On the CPU side of things, AMD talked extensively about the forthcoming Zen 2 architecture. The goal of the original Zen architecture was to get AMD, at the very least, competitive with what Intel had to offer. AMD knew that Zen would not take the performance lead from Intel, but the pricing and features of its chips made them nonetheless attractive, especially in workloads that highlighted certain shortcomings of Intel’s parts (fewer memory channels, less I/O bandwidth). Zen 2 promises to be not merely competitive with Intel, but superior to it.
Key to this is TSMC’s 7nm process, which offers twice the transistor density of the 14nm process the original Zen parts used. For the same performance level, power is reduced by about 50 percent, or, conversely, at the same power consumption, performance is increased by about 25 percent. TSMC’s 14nm and 12nm processes both trail behind Intel’s 14nm process in terms of performance per watt, but with 7nm, TSMC will take the lead.
These Zen 2 processors using the 7nm process will hit the market in 2019, so it seems like next year is the perfect moment to make any transitions from Intel to AMD. Intel has been milking its 14nm process for all its worth, because it just can’t seem to get its 10nm process to work properly. With AMD moving to 7nm, it definitely seems the company will actually leapfrog Intel next year.
Nice to see AMD making a comeback.
I’ll be needing to build a new system sooner or later to replace my first gen i7 setup.
With Intel profiteering and ripping folk off like there’s no tomorrow, and with the advances AMD have made and reasonable pricing, I will be looking at building a AMD based system this time around!
What they really need to push at the same time is VEGA. I’ve seen a hack pushing 2.4 times the normal power rating into a VEGA 56 and it was beating a RTX 2070 in some cases. that means if they get the power draw down 50% it could compete.
Their next generation of GPU could get on par with nVidia performance wise, then the war is going to be on technologies (namely realtime ray/path tracing) instead of raw power and that’s a good thing.
Things may change a little when Navi is released, that next generation of cards is aimed more at the gaming market and should close the gap between them and Nvidia some! Vega is been targeted more at the compute and business end by the looks of things.
Navi says “Hey! Listen!”
Too bad most laptops will still be Intel due to shady under the table deals…
Corruption, sorry, “lobbying” in the USA ? Impossibru!!!
Given my own experience, AMD is not yet consistently beating Intel at energy efficiency, although that seems to be at least partly because they still aren’t quite as good at single-core performance.
Is single core performance still a thing nowadays ? I can understand that showing a big score in benchmarks is a marketing plus, but I tend to favor multi-threading capability to handle complex real-life problems. There it’s hard to differ between Intel and AMD because both are rather good.
The software side will also have a big impact on performances.
I also concur with the power usage their should improve (aka lessen).
It really depends on what you are doing. There are still a lot of things out there (especially on Windows) that really don’t parallelize very well (if at all). Games are still particularly bad about this.
Games are getting multi-threaded more on more (business logic + 3d object management + physics + ai + …) because Unity, Unreal engine or else are. Single threaded performance would only impact old games than used to ran on much weaker hardware, hence should benefit the performance increase anyway.
I know I’m kinda picky on this, but even WinZip or WinRar are multi threaded now, so it would make sense only if you are still using a pretty old configuration with no updated software from the 32 bits era. If that’s the case, I repeat that those would benefit the increased performance anyway.
Now if you insist on using an old software stack for strange reasons (ie. running old legacy binaries) I can understand your concerns, but I bet upgrading or finding an up to date alternative that will benefit added cores is the way to go.
Amdahl’s Law says ‘hi’
Single thread performance will always be important in the consumer space, because there are limits to parallelism in most consumer/gaming workloads.
That being said, with Ryzen AMD now has a very strong performance/value proposition. And in some cases they are now reaching better performance per watt than Intel.
Unless there are some very specific cases which require the higher single thread performance or the integrated graphics, I don’t think that there is much merit to Intel’s current 8-core consumer platform vs AMD’s. The price differential between the two vendors is completely out of whack with the performance delta between them.
I fail to see the “proposition” of spending twice as much for 12% performance jump. Alas…
Edited 2018-11-07 20:47 UTC
tylerdurden,
Yep, while I’m pleased to see such powerful hardware. We’re certainly seeing the limits of SMP scalability.
For example, if you go from 16cores to 32cores, you’ll consume much more power and actually see negative performance gains.
https://www.cpubenchmark.net/compare/AMD-Ryzen-Threadripper-2950X-vs…
Intel has the same bottleneck. When you think about why, it makes sense because they’ve got too many cores thrashing around in the same shared memory. Except in localized algorithms that run entirely from cache, there’s just no way that a massively parallel SMP chip can perform well using shared memory.
While NUMA can solve this bottleneck, software needs to be designed to use it effectively. You don’t get more SMP scaling simply by adding tons cores.
