It's because it's based on the Bulldozer family of CPUs. AMD stirred up controversy in claiming that Bulldozer processors were made up of "modules" and that each module had two "CPU core." The problem is they counted the "CPU core" as just the integer execution unit. There are two of them, but there's still one floating-point unit and one instruction decoder/dispatcher:
Obviously this is a bad choice for cpu.
It's because it's based on the Bulldozer family of CPUs. AMD stirred up controversy in claiming that Bulldozer processors were made up of "modules" and that each module had two "CPU core." The problem is they counted the "CPU core" as just the integer execution unit. There are two of them, but there's still one floating-point unit and one instruction decoder/dispatcher:
That's the essence of the whole controversy. Some people say that because the modules share resources they're not really a 4 core CPU but a 2 core CPU. That seems specious to me since all multi-core CPU's share resources to some degree...if nothing else some caches are shared for instance. That's why the Windows' scheduler was changed to be aware of CPU architecture so not to move processes to cores with unshared resources and introduce unnecessary lag.
The problem with going down this route is this allows SMT to be considered multi-core as well, even though at the minimum the only thing that's unique are registers needed to contain state information. I don't think any computer scientist is going to seriously consider a processor with SMT as a "multi-core" one.
I'm willing to bet Windows's scheduler was already built for this... for the Server editions. x86 based multi-processor systems were a thing since the mid 90s at least.
That's pretty much it. Or more like there wasn't really a definition with enough legal weight that marketing could do anything they want.
I can't believe they would either. Aside from the decode/dispatch portion I recall a lot of the argument for declaring a Bulldozer module one core was due to the floating point unit being shared. Well, by that logic 8080's all the way up to 80386's should no longer be considered even a one-core CPU because they had no FP unit; they had to be added as an "X87" math co-processor accessory.
The way I see it, a CPU core is a processing unit you can isolate by itself and it would be able to function in the same system, with the caveat that anything in the uncore (like memory controller, PCIe controller, etc) is still there. That is, I should be able to take what you've identified as the "cores", remove everything but one, connect it to the communication backbone, and it'll still work.
Facebook
Twitter
Instagram