1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

AMD Ryzen Discussion

Discussion in 'Hardware' started by Gizmo, Feb 6, 2017.

  1. Gizmo

    Gizmo Chief Site Administrator Staff Member

    Joined:
    Dec 6, 2012
    Messages:
    2,230
    Likes Received:
    156
    Trophy Points:
    63
    Location:
    Webb City, Missouri
    Home page:
    I haven't fully committed to the Ryzen build yet, and won't until I see some actual real benches, but I'm cautiously optimistic about what I've seen.

    I'm also hoping to go with one of the Vega GPUs, now that AMD's Linux drivers seem to have decent performance.
  2. ThunderRd

    ThunderRd Irreverent Query Chairman Staff Member

    Joined:
    Dec 17, 2012
    Messages:
    2,756
    Likes Received:
    87
    Trophy Points:
    48
    Location:
    Northern Thailand, the Land of Smiles
    Home page:
  3. Gizmo

    Gizmo Chief Site Administrator Staff Member

    Joined:
    Dec 6, 2012
    Messages:
    2,230
    Likes Received:
    156
    Trophy Points:
    63
    Location:
    Webb City, Missouri
    Home page:
    IMO, that's complete rubbish. That's just my opinion, though. I don't know any more than the next guy, but there's just a lot of things that don't make sense.

    Some of it MIGHT be believable, if you squint hard enough, but the big sticking point for me is the 6-core thing. So far, everything we've heard on Ryzen is that the CCX is designed around a 4-core module (that's directly from AMD, not rumor mill). That means that the 8-core version uses 2 CCXs. Presumably, AMD can make a 4-core with 1 CCX and the memory controller, which will be about 60% the size of the 8-core version (because of the logic which can be shared between 2 CCXs, but HAS to be present for a single CCX).

    The 8C/16T versions will definitely be the SR7 high-end chips, everyone agrees on that. But there are people suggesting that AMD will be producing a 6C/12T version for the SR5 segment.

    That makes ZERO sense from a manufacturing perspective.

    Think about it: SR7 is high-end; those people will want the full monty. SR3 is low-end, those people are building cheap-as-chips systems that are low power both in terms of watts and capability. 4C/8T is probably overkill for them, but ok, we can maybe see that argument. That leaves SR5 as the 'mainstream' market, the place where the MAJORITY of chips will be sold. And that's squarely where any hypothetical 6C/12T chip would land, if you assume the above layout.

    The argument for a 6C/12T chip is that it would allow AMD to salvage defective 8C/16T chips. That's certainly true, but consider this from a manufacturing perspective: AMD would be creating a market that it could only profitably serve with defective chips. Any other scenario would require AMD to DELIBERATELY CRIPPLE the MAJORITY of their production.

    That seems grossly stupid.

    What makes more sense to me (and remember, this is just my guess, which is no more accurate, and probably less accurate than most), is something like this:

    SR7 - High end 8C/16T, dropping to 8C/8T at the lower end.
    SR5 - High end 4C/8T, dropping to 4C/4T at the lower end (defective 8C get sold here, as well as purpose-built 4C).
    SR3 - Not to be seen until possibly 2Q or 3Q, but 2C/4T APU. (APU takes up half the CCX)

    I could see an argument for SR5 being 8C/8T, and maybe SR3 being 4C/4T, primarily because Intel have 4C/4T i5s, and 2C/4T i3s. Thing is, they NEED a 4C/8T part at the SR5 price-point, in order to recover the most money from the defective 8C parts.

    There's also some room in here for reduced-cache chips; the current architecture has 8M of L3 cache per CCX, but rumor says that half the cache can be disabled (presumably for defect recovery). This makes sense, since the cache is roughly half the total chip area, and is FAR more likely to have a manufacturing defect that any of the cores.

    BTW, I've seen some rumors suggesting that AMD's manufacturing yields are north of 75%, which is roughly in line with what I'd expect given their aggressive positioning. That makes it easier for AMD to manufacture good product, but conversely makes it more difficult to justify schemes that involve deliberately crippling cores in order to 'create' a product.

    Again, this is just my thinking, and I'm old and stupid, so I could be WAY off base.
  4. ThunderRd

    ThunderRd Irreverent Query Chairman Staff Member

    Joined:
    Dec 17, 2012
    Messages:
    2,756
    Likes Received:
    87
    Trophy Points:
    48
    Location:
    Northern Thailand, the Land of Smiles
    Home page:
    But what will they have to compete with the socket 2011 octos?

