As we’ve shown in the review, this means that we get some CPUs. The Ryzen 3 3300X and Ryzen 3 3100 are odd elements to the Ryzen family, especially the 3100 with its awkward CCX and core configuration, but both parts offer a lot of performance for their pricing. At $120 and $99 respectively, using AMD’s latest Zen 2 microarchitecture and the power efficient 7nm TSMC process, AMD is defining a new base line in budget performance.
AMD now leads in budget, mid-range, high-end, crazy server processors, and game consoles.
The lead, yes, for how much longer? I wait for more innovations, like improving ISA like they did with AMD64, like neural processing units as we have already integrated DSP, and stuff.
Thom Holwerda,
I don’t like having to keep correcting you on this Thom, but you keep suggesting AMD leads across the board when it isn’t totally true. While AMD is doing great, they continue to lag intel on single threaded performance. For tasks that are bottle-necked by sequential processing, which is often the case for desktop applications and games intel still leads!
Benchmark after benchmark shows AMD Ryzen with great multicore performance but loosing to intel on single core performance. For example:
https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+9+3900X&id=3493
https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+9+3950X&id=3598
https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i9-9900KS+%40+4.00GHz&id=3593
This isn’t an anomaly, it actually translates to at least marginally higher FPS for intel in a lot of games and desktop applications that to this day still don’t take advantage of high core counts.
https://www.anandtech.com/show/14605/the-and-ryzen-3700x-3900x-review-raising-the-bar/14
You can (and should) blame developers for this, but sometimes devs have their reasons, like avoiding notoriously complex multithreading issues and the fact that a lot of game & graphics software targets the GPU for parallelism rather than the CPU. So while there’s obviously a lot of untapped potential for games running on ryzen, it still leaves intel in the lead for single threaded workloads regardless.
Granted intel’s lead in sequential process performance may well be outweighed by Ryzen’s advantages everywhere else, but it’s a lead nevertheless and to suggest otherwise is putting favoritism over facts. If I were to buy a new computer today, it would probably be a ryzen because they’re really pushing ahead in so many categories, but it is not a win across the board as you imply.
Actually I am surprised that AMD hasn’t been able to beat intel’s single threaded performance. One would think that AMD’s 7nm fab process would enable it to bump up the speed over intel’s 14nm. Clearly AMD is able to fit a lot more cores in the same area, but why haven’t they been able to turn up the core speed to beat intel there too? I’m just speculating here, but I guess there’s some very difficult barriers at 7nm to overcome for higher core speeds that neither intel nor AMD have been able to solve yet.
In any case, I agree with Kochise that the next technological revolution isn’t going to be about traditional core speeds or counts it’s going to require an architectural shift. Things like GPUs and FPGAs are so fast not because they’re clocked fast (they’re relatively slow), but because they have so much more explicit parallelism than a sequential processor. CPUs are rather quaint in this regard, yes even the 64core/128 thread processors don’t hold a candle to the processing power of GPUs. I think it’s inevitable that CPU must eventually embrace more efficient & explicit parallelism far below the SMP abstraction. The main obstacle here isn’t technological so much as software developers being very resistant to change. The last time intel invested in VLIW architecture (itanium) was an abysmal flop because the market judged it by it’s ability to run traditional software. Nobody rewrote their code to take advantage of VLIW. Virtually no commercial software would be optimized and it failed miserably because of this. Nvidia kind of had an advantage with GPGPU since there was already demand for it’s GPUs and many programmers were already programming them. They didn’t face resistance when they added compute capabilities. I’m not sure how AMD and Intel can address this problem for CPUs, but I think one of them eventually has to so that CPUs can continue to evolve rather than get stuck in a rut.
Alfman,
Whoa there Alfman, coming off as a bit snippy there mate. I try to be friendlier than this but It would seem I wasn’t in the best mood when I wrote that. And then wordpress moderated it forcing you to approve it, haha.
Thom, sorry about being unfriendly without cause, I do appreciate what you do for us 🙂
Again this thing about single-threaded performance!
Alfman,
AMD only loses by negligable amounts to Intel in single-threaded loads, save for some exceptions. (These exceptions, e.g. Matlab until recently, exist because Intel can gain an unfair advantage by not playing fair in allowing non-Intel CPUs to use their most efficient code path via the standardised x86 instruction set and extensions, or exist because the software hasn’t really been optimised for use with AMD because there was no reason for developers to do that until March 2017 when Zen 1 was launched. A lot of reasons that admittedly don’t make a difference in the end, but can be explained as something AMD isn’t really responsible for.)
However, when AMD wins it’s by a far greater difference than when it loses. This is where the difference in favour of AMD is made. This is what everyone is talking about.
My last point is that near-as-makes-no-difference comparable performance from AMD costs significantly less than what Intel charges for this top single-threaded performance. It also consumes a lot less power, requires less expensive cooling, etc. There is no reason to shell out hundreds extra of your local money for a few extra percentages of single-threaded performance at many extra costs, except for when you are SO time constrained every single second is worth that kind of money – which I doubt is a lot of people.
Gargyle,
Absolutely, it deserves to be mentioned because many desktop applications in particular are bottleneck by it and it is reflected in the empirical data!
The 13-14% advantage for intel suggested by the data may or may not be negligible to users, as usual it depends.
We can agree that turning off optimizations on AMD CPUs was an egregious & deceptive business practice by intel, really despicable stuff… However the last I’ve really heard of this was a decade ago when intel lost a major lawsuit to AMD.
https://www.agner.org/optimize/blog/read.php?i=49#49
Is this still an issue and if so do you have any data to back up that claim? If so that would be important to talk about. Anyways, the benchmarks I cited above are compiled under visual studio.
I know what everyone’s talking about. AMD’s multicore performance is fantastic, but it’s rather misleading to label AMD as the winner across all categories. It’s like a rigged Olympics where you throw out the scores when declaring a winner – this bugs me and it should bug you too! AMD is clearly the new darling in the media, but the truth is intel still wins for a lot of desktop use cases. Like I said earlier, it isn’t a huge lead. Does AMD have a big multicore lead? Clearly! One just needs to factor in whether the software they’re running can actually take advantage of it in order for it to matter.
There’s a relevant truism here: all software can benefit from increased core performance whereas some software can benefit from having increased cores. People might think I’m expressing bias in favor of single threaded needs, but I’m really more keen on emphasizing both. It’s just that when everybody’s focusing exclusively on multicore benefits I’m forced to pull in the other direction in order to bring some balance to the discussion. If the discussion were more balanced in the first place, I’d be more of a centrist right now 🙂
Sure, I said that much in the past too. It’s one of the reasons I’d consider AMD a better buy right now. However when people speak as though AMD wins in every category, that’s just not true even if they want it to be true.