Talking 12nm and Zen+

One of the highlights of the Ryzen 2000-series launch is that these processors use GlobalFoundries’ 12LP manufacturing process, compared to the 14LPP process used for the first generation of Ryzen processors. Both AMD and GlobalFoundries have discussed the differences in the processes, however it is worth understanding that each company has different goals: AMD only needs to promote what helps its products, whereas GlobalFoundries is a semiconductor foundry with many clients and might promote ideal-scenario numbers. Earlier this year we were invited to GlobalFoundries Fab 8 in upstate New York to visit the clean room, and had a chance to interview Dr. Gary Patton, the CTO.

The Future of Silicon: An Exclusive Interview with Dr. Gary Patton, CTO of GlobalFoundries

In that interview, several interesting items came to light. First, that the CTO doesn’t necessarily have to care much about what certain processes are called: their customers know the performance of a given process regardless of the advertised ‘nm’ number based on the development tools given to them. Second, that 12LP is a series of minor tweaks to 14LPP, relating to performance bumps and improvements that come from a partial optical shrink and a slight change in manufacturing rules in the middle-line and back-end of the manufacturing process. In the past this might not have been so news worthy, however GF’s customers want to take advantage of the improved process.

Overall, GlobalFoundries has stated that its 12LP process offers a 10% performance improvement and a 15% circuit density improvement over 14LPP.

This has been interpreted in many ways, such as an extra 10% frequency at the same power, or lower power for the same frequency, and an opportunity to build smaller chips.

As part of today’s launch, AMD has clarified what the move to 12LP has meant for the Ryzen 2000-series:

  1. Top Clock Speeds lifted by ~250 MHz (~6%)
  2. All-core overclocks around 4.2 GHz
  3. ~50 mV core voltage reduction

AMD goes on to explain that at the same frequency, its new Ryzen 2000-series processors draw around 11% less power than the Ryzen 1000-series. The claims also state that this translates to +16% performance at the same power. These claims are a little muddled, as AMD has other new technologies in the 2000-series which will affect performance as well.

One interesting element is that although GF claims that there is a 15% density improvement, AMD is stating that these processors have the same die size and transistor count as the previous generation. Ultimately this seems in opposition to common sense – surely AMD would want to use smaller dies to get more chips per wafer?

Ultimately, the new processors are almost carbon copies of the old ones, both in terms of design and microarchitecture. AMD is calling the design of the cores as ‘Zen+’ to differentiate them to the previous generation ‘Zen’ design, and it mostly comes down to how the microarchitecture features are laid out on the silicon. When discussing with AMD, the best way to explain it is that some of the design of the key features has not moved – they just take up less area, leaving more dark silicon between other features.

Here is a very crude representation of features attached to a data path. On the left is the 14LPP design, and each of the six features has a specific size and connects to the bus. Between each of the features is the dark silicon – unused silicon that is either seen as useless, or can be used as a thermal buffer between high-energy parts. On the right is the representation of the 12LP design – each of the features have been reduced in size, putting more dark silicon between themselves (the white boxes show the original size of the feature). In this context, the number of transistors is the same, and the die size is the same. But if anything in the design was thermally limited by the close proximity of two features, there is now more distance between them such that they should interfere with each other less.

For reference, AMD lists the die-size of these new parts as 213mm2, containing 4.8 billion transistors, identical to the first generation silicon design. AMD confirmed that they are using 9T transistor libraries, also the same as the previous generation, although GlobalFoundries offers a 7.5T design as well.

So is Zen+ a New Microarchitecture, or Process Node Change?

Ultimately, nothing about most of the Zen+ physical design layout is new. Aside from the manufacturing process node change and likely minor adjustments, the rest of the adjustments are in firmware and support:

  • Cache latency adjustments leading to +3% IPC
  • Increased DRAM Frequency Support to DDR4-2933
  • Better voltage/frequency curves, leading to +10% performance overall
  • Better Boost Performance with Precision Boost 2
  • Better Thermal Response with XFR2
New CPUs, New Process, New Competition Improvements to the Cache Hierarchy: Lower Latency = Higher IPC
Comments Locked

545 Comments

View All Comments

  • ACE76 - Friday, April 20, 2018 - link

    Yes...my original memory was 16...I eventually got it to run at 2933 which is fine by me...I might upgrade to 3400 15 memory in the future...
  • TheJian - Thursday, April 19, 2018 - link

    Not sure why you're running Spectre/Meltdown for Intel when AFAIK both were yanked by MSFT and INTEL (microsoft first, then Intel finally also as they're not working properly) last I checked.

