AMD Ryzen 14nm Wafer Yields Pass 80% - Threadripper CPUs on track

Published by

Click here to post a comment for AMD Ryzen 14nm Wafer Yields Pass 80% - Threadripper CPUs on track on our message forum
https://forums.guru3d.com/data/avatars/m/270/270008.jpg
Killing me I'm trying so hard not to buy / build a new PC all this news is killing me this year. I guess we should see what Skylake-x has to offer up prior to Threadripper landing. Im so glad every time I went to build an Intel HEDT platform I stopped myself over the last 3 years and just saved the cash.
https://forums.guru3d.com/data/avatars/m/240/240605.jpg
Good for amd and gotta love them internet gifs 🙂
https://forums.guru3d.com/data/avatars/m/268/268848.jpg
So next is over 3000 points in R15? ;-)
https://forums.guru3d.com/data/avatars/m/115/115710.jpg
14 core and 10 core Threadrippers are not going to happen. (3+2)+(3+2) or (4+1)+(4+1) for hypothetical 10 core is not possible. Same logic applies to 14 core - there's 5 core Zeppelin is not possible, nor 7 core. Also both Zeppelins must have same core configuration so all 12 core variants are (3+3)+(3+3). 12 and 16 are possible, 8 might just make sense but it's overlapping a bit with Ryzens. Extra PCI-E lanes and quad channel memory would be the main difference so for some power users it might make sense...
https://forums.guru3d.com/data/avatars/m/243/243702.jpg
14 core and 10 core Threadrippers are not going to happen. (3+2)+(3+2) or (4+1)+(4+1) for hypothetical 10 core is not possible. Same logic applies to 14 core - there's 5 core Zeppelin is not possible, nor 7 core. Also both Zeppelins must have same core configuration so all 12 core variants are (3+3)+(3+3). 12 and 16 are possible, 8 might just make sense but it's overlapping a bit with Ryzens. Extra PCI-E lanes and quad channel memory would be the main difference so for some power users it might make sense...
You missed one important thing. Ryzen cores can be disabled via BIOS in any combination. If BIOS can do it, laser can as well...
https://forums.guru3d.com/data/avatars/m/259/259654.jpg
14 core and 10 core Threadrippers are not going to happen. (3+2)+(3+2) or (4+1)+(4+1) for hypothetical 10 core is not possible. Same logic applies to 14 core - there's 5 core Zeppelin is not possible, nor 7 core. Also both Zeppelins must have same core configuration so all 12 core variants are (3+3)+(3+3). 12 and 16 are possible, 8 might just make sense but it's overlapping a bit with Ryzens. Extra PCI-E lanes and quad channel memory would be the main difference so for some power users it might make sense...
It isn't possible or just not very handy for compilers?
data/avatar/default/avatar28.webp
14 core and 10 core Threadrippers are not going to happen. (3+2)+(3+2) or (4+1)+(4+1) for hypothetical 10 core is not possible. Same logic applies to 14 core - there's 5 core Zeppelin is not possible, nor 7 core. Also both Zeppelins must have same core configuration so all 12 core variants are (3+3)+(3+3). 12 and 16 are possible, 8 might just make sense but it's overlapping a bit with Ryzens. Extra PCI-E lanes and quad channel memory would be the main difference so for some power users it might make sense...
I think they would sell a bunch of 8 core chips with 44 lanes and 128GB RAM support. There are a lot of workstation workloads out there where more RAM, a fully dedicated M.2 slot and more GPUs are far more important than more CPU cores. Even so, they probably won't do it just to avoid confusion. Maybe next year, I guess.
data/avatar/default/avatar04.webp
Why not release single CCX quad core CPUs with proper clock speeds? The problem with that company is that they try to awe people instead of trying to understand the market's needs, and that's why they screw it all up every time. They can't release mainstream Vega because HBM memory supply is quite inadequate for mainstream Vega, so they delay it. They could've used a 384bit wide memory bus for Vega, which would result in 384 GB/s of memory bandwidth when coupled with Samsung's GDDR5 memory. If their delta color compression improvements claims are true, then that all they would've needed to release a competitive on par or faster than the GTX1080. Ryzen blows for gaming because they opted to impress people with cores instead of releasing proper quad core CPUs at high clock speeds, without all that cross-CCX latency that's causing slow L3 cache access and slow thread migration across CCXs.
