Latest rumors suggest DDR5 isn't coming for AMD until 2022 though, and 2021 might see "Zen3+"-like refresh instead (which would not be supported on 400-series motherboards
).
- Status
).event then I think the 1700x will last into 2022 , I just rather invest in a new platform than stick to the older one. I can always get a 3x00 series part cheaper next yearLatest rumors suggest DDR5 isn't coming for AMD until 2022 though, and 2021 might see "Zen3+"-like refresh instead (which would not be supported on 400-series motherboards
D
Deleted member 13524
Guest
Still, it does feel somewhat wrong to invest in a dual-channel DDR4 platform in late 2020 / early 2021.Latest rumors suggest DDR5 isn't coming for AMD until 2022 though, and 2021 might see "Zen3+"-like refresh instead (which would not be supported on 400-series motherboards
At least with a quad-channel DDR4 platform with 8 dimm slots I'd get enough headroom for gradually upgrading the memory amount and not staying behind on bandwidth.
hoom
Veteran
As much as I've always liked having the possibility of upgrading CPU on same mobo I've literally never done it despite building my own PCs since late '90s, every time I've bought a new CPU I bought a new Mobo with it.
I actually really wouldn't mind if we got CPUs permanently attached to Mobos as long as it came with a decent cost/power efficiency benefit & didn't excessively cramp my ability to get the price/feature combo I want.
I was reading some stuff: 'the return of chiplets', we're talking IO chip + chiplets right, not return to Zen1 style single-chip in different configs? I still kinda like the latter better.
Brand new 'unified' cache, suggested as L3 shared between 8 cores instead of split between 2* quad-core CCXes.
But could that be something else like a huge unified L4 on (/HBM off) the IO chip that we got excited about the possibility of last year?
Perhaps more practical & still major improvement would be if L3 got changed to an Inclusive pre-fetched cache without wrecking latency.
I'm a bit suspicious the claims of massive IPC increase is largely due to doubled FPU (will it be 2* 256bit or will they be supporting 512?) which may mean the actual practical IPC increase is much lower than the headline numbers.
I actually really wouldn't mind if we got CPUs permanently attached to Mobos as long as it came with a decent cost/power efficiency benefit & didn't excessively cramp my ability to get the price/feature combo I want.
I was reading some stuff: 'the return of chiplets', we're talking IO chip + chiplets right, not return to Zen1 style single-chip in different configs? I still kinda like the latter better.
Brand new 'unified' cache, suggested as L3 shared between 8 cores instead of split between 2* quad-core CCXes.
But could that be something else like a huge unified L4 on (/HBM off) the IO chip that we got excited about the possibility of last year?
Perhaps more practical & still major improvement would be if L3 got changed to an Inclusive pre-fetched cache without wrecking latency.
I'm a bit suspicious the claims of massive IPC increase is largely due to doubled FPU (will it be 2* 256bit or will they be supporting 512?) which may mean the actual practical IPC increase is much lower than the headline numbers.
D
Deleted member 13524
Guest
I actually did that just last upgrade even though I had decent motherboard with Zen 2 BIOS, went from Gigabyte/Aorus AX370 Gaming 5 (or something) to Asus Prime X570 Pro. The compatibility did allow me to hunt down nice price for the X570, though, saved me over 50 euros.Let me guess, you've been purchasing Intel consumer platforms all this time?
D
Deleted member 13524
Guest
Erm.. but you also spent a bunch of money on an upgrade you didn't need to do?
Yes, I did spend some money I didn't really need to, but in turn I got in my opinion a better and more importantly better looking board.
(I like having lighting in my case, but I prefer it plain white (+ sadly one effect color because can't change the LED colors on my sound card or video card, lucky for me they both have red LEDs), the Gigabyte board had alone 3 different shades of "white" in it's lighting)
D
Deleted member 13524
Guest
Hey! 3 different shades of "white" LED light on motherboard alone, who could possibly endure such horrors?!Dude..
hoom
Veteran
I think I've only had 3 Intels: a Core2 Duo, a Core2 quad & this i5.
