Printable View
So it looks like Zen2 is 2019 and Raven Ridge is 2018:
https://videocardz.com/72934/rumor-a...n-5-pro-mobile
Also,the Ryzen 5 PRO appears to not have a much faster GPU than an AMD A12? It seems 11 CUs are confirmed.
https://i.imgur.com/i30UKES.jpg
https://i.imgur.com/JDzfyFe.jpg
Perhaps the GPU performance is with the lower TDP though? Just a guess.
Also why did I think Raven Ridge was coming sooner than that? I thought we'd see it in laptops at least over the next quarter?
Edit: I knew I wasn't imagining it! http://ksassets.timeincuk.net/wp/upl...formance-2.jpg
Interesting stuff. Looks like AMD are starting their own "tick-tock"Quote:
Originally Posted by CAT-THE-FIFTH
Any leaks about what we can expect from Pinnacle ?
Don't think so. There have been device refreshes like Richland between the major updates for as long as I can remember (K5-166 was an update over K5-133 not a clock change), but they don't seem to tie in with process changes.Quote:
Originally Posted by Biscuit
Ryzen 5 Mobile might not have all 11 CUs enabled (I'd imagine that'll be saved for Ryzen 7 Mobile). If it was an 8 CU part it'd make sense that it's got similar IGP performance to BR A12.Quote:
Originally Posted by CAT-THE-FIFTH
Don't we have confirmed news from AMD that the next round of enhancements will come from a move to GloFo's 12LP process though? I mean, it's not really even a half-node move, but it is a process change, which would make Pinnacle Ridge a ... well, whichever part of tick-tock was the process bit ;)Quote:
Originally Posted by DanceswithUnix
EDIT: actually, 14nm -> 12nm probably *is* a half-node, isn't it...? :confused:
If Pinnacle Ridge is a straight process change, then how does that map in with the APU Raven Ridge to Picasso which is one cycle out of phase? That would make both cycles a process change, which if 7nm comes in for 2019 could be the case but it throws out the whole nonsense of tick-tock. Intel never actually did tick-tock other than on marketing slides, their process change devices were covered in tweaks and their architecture changes were boringly incremental such that it was hard to tell the difference without checking the marketing slides. I doubt AMD are making any attempt at tick tock when they haven't in the past and it is clearly failing Intel now.Quote:
Originally Posted by scaryjim
Basically masks are damned expensive to make, so I expect cash strapped AMD to make good use of spinning new masks, and just change what needs to be changed rather than tick tock which was supposed to be about risk reduction. It might even be going the other way, perhaps they have a bunch of fixes that they want in place and felt they may as well do them on the latest 12nm platform, we can only guess.
My best guess is that Raven Ridge is on 14LLP (LPP?) - it's been sampled this year and was meant to be released this year, but perhaps it's been pushed back to early next year "for some reason". Then middle of next year Pinnacle Ridge launches on desktop on 12LP, and start of 2019 Picasso launches on mobile on 12LP, and Matisse intros Zen 2 ... also on 12LP? Or perhaps on 7nm. So they're pushing desktop first on each new process, then migrating mobile later.Quote:
Originally Posted by DanceswithUnix
I doubt AMD will just settle for a process change though - previous slides have mentioned Zen+ cores, and they've publicly said they've identified key areas where they can architecturally improve the performance of Zen, so I'd guess they'll make small quick-win tweaks at the same time as shifting to 12LP.
According to http://wccftech.com/amd-pinnacle-rid...g-raven-ridge/ Matisse is on 7nm and Picasso is on 12nm down from the 14nm Raven Ridge and 28nm Bristol Ridge. So going left to right both for desktop and mobile every part is a process change as well as sometimes a sizable architecture change!
Lawks, that's a bold strategy - although given their current IPC and clock deficit to Intel I can understand why they'd do it.Quote:
Originally Posted by DanceswithUnix
Interesting to see them staggering the shift to a new node between desktop and mobile though ... I guess on desktop you've got more TDP to play with so you get a bigger margin on your bins while the process beds in, then when you shift to mobile you have a more mature process so you can do more fine tuning for lower TDPs....?
I do have one bone to pick with that article though:
That's not amazing at all. Maintaining compatibility between generations have pretty much always been AMD's bag (which is why 2010's Phenom II x6 processors will work in 2006's AM2 motherboards), and besides I'm sure they've explicitly stated that socket AM4 will span several generations of Zen-based processors. So it's absolutely expected and predictable that 2019's Zen 2 processors will be a drop in upgrade on AM4 motherboards (although I suspect they'll require BIOS support from mobo manufacturers, at which point YMMV).Quote:
In 2019 we’ll see AMD launch its first CPU products based on the Zen 2 core built on 7nm technology. Amazingly, these processors will be compatible with AMD’s current AM4 socket
Guess its a product of them being out of the market so long, and Intel having free reign to squeeze as much cashola out of customers as they possibly can. Its now "amazing" that things have gone back to the way they were YEARS ago :/Quote:
Originally Posted by scaryjim
A socket that is good for a 28nm APU and a 7nm CPU, 4 full process iterations, that has to be worth at least a raised eyebrow even for AMD? Perhaps modern VRM requirements mean boards should be able to cope with a wide requirement range anyway.Quote:
Originally Posted by scaryjim
It's pretty good I guess, but I strongly suspect they were designing the socket ground-up with longevity in mind after the rapid churn through AM3+ and FM1/2/2+. AM2 had CPU compatibility for three full process nodes (90nm down to 45nm), even if it wasn't the primary socket for all of that period.Quote:
Originally Posted by DanceswithUnix
Of course the rather disjointed nature of AMD's CPU development over the last few years means they're covering those four process nodes within 2 years of retail launches, and 2 years - or even 3 - isn't that long for platform longevity...
*sigh*, fencepost error:Quote:
Originally Posted by DanceswithUnix
28, 20, 14, 10, 7
That's five nodes worth. Apologies for brain fade...
https://www.youtube.com/watch?v=S6yp7Pi39Z8
https://www.youtube.com/watch?v=S6yp7Pi39Z8
Good video on RAM speed and latency. It seems 3200MHZ DDR4 and low latency seems the best combo.
https://i.redd.it/dg8tg16akmpz.jpg
The chart is from this investor presentation slide pack:
http://phx.corporate-ir.net/External...25811537948956
Not exactly on topic but I wish I knew that a long time ago. :(Quote:
Originally Posted by scaryjim