    Also, why 8700k machine having 4x8 dimms instead of AMD 2x8? If you're not comparing like systems, it kind of introduces things that cause weird benchmarks. Just read Hexus, no 1080p game lost by Intel, the rest even at 4k which nobody uses & is GPU limited anyway (I call it nobody until 10%, miles from that, even 1440p miles from that). Despite what you guys have been claiming since 660ti, people still running 1920x1200 or lower (feel free to check steam survey and add up ALL others above 1920x1200), or they have TWO cards or more. AMD claiming 1440p best res but you guys do 1080p & 4K? LOL.

    One more point after a quick read here and elsewhere: You should Have tested Intel at 2933 mem also as it is EASILY doable on Intel boards even for the Crucial. Just do timings yourself, heck you could have used the exact modules for both. People who use SPD's are lazy. Nearly every module can run on every board if you IGNORE SPD's and set it up yourself. IE PCper ran 8700k/2700x at 3400 actual.

    Darn good chip either way, not sure why AMD insists on making no money by charging less than their stuff is worth. This chip should be $400 with fan. You just pissed away much of your NET INCOME. Nice work again AMD retarded management (and that's an insult to retards). I say retarded, but really just stupid (as in can't learn from past pricing mistakes). Ignorance would mean they can learn...You're not in business to be our friend, you need to make MONEY for a few years to have R&D. Over the life of the company AMD has NOT made a dime. I don't think people get this. We need them to make NET INCOME and 60% margins like NV/Intel, not this 30% crap. Market share does you ZERO good if you make ZERO from having it. I'd rather be Apple/NV/Intel and OWN the high end and most of the profit.
  • Dragonstongue - Thursday, April 19, 2018 - link

    depends on the way of looking at things, AMD gives back what they can where they can, they absolutely need to make money, one can "assume" they are not, but then again do we really know how much Intel/Nv or Apple are making, nope, AMD has a smaller staff size far less overhead (generally speaking)

    they absolutely need to bank some, but who is to say that AMD charging $400 for product X is not directly the same thing as Intel/Nv or Apple charing $600, we do not know simple as that.

    in my books, you are in business to make money for sure, but there is no reason to gouge the crap out of those buying said product, doubly so for a company like AMD that seems to give a great deal towards the industry that benefits everyone including themselves, this is "priceless"

    We already have enough mega greed corporations out there, am glad at least one of them is charging a fair price for the product instead of gouging "just because they will buy it anyways"

    like big pharma type deal, screw that lol...if they only need to sell at say $400 to "bank" even 20%, they still are profitable, now when it comes to the very high end likely their margins are MUCH higher with little overall to them increase in cost (such as Threadripper/EPYC)

    either way they are being fair to themselves AND their customers, at the very least they are not selling more or less at cost ~5% at most margins (Bulldozer as an example) basically doubled the price likely about 1/3 less to produce them still is a win win for them, they put $ in the bank and towards RnD and we get some shiny new toys as well.

    quit your bitchin ^.^
  • TheJian - Thursday, April 19, 2018 - link

    Umm, no, we do know what they Q reports say and as noted they haven't made a dime over the ENTIRE life of the company. We also know they've lost ~8B in the last 15yrs. We also know they've released multiple new products and still aren't making money for a full year yet and barely making money in quarter here and there.

    https://finance.yahoo.com/quote/AMD/financials?p=A...
    Net income for the last 4 quarters:
    61,000 71,000 -16,000 -73,000
    Get the point genius? I'll keep bitchin until AMD sets an appropriate price to actually make 100mil for the year...LOL. NO not GROSS, that doesn't count..NET INCOME is the only thing that matters in the end. Commonly known as "THE BOTTOM LINE".

