AMD Ryzen 5 CPUs Launching Next Month, 4 & 6 Core Options

[L_A_G]

The sad reality is after over a decade of multi core CPU's in the consumer space and decades before that of multi- CPU rigs in the high end space we still can't get a handle on parallel programing. And time and again test after tests show rapid declines in scaling past 4 cores and particularly after 6.

Your best all around bet even to this day is simply buy the clock. In other words....buy the fastest CPU you can afford. And base speed at that and don't rely on any turbo mode. And don't worry about more than 4 cores.

I wouldn't be that negative... There's a lot of applications in common use that still aren't properly multithreaded despite being suitable for it, but it's way better than what it was only a few years back. Let's not forget that proper multithreading generally requires you to re-think not only the implementations, but the underlying algorithms/processes. Some jobs simply have to be done in sequence rather than parallel due to race conditions et cetera.

Admittedly we still haven't gotten to the point at which the field of high performance computing, who had to tackle the issue of parallelism head-on over a decade ago, is today and probably will never get to that point due to many of the algorithms and processes we use simply not being suitable for a high level of parallelism. However just we've seen like in HPC, the march towards greater parallelism is not an overnight thing, but a gradual process taking years to complete. However I will admit that it has been a somewhat slower process as HPC uses a lot of software purpose made/modified by the people using it and slow software wastes very expensive run time on multi-million dollar supercomputers rather than personal machines costing only a few hundred dollars.

[VikingGe]

Let's not forget that proper multithreading generally requires you to re-think not only the implementations, but the underlying algorithms/processes.

From the code I've seen in the past I bet 95% of all thread safety issues stem from programmers throwing mutable state at absolutely everything for no reason. Look at FreeType2, for instance. Nice library for font rasterization indeed, but implemented as a state machine with lots and lots and lots of unnecessary global state.

Granted, FreeType2 is quite old and multi-threading wasn't really a thing when it was first released, so it does have an excuse for its outdated API design. Modern libraries and applications don't have that excuse, yet people love their over-complicated state machines with absolutely zero thread safety, almost no extensibility and millions of illegal state vectors.

[sordna]

The 4 core version would be interesting to test. It is one CCX whereas the 6 core is still two CCXs with one core disabled on each CCX. The 4 core version should suffer from less cache bottlenecks and thread migration issues between CCXs.

Nope, unfortunately according to several articles, such as http://www.tomshardware.com/news/amd...500,33913.html even the 4 core Ryzen 5 has 2 CCX, unfortunately, with 2 cores each. It's also disappointing the clocks on the 4-core are so low, it's as if they want to keep the low core ryzens from performing better than the high core ones in games ! Anyway, for me it seems the 6 or 8 core Ryzens are the ones to get.

[bug77]

From the code I've seen in the past I bet 95% of all thread safety issues stem from programmers throwing mutable state at absolutely everything for no reason. Look at FreeType2, for instance. Nice library for font rasterization indeed, but implemented as a state machine with lots and lots and lots of unnecessary global state.

Granted, FreeType2 is quite old and multi-threading wasn't really a thing when it was first released, so it does have an excuse for its outdated API design. Modern libraries and applications don't have that excuse, yet people love their over-complicated state machines with absolutely zero thread safety, almost no extensibility and millions of illegal state vectors.

That's not the only reason, though. Even properly threaded code is inherently harder to test, that's why migration is slow.
For example, in the traditional, sequential world, you' do:

x = 1 + 2
assert x == 3

In a threaded world, you don't know when x equals three...

[torsionbar28]

The sad reality is after over a decade of multi core CPU's in the consumer space and decades before that of multi- CPU rigs in the high end space we still can't get a handle on parallel programing. And time and again test after tests show rapid declines in scaling past 4 cores and particularly after 6.

Your best all around bet even to this day is simply buy the clock. In other words....buy the fastest CPU you can afford. And base speed at that and don't rely on any turbo mode. And don't worry about more than 4 cores.

