Hacker News Reader

Comment by sub7 | Hacker News Reader

The 3700X is very slightly lower on performance per $ but it does have 8 cores and 16 threads so if you're going to be compiling code it's totally worth it.

The 3900x has been selling for $400 which is also amazing value.

Intel CPU market is behind after many years. Long live AMD design + Taiwan manufacturing.

View on HN

> Intel CPU market is behind after many years. Long live AMD design + Taiwan manufacturing.

Except Intel still delivers better and more consistent gaming performance largely based on better overall architecture (L3, IMC and what connects these).

Both Intel and AMD are billion-dollar publicly traded companies. I don't get why people would fanboy either of them. I don't want AMD to win, because then they are going to abuse their position, just like Intel did. Conversely, I don't want Intel to win. Two players in a market is already problematic; trying for one player is just suicidal.

I don't think anyone actually wants AMD to "win", regardless of how they express themselves. What they want is AMD to have a lead on Intel, since it's been so long since that's been the case, and we'll all benefit from Intel not being continuously in a hefty lead, like it has been for decades.

I'm still thinking about my cheap Ryzen-based PC only as a stop-gap solution, until the "real deal", aka Intel 7nm CPUs, premieres in 2021 (fingers crossed) .

Why is the "real deal" not the 7nm+ AMD CPUs coming out this year, or the 5nm AMD CPUs likely coming out next year?

Because AMD is a company with a long history of making crappy processors, that only recently created a decent one, by lucky coincidence. Intel is exactly the opposite.

Even now Intel architecture is still superior in some areas, the biggest downside of it being the fact that they are stuck on the outdated 14nm++++ process.

Once the Intel's 7nm comes out there will be no reason to buy AMD, except for the price (I expect Intel's 7nm process to be roughly equivalent, or even superior to TSCM's 5nm)

>that only recently created a decent one, by lucky coincidence.

They have been repeating that "lucky coincidence" for the past 3 generations of Ryzen (ofc you can nitpick that one of them wasn't a true gen, but more of a gen 1.5, but whatever).

3 extremely successful generations in a row seems a bit more than just a "lucky coincidence" to me.

>Because AMD is a company with a long history of making crappy processors

Tell that to Microsoft and Sony.

Like it or not the Xbox and PlayStation shows what the future will bring for games on PC which is exactly the opposite of what Intel tries to sell.

An alternative point of view is that AMD took the long view and invested heavily in future gains at the expense of short-term gains, allowing the market to be even less competitive for Intel during that time.

Perhaps their current situation isn't as much a "lucky coincidence" as the payoff for a very successful long-term strategy?

You must not remember the Athlon 64 era.

I don't think most people want AMD to win completely, as no one expects Intel to ever fold. Also historically, Intel has been much scummier, and it's paid off for them. When AMD was previously top dog (most recently with Athlon64), I don't remember them being as anticompetitive as Intel has been.

Intel has 5x the market cap and 10x the revenue compared to AMD, and AMD is the underdog in its dGPU competition with nVidia too! If you want innovation, you'd want to equalize investment in R&D for both these two competing pairs, it makes sense to be pro-AMD.

AMD is so far behind intel in market share, money, resources overall that worrying about "you don't want AMD to win (either)" is looking too far ahead.

Ended up with a 3700x. They seem to run at the very edge of their thermals straight out of the box though.

I tried a light touch of overclocking and got absolutely nowhere. (Stock cooler)

Running that close to the edge is an achievement in itself though I guess

I see this as a good thing, the chip is tuned well enough to hit its limits out of the box, without having to mess around with manual overclocking to get optimal performance.

Yeah, and this is advertised in the 700X moniker. If you want an overclockable chip you need something from the *600 series.

This might be slightly wrong: I think the X is the key indicator, not the 600/700 distinction.

I can confirm this. I'm running a 3700x with the stock cooler and It's idling at 50-60 degree at 3.6 Ghz. If I enable Game boost mode in the motherboard (it basically push the frequency to 4.2 and turn up all fans) it increase the performance of the CPU in cinebench from 4662 to 5087 points. That's said, the CPU reaches a whopping 100 degree!!! Probably need to get better cooler before enabling that again.

>so if you're going to be compiling code it's totally worth it.

This always gets brought up as a use case, but how many people are building non-trivial codebases where compile times matter? If build times are under 5s, near perfect parallelization only nets you a 1.25s gain between 3700x and 3600.

Most web developers. Webpack + typescript compiling and bundling all the assets.

You want to see a preview as quickly as possible. Best if you can keep a live view on a second monitor.

I doubt much of this is parallelised. Would love to be wrong though!

Cannot speak for everyone, but the builds of my team are heavily parallelized, and we aren't even running a monorepo. For any non-trivial project that has its own somewhat complex build pipeline with many internal dependencies and such, it is pretty much a given.

The only scenario I can foresee where people wouldn't be doing parallel builds is if they are just working on their personal side project with no other people involved, where the need for parallel builds is non-existent either way. You don't want to have one giant monolith project for your whole complex web app, it becomes a nightmare very quickly.

is any of that compute-bound? I would imagine that it's mostly I/O-bound.

My CPU definitely spikes when compiling JS.

Any decently sized C++ or Swift code base is going to have compile times way over 5s. Many just moderately sized projects have build times over 10 minutes.

This is for a full rebuild, right? If you're changing a few files it shouldn't have to recompile + relink everything.

For moderate to large projects, link time can be a huge factor even for incremental builds. It might be that faster IO is the better investment, though.

My current codebase is 28 hours for a full build on one machine...

I dread having to update the compiler or some core "used nearly everywhere" header file, and wait a whole another 28 hours...

Reply on news.ycombinator.com