    I don't see any indication there will be anything soon.
  5. cloasters

    cloasters Moderator

    Joined:
    Jul 3, 2013
    Messages:
    8,383
    Likes Received:
    82
    Trophy Points:
    48
    Very interesting, thank you gentlemen!
  6. Gizmo

    Gizmo Chief Site Administrator Staff Member

    Joined:
    Dec 6, 2012
    Messages:
    2,230
    Likes Received:
    156
    Trophy Points:
    63
    Location:
    Webb City, Missouri
    Home page:
    I'm not following you? The Ryzen is DESIGNED to DIRECTLY compete against the Intel 8C/16T CPUs. Their demos have showed them competing with and even BEATING the 6900K, which is a socket 2011 v3 i7 chip (broadwell-e).
  7. ThunderRd

    ThunderRd Irreverent Query Chairman Staff Member

    Joined:
    Dec 17, 2012
    Messages:
    2,756
    Likes Received:
    87
    Trophy Points:
    48
    Location:
    Northern Thailand, the Land of Smiles
    Home page:
    You're right, of course, that wasn't what I meant to say.

    I was actually trying to refer to the S2011 hex cores, like the 6800K and 6850K. If you're right [about the 6-core status], there won't be a horse in that race.
  8. Kaitain

    Kaitain Active Member

    Joined:
    Jul 24, 2016
    Messages:
    373
    Likes Received:
    34
    Trophy Points:
    28
    AMD do have form in the direction of deliberately down-rating hardware to fill the market. This was, as I vaguely recall in the dim and distant past, why their Athlon and AthlonXPs were so good for overclocking...

    I suppose commercially the argument is, which is cheaper - running one wafer design for all 8C Opteron/Ryzen systems, then getting busy with the laser to eliminate the features you don't want to sell, or having a dedicated template for each product line.

    As usual, I speak from ignorance, but other industries would tend to make one product and downgrade it to suit the market if there's negligible difference in material cost and big difference in operating cost.
  9. ThunderRd

    ThunderRd Irreverent Query Chairman Staff Member

    Joined:
    Dec 17, 2012
    Messages:
    2,756
    Likes Received:
    87
    Trophy Points:
    48
    Location:
    Northern Thailand, the Land of Smiles
    Home page:
    I'm also actually interested in whether there will be a 6 or 8 core Kaby Lake for S1151; that might be important for me. Right now there isn't anything over 4 cores that uses the platform.
  10. Gizmo

    Gizmo Chief Site Administrator Staff Member

    Joined:
    Dec 6, 2012
    Messages:
    2,230
    Likes Received:
    156
    Trophy Points:
    63
    Location:
    Webb City, Missouri
    Home page:
    There's a bit of difference, I think, between deliberately destroying hardware in order to make a product that you are going to sell more cheaply, and downclocking an existing product because you have more parts than you can sell at the high end.

    In the case of the 4C vs. 8C scenario, I would think the economics would make it eminently sensible to have a specific design for 4C and 8C chips (one CCX vs. two), given the relative difference in die area and the fact that the 8C chip is rumored to be around 225 square millimeters. However, like you, I don't know enough about the economics involved to really say with any authority.
  11. Gizmo

    Gizmo Chief Site Administrator Staff Member

    Joined:
    Dec 6, 2012
    Messages:
    2,230
    Likes Received:
    156
    Trophy Points:
    63
    Location:
    Webb City, Missouri
    Home page:
    I think the AMD 8C chips will be priced to compete with the Intel 6C chips. Pay the same price, get 2 extra cores, what's not to like? That's just a guess though.
  12. Kaitain

    Kaitain Active Member

    Joined:
    Jul 24, 2016
    Messages:
    373
    Likes Received:
    34
    Trophy Points:
    28
    You're quite correct wrt the 4C vs 8C scenario, so yes there would be two clear product lines based on number of CCX. I was thinking along the lines of 8C vs. potential 6C, and 4C vs. potential budget 2C, for which there's no significant difference between a faulty chip and a deliberately degraded chip.
  13. Gizmo

    Gizmo Chief Site Administrator Staff Member

    Joined:
    Dec 6, 2012
    Messages:
    2,230
    Likes Received:
    156
    Trophy Points:
    63
    Location:
    Webb City, Missouri
    Home page:
    Agreed, there's no significant difference between a faulty chip and deliberately degraded one from the standpoint of the manufacturing process. However, typically when you contemplate a maneuver like that, you are doing it in anticipation of creating parts via an alternative process that allows you to increase your effective yield. The goal is to increase your net profitability, not reduce it. That is precisely what would happen in the case of a 6C product, which can only exist by deliberately crippling product (whether that product contains already defective items, or a surplus of otherwise good silicon). On the other hand, taking failed 8C and turning them into 4C IS an effective way of bolstering supply of a product that you are ALREADY making, just via an alternative method. This is more akin, IMO, to your analogy earlier of downclocking.