Time to get the soldering iron out.
Rootax
Veteran
Zen 3 hasn't been delayed, they re-confirmed a week ago that Zen 3 and more specifically Milan for late 2020. They haven't made any similar launch window promises for Ryzen-Zen3, so even if it would be released in 2021, it wouldn't mean it's delayed.
But at the moment it looks like anything any random chinese forum says and gets translated is treated as gospel and preached by the usual suspects.
D
Deleted member 7537
Guest
Hopefully it doesn't get delayed. I was reading a post on reddit from guy that has upgraded from a I7 3770k to a I7 10700k and has experienced a 15% performance improvement in AC:Odyssey. I really don't want to upgrade for that kind of improvement lol.
Maybe Ryzen was postponed, but not Epyc?
From a pure business perspective... why compete with yourself? Of course it would be nice if they kept their cadence but, despite the progress they've made in sales and paying off debt, AMD is still cash strapped. Unless they have slots at TSMC that can't be renegotiated it would be a valid option.
Epic on the other hand would probably do well to be released. AMD is gaining on the consumer side as Intel falters, but on the server side Intel has a majority stake with, I imagine, swathes of Intel infrastructure already invested by these companies. AMD would do well to be so performant as to be the only natural choice despite the cost of infrastructure adaptation.
Intel will get back on its horse in due course, and in a much shorter time than AMD did. So this might be one of few chances for AMD to maximise their profitability by giving up cadence for the foreseeable future on the consumer side.
Of course, I'm only speculating. Wildly.
Epic on the other hand would probably do well to be released. AMD is gaining on the consumer side as Intel falters, but on the server side Intel has a majority stake with, I imagine, swathes of Intel infrastructure already invested by these companies. AMD would do well to be so performant as to be the only natural choice despite the cost of infrastructure adaptation.
Intel will get back on its horse in due course, and in a much shorter time than AMD did. So this might be one of few chances for AMD to maximise their profitability by giving up cadence for the foreseeable future on the consumer side.
Of course, I'm only speculating. Wildly.
hoom
Veteran
Disagree.
The last time they had real market share they eased off on cadence about when Athlon 64 X2 came out and its taken this long to get back to being really competitive.
They need to keep executing with high cadence, really gain & sustain a big chunk of market share.
If its true & not due to some late technical issue I'd prefer it to be because Intel has some kind of 'slow the fuck down' clause in the X86 license than AMD thinking its safe to go-slow.
The last time they had real market share they eased off on cadence about when Athlon 64 X2 came out and its taken this long to get back to being really competitive.
They need to keep executing with high cadence, really gain & sustain a big chunk of market share.
If its true & not due to some late technical issue I'd prefer it to be because Intel has some kind of 'slow the fuck down' clause in the X86 license than AMD thinking its safe to go-slow.
I don't think AMD eased off as much as Intel flexed their muscles.
Intel launched the Core 2 with a big leap in memory speculation (alias detection of speculated loads before the address of pending stores is known) which caused a massive leap in performance. This brought Intel on par with AMD even though AMD's cpus had the memory controller integrated on the CPU die. When Intel brought the memory controller on die with Nehalem, they crushed AMD performance-wise. Intel finally drilled the last nails in the coffin with Sandy Bridge, which reduced power consumption (the Westmere refresh of Nehalem was mostly sold in the workstation/server market because of the 100+W TDP).
All Intel's troubles stem from its process mishaps. Their 10nm process node took things too far; Quad-patterning finfets, contact-over-active-gate and cobalt interconnect resulted in a big negative impact on yield. To this day, their biggest 10nm CPU is a quad core with integrated GPU (where you can easily fuse off broken EUs) with a modest 146mm² die size.
I expect EUV to be the great equalizer in the next few years.
Cheers
- Status
