I mean.. At this point, three or four generations on one socket is pretty good going. I quite liked being able to throw the 3600 in without batting an eye (well, aside from removing the monoblock in a hard line build...) and carry on. If the same commitment is made by AMD to AM5, if it's called that in the end, I'd be happy to buy into that ecosystem when the time comes too.
Hmmm... 5900x is on the cards for me with a 3080 as and when I can get one. If the 5900x is going to over big boosts over an 8700k in games I'll just stick with a titanxp for the time being and go about sorting myself half a build in the coming weeks. It also means I can just get a board and CPU reusing my AIO temporarily, not bother with RAM and GPU and sit with that until the 3080 stock resolves itself so I can go full water-cooling. At which point I'll not need to upgrade again for 3 years or so. Works for me nicely actually
We'll believe it when we see independent results over multiple benchmarks - and when they're done on the same day and not 4 months apart. It's also a cop out that the AMD bench has double the RAM, so it's anything but a fair test and that's before we consider the lack frequency information (like whether the Intel chip was overclocked or running at just base - that leak has nothing to clarify what the settings were). Ryzen 4000 APUs exist but are Zen 2, same as Ryzen 3000 CPUs. Ryzen 3000 APUs are Zen+ and a gen behind - which is far more confusing. I assume that the 4000 series name is being skipped for CPUs so that they can sync all Zen3 CPUs and APUs under the Ryzen 5000 brand.
I will be highly surprised if we see a 20% IPC time! Thats a huge reduction for something thats becoming harder and harder to manage. Its like saying they'll be running at nearly 5.7GHz effectively, which seems dubious at best and laughable at worst...
Facts are starting to emerge as to why the huge IPC hike. And it makes perfect logical sense. Latency is down due to one piece of 32mb cache rather than two 16. Instead of a single core having to share cache between two CCX it's now one CCX on one CCD with full access to the full 32mb cache. Latency has been AMD's biggest enemy. It's only a huge problem in gaming of course, as things like Cinebench don't care. https://forum.overclock3d.net/showthread.php?p=1027411#post1027411
That would make sense, but that would also require AMD to do some consistent and logical branding, which is the one area they have always matched Intel.
If you only build a PC for gaming, sure. For anyone else who also does other productivity stuff, then Ryzen is the go to choice.
A 10600k/10900k does not exactly suck at productivity, particularly when all those cores active can often run a lot faster than Ryzen at full load, quite a lot of software doesn't know what to do with 12 threads never mind 20.
I'm still eyeing up a 16C32T upgrade. I do a fair bit of bulk processing, and that parallelises *really* well - like, it's 8.5 times faster running as 16 workers on my 8C16T CPU than running a single worker, even when the processing software has its own native multithreading.
For sure there are cases just making the point that Intel chips don't suck balls at productivity because they have 2 less cores at a similar pricepoint, productivity being a very broad term that sweeps up all sorts of applications, there are still cases where 20 faster threads will trump 24 slower threads, so you need to know your workload.
The age of "moar of the same cores" is coming to an end anyway... https://bit-tech.net/news/tech/cpus/intel-launches-lakefield-hybrid-processors/1/ Yes, that is mobile stuff with Atom stink but it would take a fool to believe that proper CPUs won't come with a pick & mix of different cores in the near future as well.
Intel's always been late to the party... Arm launched big.LITTLE back in 2011. Okay, it wasn't mixing process nodes, but the idea of having some high-performance and some low-power cores was there.
Yep they are certainly late to that game, but also inevitable that they eventually came around to the idea.
That's great for mobile devices, or desktops which are basically just a glorified browser terminal, but mixing cores ends up being a collosal headache for more demanding stuff. Core hopping is still massively expensive, so schedulers will need to become an order of magnitude better at sorting out what goes where, and OSes will need to obey either program demands for running on a specific core and/or obey the user specifying the same. Behaviour of Linux on big.LITTLE arches (like in the ODROID XU4) has not endeared me toward the idea of it for x86. As I said, great for mobile devices where every mW counts, but otherwise?
So today is the day. 12PM Eastern time, or to us in the UK 5PM. Who else is excited and looking to start a new build with Zen 3? Tweet— Twitter API (@user) date Premieres here at 5PM UK:
I'm interested in the performance, doubt I'll do a buy though, I'm full of NVMe and the platform can't handle it, GPU is where I'll be getting my performance gains this year, my CPU typically spends its time waiting for that most of the time anyway.
I guess that will depend on how many PCIe lanes the CPU and new Chipset has, if there is an X670 coming. Current Gen has 20+4 Gen 4.0. But with PCIe 4.0, running a 3080 at x8 lanes will not effect the performance, so therefore, you could run 6x NVMe PCIe 3.0 x4 drives with current x570 and 3xxx series CPU's.
