PCPER discussion @ https://www.youtube.com/watch?feature=player_detailpage&v=Pze0QbfLpD8#t=1976
-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
Leaked AMD roadmap schedules 14-nm bonanza for 2016
- Thread starter LordOfChaos
- Start date
GhostTrick
Banned
Apparently K12 is a much 'fatter' core compared to A57, given how AMD pack 2x K12 (14nm) vs 4x A57 (20nm) in the same SDP (God, I hate this metric).
(a cross-post from the other similar thread)
I think it's the same way as Nvidia Denver or Apple Cyclone. Bigger cores, customised but based on ARMv8.
It's interesting to note that AMD gave up on x86 in the ultra low power segment.
I'm wondering if we could see Amur in the next nintendo handheld.
blu
Wants the largest console games publisher to avoid Nintendo's platforms.
It's perfectly logical to me : )
GhostTrick
Banned
TheDarkTriforce
Banned
Can someone please explain me the benefits of building CPU's in smaler nm sizes? Thanks 
Can someone please explain me the benefits of building CPU's in smaler nm sizes? Thanks
Lowers production cost and reduces power usage of the chip.
Nostremitus
Member
Wow, so the quad core will be 8 threaded and the 8 core will be 16 threaded? I wasn't expecting that...
Can someone please explain me the benefits of building CPU's in smaler nm sizes? Thanks
Putting it simply, smaller transistors means you can either pack more of them in the same area or the same amount of transistors take up less space. It also leads to lower power draw and/or higher performance.
We're slowly heading towards an age where the methods to shrink nodes become increasingly more expensive and harder to pull off. This is why companies who can't afford these new bleeding edge nodes are sticking to older ones; they're a lot cheaper even when your chips end up faster/smaller. At these small scales quantum mechanics are starting to become a problem too. Electrons are harder to contain to make clear on or off states (0 or 1). Seriously, transistors are becoming so small you can almost count the individual atoms.
In reality this stuff is extremely complicated and it even boggles my mind. If you, or anyone else, wants to read up a bit more about how these things get designed and produced, read this TechReport article.
As for Zen, if AMD can come close to mainstream platform Skylake (probably Cannonlake by then) at a good price, they did well in my book. Skylake was my original plan to consider upgrading, but I might as well just wait to see what Zen does before I retire my 2500k.
TheDarkTriforce
Banned
TheDarkTriforce
Banned
& I ask myself "can I do everything I want with my 5 year old 875k + 4 year old 7970 and 4 year old mb-pro?". So far the answer has been yes. Back in the day when I'd encode videos, games pushed hw, and tabs/smphones weren't a thing, I'd want new some new piece of PC kit every six months or so. Anand is right, enthusiast PC stuff is dull as wet paint now.
LordOfChaos
Member
Source? I was under the impression that it would not be faked out by hyperthreading.
https://www.youtube.com/watch?v=JxUPJdcChzE
They needed 4 threads, not 4 physical cores. So it can be 2 core 4 threads (i3) or 4 core 4 threads (i5).
Does this have anything to do with the way console versions are designed? I remember reading a post about this somewhere a while ago. The person mentioned that some PC ports straight up don't work on anything that has less than four cores/threads because they're designed to run off of core 3 (core 0 and 1 being reserved for the OS). I don't know if that's the reason, but it would explain why recent games require four cores/threads. Obviously four cores will give you more performance if the game takes advantage of them, but that seems like such an oversight.
I don't know an awful lot about the software side of things, which is why I'm asking.
LordOfChaos
Member
Hmm. So a dual socket P4 with HT would boot that thing, but that Pentium Anniversary wouldn't
Anyways. I do hope that the hard minimum for these things doesn't become a trend, just trust us to know if we want to run it or not. They could lock it to four threads, but one could be using a very powerful dual core that bests their requirement for a quad core.
Nostremitus
Member
That's a bummer...
Irobot82
Member
Could just be PR control, nobody likes to have their big reveal stolen from under them.
Nostremitus
Member
Makes sense. I plan on replacing my 2500K next year and I would love a competitive market.I'm really hoping AMD bring their A game and learned from the mistakes on their current CPUs.
