the display output they have is not wired up to the pins required to use APU graphics.

[t.s.]

It depends on what you're doing in the first place. Even for Android development, 8 GB is not enough.

Android Studio + Gradle daemon + emulator + multiple browser tabs and windows exceed 8 GB + ZRAM.

Most probably what he is talking about is system that has 8 or more core/thread.

[sandy8925]

Most probably what he is talking about is system that has 8 or more core/thread.

Yeah, but a higher number of cores doesn't magically necessitate a higher amount of RAM. And a lower number of cores doesn't mean a lower amount of RAM will be fine. My laptop just has a dual core hyperthreaded CPU but I sure wish it had more RAM because that's the limiting factor more than the CPU for me.

[coder]

Yeah, but a higher number of cores doesn't magically necessitate a higher amount of RAM.

It does in the following way.

If you get a high-core-count machine to accelerate an existing workload - say, building the linux kernel - then the presumed way you're speeding it up is by running more of the existing compilation jobs at any given point in time. Each of these jobs requires a certain amount of RAM, so that amount will scale linearly with the number of cores.

To give another example, maybe you buy extra cores so that you can spin up more VMs, each of which does a set task - like running a suite of continuous-integration tests.

Of course, there are other, multi-threaded workloads that already had to be loaded entirely in RAM for the duration of the work. In this case, splitting up the work among more threads might not use much more RAM (at minimum, you need a stack for each thread, but that's only like 8 MB, last I checked).

So, as a general rule of thumb, having more cores is going to drive demand for more RAM, but it's definitely workload-specific.

[sandy8925]

It does in the following way.

If you get a high-core-count machine to accelerate an existing workload - say, building the linux kernel - then the presumed way you're speeding it up is by running more of the existing compilation jobs at any given point in time. Each of these jobs requires a certain amount of RAM, so that amount will scale linearly with the number of cores.

To give another example, maybe you buy extra cores so that you can spin up more VMs, each of which does a set task - like running a suite of continuous-integration tests.

Of course, there are other, multi-threaded workloads that already had to be loaded entirely in RAM for the duration of the work. In this case, splitting up the work among more threads might not use much more RAM (at minimum, you need a stack for each thread, but that's only like 8 MB, last I checked).

So, as a general rule of thumb, having more cores is going to drive demand for more RAM, but it's definitely workload-specific.

Right, but you're talking about specific workloads that require more RAM if you choose to run more jobs in parallel. Therefore not a general rule of thumb.

I gave you an example that would benefit from more RAM even with lower core counts, but beyond a certain point, more RAM doesn't help and higher core counts don't necessitate larger amounts of RAM. Therefore I've proved you wrong.

[Tomin]

Right, but you're talking about specific workloads that require more RAM if you choose to run more jobs in parallel. Therefore not a general rule of thumb.

I gave you an example that would benefit from more RAM even with lower core counts, but beyond a certain point, more RAM doesn't help and higher core counts don't necessitate larger amounts of RAM. Therefore I've proved you wrong.

But a comment quoted talked specifically about building Gentoo so this is really beside the point.