This is the Navigation Sidebar
Links:
This is the Navigation Sidebar
Links:
Sidenote: This is a really rough first draft, I'll likely add some more thoughts as time goes on. I'm mostly transcribing thoughts but without the structure that guided them on the forum. I'll ultimately add some of that, sections, clean things up.
As happens from time to time, some discussion has occurred on The Forum. I won't link to it here, but largely, the context is as follows:
A forum member is interested in buying a new computer to run some piece of software that only runs on Windows. It's been a few years since they've had to buy a new computer, where should they look? In addition, the software recommends some moderately beefy computing power.
I chimed in, as I usually do, with information about The Long Plateau. Basically, computers that will do fine at that software have been around for a long time. You should check with the vendor or the user community to see what they recommend to decide where your money is best spent.
My general recommendation was to look at Dell OptiPlexes. I cited a couple arguably very old models that you can get for close to $0 now that will meet those specs, a new high end model that allows high end gamer graphics cards, and gave example of a couple off lease models in the middle, which you can buy directly from Dell.
Nearly immediately, recommendations for Ryzen came in.
To be clear here, I don't think Ryzen is bad, nor if it meets your use cases and the systems you can buy or build with it meets your needs, should it be avoided. My recommendations were based on systems you could buy off the shelf that came configured to meet the needs of the software.
This is a scenario where for most use cases, it's really a wash. The Long Plateau has meant that for most people's needs, average computers have been “fine” for the better part of a decade. Most people consider the start of this to be with Sandy Bridge, but even a high end Core2 computer from 2007 can run Windows 10 and browse the web fine. I'm writing the first draft of this page on such a machine.
The areas where Ryzen really shines is when you need a lot of threads, which most consumer workloads just don't and where your workload isn't sensitive to having the highest possible single thread performance, where Intel is still leading.
So, largely, if you're an enthusiast and need a computer for content creation and your workload is forgiving of giving up some IPC in favor of more threads (think: 3d rendering, video editing, compiling software) then Ryzen is a great deal. The real clincher here is absolutely that high core count Ryzen chips are available for less than Intel's highest core count chips.
OP ultimately shared their software title and it's not entirely clear what the optimal choice is. The software is new and it's the kind of software where its user community is already largely using fairly powerful computers.
I got called an Intel shill, which, I guess.
I've got a lot of reasons for it. Most of which revolve around being able to just buy a computer (either used or new) and have something reliable. There aren't very many Ryzen based business desktops, and the one Dell introduced (OptiPlex 5055) just a few weeks ago (As of late december 2019) is only available in extremely low end machines. HP's EliteDesk 705 is a bit better off, but when you go to this kind of platform, you're exchanging most of the cost benefit for a pre-built machine that comes with support from the vendor.
So, if you want to build your own machine and support your own computer, then you can get some price benefit if you need a 6+ core machine, all other parts being equal.
The gotcha here is that that's not entirely true at the low end. You can get an H310 or a B450 motherboard for around $50-70, and a Ryzen 3 1200 for around, or a Ryzen 3 2200G for around $100 or an i3-8100 for around $130. The barest bones B450 motherboard costs around $10 more for a B450 than for H310. So, for an IGP system, Ryzen will save you about $20 of the entire cost of your build.
Ryzen, along with the other Zen platforms - Threadripper and Epyc, have some other benefits, but getting them largely requires you accept self-built machines or make other product and form factor compromises. For example, as of this writing, Dell/HP/Lenovo aren't building Ryzen or Epyc based servers. There's one Ryzen Pro based workstation, the EliteDesk 705 G4 Workstation Edition, which is basically a normal office desktop platform equipped with ECC and a Radeon Pro or Quadro graphics card. It's good that exists but it's not a competitor to, say, a Precision 5820. (Especially since AMD's Ryzen platforms aren't being qualified for over 64 gigs of RAM.)
Same deal in servers: You can get 1p and 2p Epyc servers, or you can build your own, but you can't buy a competitor to a low end 1u or any tower server with Zen in it. The closest thing is the HPE Microserver Gen10, which has an old Excavator core in it, ironically, exclusively because performance doesn't actually matter to people who buy and use that particular product. (Except for the ones who do care about performance, who buy older Intel-based Gen8 MicroServers instead, because the older Intel version outperforms the Excavator version very roundly.)
And, actually, a Ryzen based server is something I'd love. A hypothetical X570 or similar platform with official support for 128 gigs of RAM, 4-16 core CPUs, and the i/o benefits of being able to split the PCIe 4.0 link into more PCIe 3.0 links, or to get some ultra high performance PCIe 4 SSD storage on top of a normal disk based array would make a great hypothetical PowerEdge T345.
The other server problem is that, and I'll admit I don't know how big a deal this is today, in some kinds of virtualization clusters, you might not be able to add AMD-based hardware to existing Intel clusters, both in terms of live migrating VMs between CPU vendors and in terms of managing them as a single unit. This scenario is made worse by the fact that AMD had well liked high-core scalable platforms years ago and then did utterly nothing to expand those platforms or prevent Intel from overtaking them entirely.
The server scenario where AMD makes most sense is individual servers that are set up for task specific needs and don't need to be cattle as part of virtualization infrastructure. For example, there are rackmount platforms meant for high NVMe storage density, and for virtualization, AMD has done a lot to build high-core-count single socket palatforms that rival what Intel does in two sockets. They're still costly, but they're cost competitive with Intel.
Back to mainstream PCs, the other aspect to all of this is that there's an entire literal decade of back catalog of suitably usable Intel-based hardware, from the Sandy Bridge machines up on forward. AMD hardware pre-Ryzen, which launched in December 2016 (and as such is mostly likely still on first owners) I'd consider to be outright unsuitable. There's both some personal experience here, I had an AMD machine (which, to its credit, was $350 including a monitor and a copy of Works) that would drop USB off the chipset if there was sufficient activity on the machine, requiring a shutdown and reboot.
Not all machines will do that, of course, and the tasks that did that were above and beyond high end workloads that arguably needed a Real Server, but it's something I wouldn't expect out of the equivalent Intel platform. (In particular, just for full disclosure, I was running an SBS 2011 Standard server for some testing, but with only 4 gigs of RAM. SBS 2011 Standard spins up Microsoft Exchange and SharePoint 2010 as well as Server 2008R2 with a bunch of roles, so it was swapping near constantly. The other work I had the machine do was as a management console and backup target for VMware management, which it did fine at the time.