Memorabilia
Member
*sigh* You sound hostile. And your comments don't jive with reality when it comes to actual, real benchmarks that are available all over the web. I'm getting smooth frame-rates on my HTPC in every game I throw at it. Battlefield 4 is pegged at 60fps/1080P/Ultra. I turn down the settings to high & the fps stays pegged at 60fps in online MP with up to 64 players. This is on a $65 760K CPU coupled with a GTX 960. Sure, my main gaming rig hooked up to a 30" Dell (3007 IPS, oldie but goodie) is a different story because I run games on that at 2560 x 1600. That's where the extra IPC grunt really comes into play. It also costs about $180 more for the i5-4690K CPU I put in that system. Plus the motherboard was about $100 more expensive than typical AMD offerings. I love both systems. They both have their place in my household. Heck, I built my fiance' a little HTPC/light gaming rig using a A8-7600. She mostly plays Dungeon Defenders. I set the rez to 1280 x 800/medium quality & it looks a lot better than the same game on the 360, runs at a stable 30fps even on the most busy maps with us playing split screen. No bad for a $90 APU. Sure, if she continues gaming on Steam, I'll probably upgrade it later by adding in a GPU...probably a single slot/low profile GTX 750 Ti. All in all, not bad for a system that is actually smaller than a Xbone.
People will always want to cherry pick Crysis 3 which is extremely CPU dependent or unoptimized shite like Unity which can bring any CPU/GPU combo to its knees. But all in all AMD still produces price/performance competitive parts at or under the $100 price point which will generally provide a smooth 30-60fps in most games at 1080P when paired with a good GPU (AMD is arguably application specific competitive with the 8320 at ~$130). *Most games* of the 300+ I own & have played on Steam will play smoothly. If I hook my most expensive system up to my big screen it is still limited to 1080P/60Hz (Vizio M series advertised at 240Hz, which is marketing nonsense). It just depends on the application/need.
AMD's problem is remaining a viable business due to the difficulty of squeezing profit margins out of extremely low cost chips (hard to do). Switching to SMT+CMT in Zen should help them be more competitive and get those margins back on track.
I think the devs checked the various HW surveys, saw a good enough installed base of 4-core/thread users, decided to use it as base spec.
Memorabilia
Member
Thanks for the link...I always find these sorts of discussions interesting.
I'm in the same boat. I'll either be upgrading my 4690K to Zen or Skylake. I'm not sure what to expect from Skylake as the rumor mill right now is just as muddied for it as it is for Zen. We've been getting the tick tock 5% increases with each new revision, so it seems estimates have a fairly wide range from 5% - 20% depending on how optimistic people are. If we split the difference and Skylake achieves a 10% IPC improvement and Zen matches Haswell, then I think AMD is back in the game. Then is just becomes a matter of pricing/binning. If there's only 10% difference at the top end and that filters down to all the price points I'd be pretty tempted to upgrade to Zen assuming there's a tad bit of cost savings to go along with it. If there's a 20%+ IPC difference then Skylake may be in my future, assuming they aren't priced crazy.
brain_stew
Member
So what does everyone think AMD have to deliver to become relevant to gamers and enthusiasts again?
For me it would be:
$125
4C/8T
Ivy Bridge level IPC
Unlocked multiplier and easy 4GHz+ clocks
I think that could be a real interesting choice for mainstream "bang for buck" gaming rigs and would be an easy recommendation to anyone that can't afford an i5. You'd take a ~20% hit to IPC vs. a Skylake i3 but being unlocked would make it easy to make that difference up and then you're getting an extra 2 cores and 4 threads at the same price point. If it means offering a mainstream chip without integrated graphics (like their current Athlon range) then so be it.
At the high end a 8C/16T option for ~$200 (unlocked of course) could offer up a good alternative to an i5. DX12 will be common by that point so an extra 12 threads could be a decent alternative to an extra 20% IPC. As long as single threaded performance is good enough for the back catalogue of DX11 games at 60fps (and 4ghz+ with Ivy Bridge level IPC absolutely is) then you'd have a really compelling option for my money.
I just hope we get this level of competition as Intel have been allowed to become complacent over the last few years. We really should have an unlocked ~$100 i3 by now and i5/i7 chips should have moved to 6 cores with Haswell but without any competition there been no need to move things forward.
