Hmmm, interesting idea haha. This actually is a more reasonable hypothesis than highly binned 8-cores imo, and that is because it makes logical sense. Releasing junk yields early, but still better due to being on 7nm, would allow AMD to 100% convince investors that their current stock valuation is justified. They could simply "glue" (lol) two defective 7nm CCX's together and release a 5+5 4.5GHz Zen 2 2800X. 7nm in 2018 while Intel won't have 10nm till 2020... That would look really bad.
However 10 cores only makes sense imo if Zen 2 has 6-core CCX's (So the 2800X is 2 x partially disabled Zen 2 cores).
Getting 10 cores isn't actually possible with AMD's current architecture - the CCXes in each Zeppelin need to be balanced, i.e. you can disable either 2, 4 or 6 cores per die, but not 1, 3, 5 or 7. AFAIK, connected Zeppelins also need to be balanced, though I'm not 100% sure about that. At the very least, AMD hasn't yet released anything with a lopsided MCM setup. With an MCM 2-die solution, you could then have 4 (3 disabled per CCX), 8 (2 disabled per CCX), 12 (1 disabled per CCX) or 16 cores, but nothing else. From what I've read on the topic, this seems like a fundamental trait of the design, and not something easily overcome, at least on the Zeppelin level.
Regardless of that, there's then the issue of having enough "junk" that still manages to clock high enough to be useful to sell as a new high-end SKU. Given that maximum clock speeds and power draw are the most common points of failure in a new node, this seems unlikely.
I am not so sure about that. It would likely meet the gaming performance of the i7-9700K (Go look at IPC tests, in some games Ryzen is ahead of Intel at the same core count and clocks), and frankly anything to steel Intel's thunder at this point could (literally!) pay dividends. Oh, and it would not use more energy than Intel's newest gen of space heaters. I can assure you those 8-cores from Intel will use 200w+ if you clock all cores to 5GHz, and 5GHz will be required to beat a 4.5GHz+ Ryzen in gaming.
Oh, you're absolutely right about the power draw of Coffee (and in all likelihood Whiskey) Lake; even die-hard Intel fans admit that they need ~1.35-1.45V to reach 5GHz on average, which means anything from ~170W to ~220W. Space heaters indeed. Heck, Intel doesn't even guarantee their own turbo speeds within TDP. My issue, though, is with what's marketed - which has an image-building effect on the public. Intel still sells their space heater-grade hardware at 95W (although it pulls far more on any motherboard that has MCE/MCT enabled by default). For uninformed users, comparing this to a 125W Ryzen would mean that AMD is "less efficient", even if that's entirely BS. It's likely that a highly binned 125W Ryzen would be very competitive against an intel 8c8t (and possibly 8c16t) CPU, but the marketing effect could just as well end up being negative for AMD. AMD has a history of selling clocked-to-the-maximum SKUs that disregard power draw for performance (both in the CPU and GPU space) and releasing a product like this would hew too close to this history.
This, of course, presupposes that it'd be possible to fit two dice in an AM4 package at all, which is ... unlikely, even with the area savings of 7nm. See my previous posts for clarification. The package, substrate and platform isn't designed to accommodate the traces required for this.
I could say: no one - just like currently. ;-)
But here's the gentle variant: the people who these chips are designed for.
You think way too much about core count and performance. 4-core Xeon CPUs sell beautifully despite consumer CPUs going past 6 and the pointless HEDT closing 20.
EPYC is gaining market share rapidly. Also, even though low core count server hardware still sells a lot, high core count hardware is gaining momentum rapidly. Virtualization and the increasing parallelization of software makes this a given. Then again, it's natural that they sell less, given that you'd need 2-3 4-core chips to do the job of one 12-core, and so on.
Because IF is very RAM-dependent. Based on how many RAM compatibility issues we've seen, I bet it will have to be reprogrammed for DDR5.
So yeah... unless AMD manages to do some microcode magic, I highly doubt any of currently available Ryzen CPUs will work with a future DDR5 motherboard.
Considering that there isn't a single finished integrated DDR5 controller in existence, it's obvious that DDR5 and current-gen Ryzen won't be compatible. I seriously doubt they'd be able to hack DDR5 support into their DDR4 controller - RAM isn't that simple. While IF (and particularly APUs) stands to gain a lot from faster RAM, we're still a few years out from consumer adoption of DDR5. 2020 seems like a well though out line.
Yup. Most don't care. I don't know if you've heard, but there's a phenomenon called "laptops" and it squashed desktops so much that most people only see desktops at work. And these office desktops are not AMD-powered, because, frankly, AMD seems not to care much about their PRO lineup.
The PRO lineup is gaining momemtum, but for markets like this, turnaround is slow. There's
a lot of validation and testing required, not to mention far more competitive volume licencing prices in the business desktop market. Still, they're arriving, slowly but steadily. Same goes for laptops. Raven Ridge is showing up in ever more designs, including premium ones. Convincing OEMs to switch takes time, but AMD is gaining. And that's good for everyone.
On the other hand, I bought a 1600X when it launched, and even if the 2600x is measurably faster across the board, I'm still very happy with mine. The 2700X is a crazy powerful processor, and even if it's bound to be eclipsed by 3000-series hardware in 6-8 months, that doesn't mean it won't be good any more at that point. It's still going to be great.