    If you're happy with AMD making ~40mil for a year, I'd say Intel/NV are laughing their butts off. New toys cost more than 40mil...ROFLMAO. Thats about the cost of taping out a 7nm chip probably as they skyrocket per shrink now. Used to be ~10mil at 40/28nm, not so now. There's a reason why Intel has problems with 14nm for a while (not now) and now 10nm too. It's not easy shrinking today and is vastly more expensive to pull it off.

    AMD has margins of 34% right now. Now look at that MASSIVE 40mil profit for the TTM (trailing 12 months since you don't seem to read balance sheets or Q reports). Comic people like you comment but don't know the numbers. You run on assumptions I can't afford as a stock investor. You know what happens when people ASSuME things right? ;)

    Oh, and AMD profits after MULTIPLE launches and how many years of re-organization? Pricing and HBM/HBM2 (kills margin on top products currently, should only be done on Titan/Quadro/Tesla type stuff with massive room to adjust), chasing console (single digit-15% margins over the life of first xbox1 according to AMD themselves), and chasing APU (which was released at $165...LOL - should have made HBM 8809g chips instead of Intel doing it - great margin on ~500 bucker). Now they massively undercharge on new 2700x. Essentially $40 off orignal $369 and free heatsink/fan thrown on top. This is DUMB, until you can prove otherwise via a Q report which you can't.

    MEGA GREED? ROFL. Add up EVERY SINGLE YEAR of AMD NET INCOME please, or please kindly shut up. They should be making a billion right now, but they chose HBM/HBM2 killing top cards margins twice, now cpus twice. The cpu looks VERY strong, why screw themselves to make a few IGNORANT people (hoping you can learn by reading a Q report, so I refrained from calling you stupid) like yourself happy? While you're doing some homework, make sure you understand how they've lost fabs, buildings, property, 1/3 engineers in the last ~5-10yrs (layoffs), and massively diluted their shares during that time also (almost a billion shares outstanding now vs. 600mil a decade ago). They are NOT healthy by any measure, so by all means prove your case or go away quietly before making a bigger fool of yourself. :) It's shortsighted to have them go out of business soon (or go completely junk status) over a few bucks on your cpu or gpu. Margins do matter, so does NET INCOME.

    https://finance.yahoo.com/quote/AMD/key-statistics...
    AMD Operating margin for the last 12 months..3.83%...LOL. Again, read something. I could go a lot deeper, but if you don't even get all this, what is the point in driving you into the ground. I have more AMD reviews to read so I can decide 8700k or 2700x. :) I was excited over Anandtech games (apps already good for me), but now have reservations again reading elsewhere...LOL.

    What money they putting in the bank? R&D dropping for last 4yrs, Nvidia/Intel up over the same.
    One more for good measure, easily understood I'd hope by ALL:
    https://finance.yahoo.com/quote/AMD/financials?p=A...
    NET INCOME last 4 years:
    43,000 -497,000 -660,000 -403,000
    So, multiple product launches last year, and barely breaking even? Never mind the previous 3 years of MASSIVE losses. If you're selling 4-5B worth of crap a year and losing 10-12% every year on that, umm, you're doing something wrong right? You're CLEARLY not charging enough correct? Doing the same thing on $250 1080 gpu next year...They are pre-announcing another BAD yearly loss...LOL. FIRE your management AMD! If NV is $350 or more next year on that speed of card, you'd better rethink your margins! What gouging? They've lost ~500-650mil a year 3 out of 4 of the last 4yrs and only 43mil in the best year in the last 5...LOL. Are you high? Your idea of margins have AMD out of the cpu race for 5yrs straight, now finally back in but still you'd have them keep doing the same stupid pricing that loses 500mil a year. UGH.

    Goodwill=PRICELESS? LOL. Stupid pricing=losses...How about charging appropriate pricing to stay in business and ADD R&D instead of reduce it? Make sense? They are NOT banking 20% and if your idea of profitable is losing 1.5B in the last 4yrs and only 43mil in what should be their BEST year in a decade, you sir are...Never mind. ;) We know margins on NV and Intel product segments. IE, gaming cards around 50% overall (80% of that from top end stuff) and workstation stuff is ~80%+. Those margins allow the low end to actually get something worth buying too.