Eh, not so much. Kiddies playing video games are about the only market that *hasnt* jumped on the parallel programming trend. Video games is the only area that doesn't scale well past 4 cores.

Transcoding video apps easily load up as many cores as you've got. Pro content creation apps do the same - why do you think pro grade workstations have been two-sockets for a long time now? Plus the exploding IT industry means you've got developers, testers, and sysadmins taking advantage of virtualization, running multiple vm's on their PC's. All these parallel processing tasks would have been unthinkable (or at least very expensive) a decade ago. Heck even the latest cell phones have eight-core chips in them. There are very few scenarios outside of playing games where having few higher clocked cores delivers better performance than more cores at a slightly lower clock.

[torsionbar28]

I wouldn't be that negative... There's a lot of applications in common use that still aren't properly multithreaded despite being suitable for it, but it's way better than what it was only a few years back. Let's not forget that proper multithreading generally requires you to re-think not only the implementations, but the underlying algorithms/processes. Some jobs simply have to be done in sequence rather than parallel due to race conditions et cetera.

Admittedly we still haven't gotten to the point at which the field of high performance computing, who had to tackle the issue of parallelism head-on over a decade ago, is today and probably will never get to that point due to many of the algorithms and processes we use simply not being suitable for a high level of parallelism. However just we've seen like in HPC, the march towards greater parallelism is not an overnight thing, but a gradual process taking years to complete. However I will admit that it has been a somewhat slower process as HPC uses a lot of software purpose made/modified by the people using it and slow software wastes very expensive run time on multi-million dollar supercomputers rather than personal machines costing only a few hundred dollars.

All true, plus the cost of writing the code itself. A few million bux worth of developer time to improve parallelization on an HPC app is not even a question, absolutely you'd do it. For a $19.99 piece of consumer retail software, not so much, the ROI isn't there. The low cost and tight margins of consumer products is probably the single biggest holdup to improved parallelization for desktop peecee apps. That said, the proliferation of 8+ core desktop chips may help force the issue in the near future.

[AndyChow]

Eh, not so much. Kiddies playing video games are about the only market that *hasnt* jumped on the parallel programming trend. Video games is the only area that doesn't scale well past 4 cores.

Transcoding video apps easily load up as many cores as you've got. Pro content creation apps do the same - why do you think pro grade workstations have been two-sockets for a long time now? Plus the exploding IT industry means you've got developers, testers, and sysadmins taking advantage of virtualization, running multiple vm's on their PC's. All these parallel processing tasks would have been unthinkable (or at least very expensive) a decade ago. Heck even the latest cell phones have eight-core chips in them. There are very few scenarios outside of playing games where having few higher clocked cores delivers better performance than more cores at a slightly lower clock.

Exactly. Plus, even playing video games, often you want to record and transcode/stream at the same time. 4 cores won't cut it.

People say 'this app doesn't scale well', but how often are you running a single app? Most computers are running 100+ processes continuously. If you run one (or several) VMs, do you want to close them because you feel like playing a game or need to do some rendering?

Moar cores FTW!

[Luke_Wolf]

Ahem, What was that about games not scaling past 4 cores?

[vein]

Imho, that's the worst option. For $249 you can get a higher clocked Intel part that will be faster in 90% of the things you throw at it (because of better per-core performance). Lower priced parts is where it's at for Ryzen 5.

Lol... You dont even know what i want it for... I want 12 threads for that price and. Clock. And i dont want to bother with overclock. For that it is a bargain.

I might even buy 2 just for the fun of it

[bug77]

Lol... You dont even know what i want it for... I want 12 threads for that price and. Clock. And i dont want to bother with overclock. For that it is a bargain.

I might even buy 2 just for the fun of it

I have a quad core and they're rarely fully used at the same time. For me, 12 would be a waste at any price point. But then again, if the price point is low enough, you can afford to be wasteful. In that case, the only thing to consider would be if, for the same $$$, you could get fewer cores, but clocked higher.

Luke_Wolf If that game needs that many resources to output that sorry gfx, it better had a hell of an AI.