https://forums.guru3d.com/data/avatars/m/56/56686.jpg
Prices wars who can perform better cheaper! still waiting for this, I been tempted by Ryzen chips, but mutli cores out the ass still mean very little
data/avatar/default/avatar40.webp
14 core and 10 core Threadrippers are not going to happen. (3+2)+(3+2) or (4+1)+(4+1) for hypothetical 10 core is not possible. Same logic applies to 14 core - there's 5 core Zeppelin is not possible, nor 7 core. Also both Zeppelins must have same core configuration so all 12 core variants are (3+3)+(3+3). 12 and 16 are possible, 8 might just make sense but it's overlapping a bit with Ryzens. Extra PCI-E lanes and quad channel memory would be the main difference so for some power users it might make sense...
Seeing as how 14 core and 10 core have been on the map and already named and specced I'm guessing that means that AMD says it's possible lol
https://forums.guru3d.com/data/avatars/m/269/269694.jpg
14 core and 10 core Threadrippers are not going to happen. (3+2)+(3+2) or (4+1)+(4+1) for hypothetical 10 core is not possible. Same logic applies to 14 core - there's 5 core Zeppelin is not possible, nor 7 core. Also both Zeppelins must have same core configuration so all 12 core variants are (3+3)+(3+3).
(3+2)+(3+2) and (4+1)+(4+1) are probably unlikely configurations. But why not (4+4)+(3+3) for the 14 core version, and (3+3)+(2+2) for the 10 core? They're currently producing (4+4) configurations for Ryzen 7, and (3+3) and (2+2) configurations for Ryzen 5. So it doesn't sound like too much of a stretch that they could match some of those together for Threadripper.
https://forums.guru3d.com/data/avatars/m/197/197287.jpg
14 core and 10 core Threadrippers are not going to happen. (3+2)+(3+2) or (4+1)+(4+1) for hypothetical 10 core is not possible. Same logic applies to 14 core - there's 5 core Zeppelin is not possible, nor 7 core. Also both Zeppelins must have same core configuration so all 12 core variants are (3+3)+(3+3). 12 and 16 are possible, 8 might just make sense but it's overlapping a bit with Ryzens. Extra PCI-E lanes and quad channel memory would be the main difference so for some power users it might make sense...
Ohhhh do i love it when people claim, as though it is fact, that something "can't" happen, based off of their supposed super knowledge. It's always great, because when they are proven wrong, they pretend they never said it.
Let me bring up topics that have already been talked about, as well as nonsense, every post i make.
Oh hey look, it's this guy here to chime in some information that has already been stated and discussed to death and replied to repeatedly through the forums, yay!
https://forums.guru3d.com/data/avatars/m/270/270718.jpg
Why not release single CCX quad core CPUs with proper clock speeds? The problem with that company is that they try to awe people instead of trying to understand the market's needs, and that's why they screw it all up every time. Ryzen blows for gaming because they opted to impress people with cores instead of releasing proper quad core CPUs at high clock speeds, without all that cross-CCX latency that's causing slow L3 cache access and slow thread migration across CCXs.
Says you...I converted my main PC from i5 2500 to R7 1700, and the performance is fabulous-- even at the stock clock. Windows 10 is lightning fast on this thing, MAME runs great, Supermodel runs great, Powerdirector performance is incredible. I have no problems holding over 60 framerates on any of my games. I can get a stable 3.9GHz no problem, but I don't see the need so I just let it be stock. All this and a kick ass bundled cooler for $320. As an added bonus I'm guaranteed mobo compatibility with the next gen Zens as well. Screw Intel and their monopolistic practices. I wasn't going to upgrade my 2500 to yet ANOTHER quad core when I could get a kick ass octocore with a great cooler for less money-- and forget a hex or octocore Intel for double the price. AMD just raised the bar for PC that Intel was holding over our collective throats.
https://forums.guru3d.com/data/avatars/m/229/229509.jpg
Is threadripper a complete new die with 4 CCXs or 2 of the 8-core dies slaved together?
https://forums.guru3d.com/data/avatars/m/243/243702.jpg
Is threadripper a complete new die with 4 CCXs or 2 of the 8-core dies slaved together?