    Let's look at it this way: when I manufacture a batch of AthlonXPs, all of the chips are identical in every respect. I know, based on historical yields, that some of those chips will clock very well but others will barely make the cut. I also know what I can expect to sell for each speed bin. I have no way filling the speed bins to make the manufactured yield equivalent to the desired yield, so I may have to take some higher clocking chips and sell them at lower prices, but that is understood from the outset, and the physics of the process don't allow me to do otherwise.

    This is quite different from manufacturing a bunch of Ryzen 8C chips, where I have to actually cripple part of my otherwise good yield in order to make a cheaper product. Here, I DO have a way of affecting the outcome by simply changing my product configuration. In this scenario, it makes little sense to design the product such that I have to destroy good yield in order to make MOST of my product.

    I suppose it might make sense if the anticipation is that the 6C CPU is a niche product with small sales figures, but I don't really see that happening. It might also make sense if chip yields are such that MOST of the 8C product is defective otherwise, but I don't really see that happening either; otherwise we end up with chip prices like Intel's 6900K.

    Of course, all of my impassioned analysis is worth sod-all at the end of the day, because I'm making it with a blindfold and earmuffs. The only way this will be definitively answered is when AMD reveals the product lineup, and I'm certain I'll be wrong. The only question is "How wrong?".
  14. Gizmo

    Gizmo Chief Site Administrator Staff Member

    Joined:
    Dec 6, 2012
    Messages:
    2,230
    Likes Received:
    156
    Trophy Points:
    63
    Location:
    Webb City, Missouri
    Home page:
    It's worth noting that Intel's Broadwell-e CPUs are apparently based around a 10-core CPU, which is then chopped down to get the 6C and 8C parts. This is almost certainly a significant portion of the reason why the 6C 6800K is running around $420, with prices rapidly going north from there until you get to the 10C 6950X at about $1700. I would expect the $420 price point represents the minimum at which Intel can make a reasonable profit on that die, which has got to be HUGE.
  15. Gizmo

    Gizmo Chief Site Administrator Staff Member

    Joined:
    Dec 6, 2012
    Messages:
    2,230
    Likes Received:
    156
    Trophy Points:
    63
    Location:
    Webb City, Missouri
    Home page:
  16. Kaitain

    Kaitain Active Member

    Joined:
    Jul 24, 2016
    Messages:
    373
    Likes Received:
    34
    Trophy Points:
    28
    Good find, thanks Gizmo!
  17. Daniel~

    Daniel~ Chief BBS Administrator Staff Member

    Joined:
    Dec 17, 2012
    Messages:
    11,172
    Likes Received:
    161
    Trophy Points:
    63
    Location:
    Greenwater WA
    Home page:
    Well somebody better pretty damn quick build better than stock or poor Daniel may just fall over.
    Come on guys, it's been ever so long since any of you took me for a ride by chasing the blue smoke.":O}
  18. ThunderRd

    ThunderRd Irreverent Query Chairman Staff Member

    Joined:
    Dec 17, 2012
    Messages:
    2,756
    Likes Received:
    87
    Trophy Points:
    48
    Location:
    Northern Thailand, the Land of Smiles
    Home page:
    Gizmo, did you ever build anything with that dual Xeon setup you bought 4 or 5 years ago?
  19. Gizmo

    Gizmo Chief Site Administrator Staff Member

    Joined:
    Dec 6, 2012
    Messages:
    2,230
    Likes Received:
    156
    Trophy Points:
    63
    Location:
    Webb City, Missouri
    Home page:
    It ran as a file server for a while before it died. I tried to do some overclocking with it, but was never very successful. Dual CPUs just aren't very conducive to OC, and the mobos aren't really designed with that in mind either.
  20. Gizmo

    Gizmo Chief Site Administrator Staff Member

    Joined:
    Dec 6, 2012
    Messages:
    2,230
    Likes Received:
    156
    Trophy Points:
    63
    Location:
    Webb City, Missouri
    Home page:

Share This Page