Budget CPUs have really fallen away in recent times. Back in the Core 2 days you could buy an ~$80 E2180, easily OC it to 3ghz and out perform a stock $300+ e6600. There's just no chips that offer that anymore. The Pentium Anniversary was nice but 4 threads has become basically a minimum spec these days so it really needed hyper threading to let it fly. Even if Zen can't beat Intel at the high end (and I don't expect it to) there's still a big gap in the market for something like a modern E2180.
For me it would be:
$125
4C/8T
Ivy Bridge level IPC
Unlocked multiplier and easy 4GHz+ clocks
I think that could be a real interesting choice for mainstream "bang for buck" gaming rigs and would be an easy recommendation to anyone that can't afford an i5. You'd take a ~20% hit to IPC vs. a Skylake i3 but being unlocked would make it easy to make that difference up and then you're getting an extra 2 cores and 4 threads at the same price point. If it means offering a mainstream chip without integrated graphics (like their current Athlon range) then so be it.
At the high end a 8C/16T option for ~$200 (unlocked of course) could offer up a good alternative to an i5. DX12 will be common by that point so an extra 12 threads could be a decent alternative to an extra 20% IPC. As long as single threaded performance is good enough for the back catalogue of DX11 games at 60fps (and 4ghz+ with Ivy Bridge level IPC absolutely is) then you'd have a really compelling option for my money.
I just hope we get this level of competition as Intel have been allowed to become complacent over the last few years. We really should have an unlocked ~$100 i3 by now and i5/i7 chips should have moved to 6 cores with Haswell but without any competition there been no need to move things forward.
Budget CPUs have really fallen away in recent times. Back in the Core 2 days you could buy an ~$80 E2180, easily OC it to 3ghz and out perform a stock $300+ e6600. There's just no chips that offer that anymore. The Pentium Anniversary was nice but 4 threads has become basically a minimum spec these days so it really needed hyper threading to let it fly. Even if Zen can't beat Intel at the high end (and I don't expect it to) there's still a big gap in the market for something like a modern E2180.
blu
Wants the largest console games publisher to avoid Nintendo's platforms.
I think you're a tad optimistic. Intel's current similar offerings run north of $300. AMD pulling that at that price would be a small revolution, and I'd personally buy a handful of those to upgrade all my home stations. Alas.
LordOfChaos
Member
Ivy Bridge level IPC is essentially saying get to within ~10% (5-15%% on average) of Haswell IPC.... And then saying offer that with four cores, for 125 dollars.
I dunno. I want to be optimistic, but that seems like a hard jump.
AMD is quite happy to throw cores at things, but that's with their Bulldozer "module" architecture with a shared floating point unit per two cores. Four cores with the new architecture would be more costly.
SapientWolf
Trucker Sexologist
So what does that mean in terms of perceived fluidity? That could all be from one frame of stutter. It is even less clear with gsync/freesync, which is designed to smooth out uneven frametimes.What are these odd 'fps' metrics that are being thrown around? Are they rendering films or playing games?
If games like Ryse are representative of future game CPU performance then all you will need is 8 halfway decent cores.
DrROBschiz
Member
Holy crap
14 nm
We are getting close to the atomic limit here guys. Whats their next step afterwards?
14 nm
We are getting close to the atomic limit here guys. Whats their next step afterwards?
blu
Wants the largest console games publisher to avoid Nintendo's platforms.
Down to 10nm they can carry on with the existing FinFET tech. Beyond that come some exotic elements/alloys like Germanium and InGaAs offering high electron mobility. Then comes the Graphene hypothesis, Cthulhu rituals, etc.
Frametimes is the time used to render a frame. 16.2/3 ms is what every frame would render at with perfect 60fps, 33.1/3ms with 30fps and 50ms with 20fps. That slide shows how many milliseconds during the test run that the frametimes where over 50ms.
It is important for percieved fluidity because those bad frames mean a lot more than fast frames (often faster than the screen itself). With FPS everything is averaged out so you could in the worst case have frametimes alternating between 8-33ms where the game would look and feel like 30fps while the FPS counter would show 60fps.
While g-sync helps with uneven framerates it doesn't fix this issue.