I’m looking to buy a new desktop system, and I’m debating whether to go intel or AMD. I’d like to support AMD for a change, and I think the threadrippers have good value compared against intel’s offerings, although they’re somewhat slower for single threaded code (which is often the bottleneck for software), which is why I’m on the fence. Anyone else care to weigh in?
Edited 2018-11-08 21:37 UTC
Yes! This is something I’m always trying to get across in various focus meetings etc.
Yes, you can go parallel but you’ll (probably) multiply your memory requirements, divide your cache efficiency, multiply your power consumption, divide your memory bandwidth and so on.
“There’s no such thing as a free lunch”.
But does that really matter in 2018? Modern OSes are multithreaded so even if your particular task is only single threaded your OS has no issues making use of those extra threads. For example I’ll use myself, even when browsing I have a CPU monitor, AMD Customer Experience (tells AMD if I have a game crash so they can analyze the crash and make patches) Gigabyte Appcenter (gives me control of pretty much everything from CPU clock to RAM to making macros for games and programs), the AV doing checks for updates…no how much any of this affects me? ZERO because my FX-8320e has 8 threads.
Thanks to having that many threads if the OS wants to do something like update the AV while I’m gaming? Don’t care, never notice. If I want to suddenly watch a movie or load a podcast in YouTube while my gameplay footage is rendering? Never think about it, I just do it and everything is smooth as butter.
If you were to try to do any of this with an i3 which at the time would have been the same price as my FX-8? It would have stuttered like a P4 and driven me up the wall! So I’d argue that having great single core really doesn’t matter if your CPU doesn’t have enough threads to handle everything both you and your OS can throw at it. Trust me once you get up to 6 threads and see how you never have to think about what you want to do, you just do it? You really won’t want to go back.
Yeah, except that single core performance still matters if you’re not using more threads than the CPU has physical cores.
Put a bit differently:
A Core-i with hyperthreading disabled can generally beat a Ryzen with an equivalent number of physical cores with SMT disabled in terms of raw performance. This is quite simply because Intel’s physical cores have better performance than AMD’s. Such differences are generally only visible to people when running only one thread, hence the use of the term ‘single-core performance’, and it generally does actually matter because any sane operating system is smart enough to use one thread per-core before it starts using extra threads on any core.
On the other hand, that same Ryzen with SMT enabled will beat that same Core-i with HT enabled if you’re actually using all the threads simultaneously, because AMD’s SMT shares fewer resources between threads than Intel’s HT and does a better job of scheduling those resources between the threads (which means that there’s less time wasted waiting on resources).
The reality is though that most users are not developers, gamers, or scientists, and thus typically won’t be using even half the threads on the CPU most of the time, which means they will generally see better performance on Intel chips.
I don’t think that is necessarily true (about consumers seeing better performance on processors with better single-thread performance). If you cast your mind back to the annual Usenet posts about the Forest curve, you’ll notice that long ago almost any single thread performance was “fast enough” for general purpose computing. Are there still things that consumers do that cause them to gnash their teeth about how slow their computer is? Of course there are, but when you dig into it those are all things that are either bottle necked by network or storage latency (or sometimes bandwidth, and frequently exacerbated by poor coding or inefficient protocols) or they are “throughput” problems that are in fact routinely well parallelized in modern software. Lots of threads require fat pipes to fast memory, of course, so it’s still worthwhile to buy well balanced, non-crippled systems. The new AMD threadrippers and Rome Zen2 parts have oddles of DRAM ports. Yay!
Not only single-thread perf is nowadays enough, computers can be more responsive to the user precisely because they are equipped with multiple CPUs; being able to multitask better benefits also daily OS interaction.
When I built a dual Pentium2 266 machine, it felt at least as smooth as supposedly ~two times faster single processor systems…
Yeah, and maybe vendors artificially limiting the few AMD variants not to come close to Intel performance, e.g. AMD ThinkPad A485: https://www.youtube.com/watch?v=ksc2LSyO8vs
..the transistor density.
AMD uses for Ryzen TSMC, not formerly inhouse Global Foundries? (are they also behind in process node?)
Global Foundries has given up on going past 14nm and is instead gonna try to squeeze as much money out of what they have such as making memory chips and making cheaper CPUs in bulk for like low cost electronics.
https://www.bloomberg.com/news/articles/2018-08-27/globalfoundries-g…
So yeah GF is out of the race, its down to Samsung, TMSC, and Intel and as we know Intel is having serious issues so its looking more and more like AMD made the right call by selling GF when they did, now they can hire companies like TMSC that have the best process and not blow billions making it all in house.
Bring back single core , nobody needs more than one core
Edited 2018-11-12 21:39 UTC