    But hey, congrats on all your "guess work", I'd rather deal in DATA.
    https://nvidianews.nvidia.com/news/nvidia-announce...
    That's how your summary for a quarter's financial results should look. See the "RECORD" stuff in there...NO ripping people off either as everyone has other options but still buy. I'm happy with my 1070ti even at $500 during a mining war in Nov. Still a great card today. The only thing I don't like about the financial summary above is giving 1.25B back to shareholders. Dividends: no point for tech co, and share buybacks while nice, don't make the next product. R&D please. Can't be bothered to correct spelling/grammar (been up all night for a hospital visit for family). Are we done? LOL.
  • Dragonstongue - Thursday, April 19, 2018 - link

    I would say AMD is doing well only for the fact they under new leadership are profitable after being under the mud for many many years, that is what counts, paying down the bills and making some on the side...comparing to Ngreedia who is VERY overvalued not just my opinion, but whatever, not worth talking about.

    lets use a company the constantly cuts corner, that constantly screws consumers for the $$$$$$$$$$$$$$$$$$$$$$$ and nothing more, way to go...when a company such as AMD who has basically been less then broke for many years turns a very high turn over profit, I say they are doing SOMETHING right, I guess the awards they have got over the least 2 years mean jack shit huh?

    They have to invest to stay ahead of the game, hard to do when you owed billions in past debts you cannot go to making billions overnight while still making a high quality product(s)

    guess AMD should price their chips at $600+ just because they should also be greedy mofos and make everything proprietary nonsense.

    glad they at least are trying to do what they can as best as they are able, or will you also "argue" on that point sir? income year over year was UP, debts were DOWN, they are doing what is right in my books period.

    mehhh whatever, am done, simply not worth it.
  • Fallen Kell - Thursday, April 19, 2018 - link

    They are running the Spectre/Meltdown patches because they were re-released as final fixes this month. Intel released and Microsoft approved of the changes mid-March for Skylake and Coffee-Lake, and Microsoft approved the updated patch mid-March:
    https://support.microsoft.com/en-us/help/4090007/i...

    For complete information of Spectre and Meltdown with Microsoft:
    https://support.microsoft.com/en-us/help/4073757/p...

    So, yes, Anandtech is running the latest released patches, which have been approved by both Intel and Microsoft and are the recommended solution. Not all systems can be patched yet, as your motherboard provider also needs to provide a BIOS update, but for the testbed Anandtech used, the motherboard provider does have the BIOS updated with the fixes, and thus, it was benchmarked using the fully supported, patched configuration for this security vulnerability.
  • TheJian - Thursday, April 19, 2018 - link

    https://threatpost.com/bad-microsoft-meltdown-patc...
    I guess the worry is win7/2008sr2 64bit ONLY still (which affects me on multiple machines).
    “Microsoft is aware of this and looking into the matter further. This issue impacts Win7 SP1 (x64 only) and Server 2008R2 SP1 (x64 only). We are actively testing a solution, and will make it available as soon as it is properly validated.”

    Maybe I missed if they have been fixed (april patch tuesday fix this?), as I've been dealing with parents (hospital crap). But with quick checks I think win10 (used here) is ok it seems. Still odd game benchmarks based on more reviews elsewhere. Either something is fishy or everyone else did it wrong? LOL. Still think mem speed and # of dimms (2x8GB) should be the same especially since you can run at 4000 on most Intel boards (apparently maybe 470 chipset boards for AMD too). The post above was from article date Mar28th. So unless it was fixed days later, guessing win7/2008sr2 64bit varieties are still buggered (not to mention all the cpus Intel abandoned).
  • Fallen Kell - Thursday, April 19, 2018 - link

    My guess is they tested with 2 DIMMs because the Ryzen's memory controller is only dual channel, and using 4 DIMMs would mean it had to use multi-plexing to access the additional DIMMs, thus creating slightly slower performance. That said, Coffee Lake is only dual channel as well...
  • Cooe - Friday, April 20, 2018 - link

    Techradar has identical results to AnandTech with the new patches.
  • msroadkill612 - Friday, April 20, 2018 - link

    Afaik, it takes the comments section to break the big story in the review - "Meltdown Smackdown"

    :(

Log in

Don't have an account? Sign up now