I recommend using fraps to record frametimes (it has an option for it and will spit out a spreadsheet) and then use a program like FRAFS Bench Viewer to graph them. If you find a game with an uneven framerate you get to experience what it feels like. Try running a game at around 45fps with vsync on and off if you can't and play around with framelimiters. Source games like TF2 are easy to run and should include a framelimiter in the console so you can see how 16-33ms (30-60fps) alternating (you can get this with 40~45fps with vsync turned on) is compared to solid 30 and 60fps.
Ryse is just one game and one that happens to be very straight forward and predictable. A lot of games are very unpredictable and if one thread suddenly requires a lot of processing (say you blew up a lot of barrels or hit space debris) it's better to have the single threaded performance than to lack it. A lot of old games will also not be updated and need beefy single threaded performance for the best experience. mkenyon did a big experiment showing how seemingly easy to run games like Dota and Tribes Ascend really benefited from an overclocked Intel CPU at 120hz.
LordOfChaos
Member
Intel was already there just in case you didn't know, plus their 14nm puts everyone elses to shame.
http://www.extremetech.com/wp-content/uploads/2014/12/Cell-SizeComparison.png
We'll still have a few shrinks after 14 on silicon before hitting those hard limits though, I think. 7nm and 5nm are being worked on, for after 10nm. After that, yeah, everyone will be scrambling for materials other than silicon to use, or other esoteric methods.
http://www.cnet.com/news/end-of-moores-law-its-not-just-about-physics/
ScepticMatt
Member
confirmed FAKE
But we'll get real information in 5 days.
But we'll get real information in 5 days.
https://www.planet3dnow.de/cms/14846-angebliche-amd-folie-zeigt-erste-details-von-zen/
ColonialRaptor
Member
What about the darn graphics cards?
I mean, cpu wise we've hit the land of massively diminsed returns and I doubt I'll be upgrading my CPU until they bring out some graphene based chips.... 10% gains on cpu speeds is just ridiculous especially since the older chips overclocked so well
I truly believed that the only solution to this no gains problem would be processors with ever increasing numbers of cores but the software simply doesn't take advantage of the multi core cup abilities yet and essentially I figured that we'd be seeing everything multi threaded and cpu's with 32 and 64 cores by now but I guess that's just a stupid idea when 4 cores can easily do any of the jobs currently needed.
I mean, cpu wise we've hit the land of massively diminsed returns and I doubt I'll be upgrading my CPU until they bring out some graphene based chips.... 10% gains on cpu speeds is just ridiculous especially since the older chips overclocked so well
I truly believed that the only solution to this no gains problem would be processors with ever increasing numbers of cores but the software simply doesn't take advantage of the multi core cup abilities yet and essentially I figured that we'd be seeing everything multi threaded and cpu's with 32 and 64 cores by now but I guess that's just a stupid idea when 4 cores can easily do any of the jobs currently needed.
brain_stew
Member
Confirmed 40% IPC increase vs. Excavator.
Focus on FX CPUs first (transistors spent on cores not integrated graphics so a mainstream 8C/16T CPU looks likely) and APUs second. There's some real potential for serious competition now if AMD can comfortably hit mid 3ghz clocks at stock (and 4ghz+ overclocked).
http://www.anandtech.com/show/9231/amds-20162017-x86-roadmap-zen-is-in
Focus on FX CPUs first (transistors spent on cores not integrated graphics so a mainstream 8C/16T CPU looks likely) and APUs second. There's some real potential for serious competition now if AMD can comfortably hit mid 3ghz clocks at stock (and 4ghz+ overclocked).
http://www.anandtech.com/show/9231/amds-20162017-x86-roadmap-zen-is-in
Oooooh. Sounds great to me. Huge IPC gains? CPU vs APU on desktop?
That's a fucking huge jump. I would love it if AMD became a rational CPU choice again. Wonder where this would put them vs Intel. IB levels of IPC?
LordOfChaos
Member
Could be above IB and close to Haswell. Excavator is 20% faster then piledriver, which puts it on par with Nehalem. If Zen is 40% faster then excavator/Nehalem that's around Haswell-e single core performance.
That makes me almost giddy. Haswell like IPC would put it in spitting range of Skylake.
I'll try to reserve judgement though. AMD promised the world with Bulldozer and look what happened there.
Thing too is by the time this launches Intel could be on Cannonlake potentially if they get Skylake out this year.
Bumped for news.