Considering number of PCIe lanes, best guess is that it has 2 new (different from regular Ryzen 7) 8-core dies connected. Image shows 4, but that may be 32 core version or something else entirely.
https://forums.guru3d.com/data/avatars/m/128/128096.jpg
Why not release single CCX quad core CPUs with proper clock speeds? The problem with that company is that they try to awe people instead of trying to understand the market's needs, and that's why they screw it all up every time. They can't release mainstream Vega because HBM memory supply is quite inadequate for mainstream Vega, so they delay it. They could've used a 384bit wide memory bus for Vega, which would result in 384 GB/s of memory bandwidth when coupled with Samsung's GDDR5 memory. If their delta color compression improvements claims are true, then that all they would've needed to release a competitive on par or faster than the GTX1080. Ryzen blows for gaming because they opted to impress people with cores instead of releasing proper quad core CPUs at high clock speeds, without all that cross-CCX latency that's causing slow L3 cache access and slow thread migration across CCXs.
Wow, you have no idea what you're talking about... For the first time in many years, I'm actually agreeing with AMDs approach for the most part and the fact that the may be a chance to return them to (slight) profitability.
https://forums.guru3d.com/data/avatars/m/115/115710.jpg
Ohhhh do i love it when people claim, as though it is fact, that something "can't" happen, based off of their supposed super knowledge. It's always great, because when they are proven wrong, they pretend they never said it. Oh hey look, it's this guy here to chime in some information that has already been stated and discussed to death and replied to repeatedly through the forums, yay!
That's according to The Stilt who definitely knows more than any of us here.
You missed one important thing. Ryzen cores can be disabled via BIOS in any combination. If BIOS can do it, laser can as well...
Any combination? No...
It isn't possible or just not very handy for compilers?
But why not (4+4)+(3+3) for the 14 core version, and (3+3)+(2+2) for the 10 core?
That's because each Zeppeling die must have matching core & cache configuration. According to The Stilt. Makes sense, since 5 and 7 core options aren't possible for Zeppelin (either one CCX is off completely or they are matching). Also in case it isn't obvious, Threadripper is two Zeppelin dies. EPYC is four.
https://forums.guru3d.com/data/avatars/m/268/268248.jpg
so far i had the idea that amd produces a wafer with ccx's and then stitching 2 together to make the 8 cores and then the less likely to be healthy stitching em up to make 4 and 6 core although that is my own speculation so for 12 core and 10 i thought they will use 3 ccx ...on the other hand i thought they will use just 1 ccx for quad cores and i was wrong so there is that ....now about clock speeds the way they stitch em together seems like once again the overclock will be again 3.9-4.0ghz as far as you have the cooling capacity. anyway we will see when those babies come out 🙂 with 80% yields it sounds to me like the ccxs that can not run with full features might be less than the demand ...so they might be laser-cutting healthy ccxs to meet the demand of the lower models at least that is what my logic says
https://forums.guru3d.com/data/avatars/m/197/197287.jpg
That's according to The Stilt who definitely knows more than any of us here.
And is according to someone not part of AMD. So no, they don't know what they are talking about. Now, if someone who actually helped develop Ryzen were to state this, sure, no reason not to believe it. But that's not what is happening, now is it... No point is saying "that can't happen" when you have absolutely no idea what you are talking about and should rather wait and see. Before ryzen was released, people were saying 6 core processors wouldn't happen either, because it'd be 4, 8, 12, 16, etc. The earth was flat, you couldn't go faster then the speed of sound and only nine planets exist. All facts that people decided to state because they either did not know what they were talking about, could not do it or were simply blind. But hey lets continue this tradition of saying what is or isn't possible about stuff we know thing about, that's cool too.