It seems Samsung and Global Foundries will both produce AMD's Arctic Islands family, including the flagship GPU; Greenland, on their joint 14nm FinFET LPP process, starting in Q2 2016.
http://hexus.net/business/news/components/89078-samsung-produce-greenland-gpus-amd-starting-q2-2016/
http://english.etnews.com/20151222200002
http://wccftech.com/amd-greenland-14nm-production-q2-2016/
http://www.theinquirer.net/inquirer...reportedly-start-making-chips-for-amd-in-2016
http://www.reuters.com/article/us-samsung-elec-amd-chips-idUSKBN0U500P20151222
It seems Samsung and Global Foundries will both produce AMD's Arctic Islands family, including the flagship GPU; Greenland, on their joint 14nm FinFET LPP process, starting in Q2 2016.
http://hexus.net/business/news/components/89078-samsung-produce-greenland-gpus-amd-starting-q2-2016/
http://english.etnews.com/20151222200002
http://wccftech.com/amd-greenland-14nm-production-q2-2016/
http://www.theinquirer.net/inquirer...reportedly-start-making-chips-for-amd-in-2016
http://www.reuters.com/article/us-samsung-elec-amd-chips-idUSKBN0U500P20151222
Seems like this should provide a pretty sizable leap in perf/watt.
Is it just me or do AMD chips always look pretty good on paper every year a thread like this rolls around yet always just ends up getting it's ass tore up in real world benchmarks?
I admit that I don't really pay too much attention to CPU benchmarks. I do video and audio rendering/encoding and have always just grabbed the fastest unlocked non extreme i7 at the time and just assumed that it would obliterate anything AMD had to offer me.
Is there any indication of this changing with this new leaked stuff or is it basically just going to continue to be the i7 cleaning house and the "Fast" AMD chips trading blows with the middle of the road i5?
I just want AMD to give me a chip that doesn't have any focus on gaming but rather something like a powerful multithreading Xeon 6 or 8 core equivalent that I can run a couple of virtual servers with or do CPU intense rendering with that doesn't cost an arm and a leg like the consumer octa core Intels.
I admit that I don't really pay too much attention to CPU benchmarks. I do video and audio rendering/encoding and have always just grabbed the fastest unlocked non extreme i7 at the time and just assumed that it would obliterate anything AMD had to offer me.
Is there any indication of this changing with this new leaked stuff or is it basically just going to continue to be the i7 cleaning house and the "Fast" AMD chips trading blows with the middle of the road i5?
I just want AMD to give me a chip that doesn't have any focus on gaming but rather something like a powerful multithreading Xeon 6 or 8 core equivalent that I can run a couple of virtual servers with or do CPU intense rendering with that doesn't cost an arm and a leg like the consumer octa core Intels.
dr_rus
Member
Well they've promised 2X. That's probably because of the process switch exclusively as I don't expect much changes to the GCN architecture unfortunately.
Most of us hope that Zen core which is launching next year will be able to compete with Intel's i7s.
Terbinator
Member
For the most part the FX series was just this.
Green Yoshi
Member
Please let it be inside Nintendo NX.
As for AMD's near term plans, the dual-Fiji card, codenamed Gemini (which is widely expected to be R9 Fury X2) its release as been delayed pushed into 2016, to be aligned with VR headset launches.
Anandtech has a big article up on the problematic state of multi-GPU SLI, CrossFire and Alternate Frame Rendering tech.
http://www.anandtech.com/show/9874/amd-dual-fiji-gemini-video-card-delayed-to-2016
Anandtech has a big article up on the problematic state of multi-GPU SLI, CrossFire and Alternate Frame Rendering tech.
http://www.anandtech.com/show/9874/amd-dual-fiji-gemini-video-card-delayed-to-2016
Gurgelhals
Member
NX development is probably too far along for any 16/14nm tech to be used. It's far more likely to be still based on old 28nm tech; it's cheap, reliable and the risk of any unforeseen problems cropping up when going into mass production is basically zero. And, yep, I think that this doesn't sound particularly exciting too... :-/
blastprocessor
The Amiga Brotherhood
Seriously this is exciting stuff. Anyone buying a GPU now is just throwing away money.
Wait for 14nm goodness and HBM2.
Wait for 14nm goodness and HBM2.