AMD Ryzen 5 3600X review
Promo: URCDKey Summer Sale Windows 10 Pro for $11
Team Group T-Force 4133 MHz 16GB DDR4 review
ASUS ROG Crosshair VIII HERO (Wi-fi) review
MSI MEG X570 Godlike review
Corsair NightSword RGB gaming mouse review
Corsair MP600 PCIe 4.0 NVMe SSD review
Palit GeForce RTX 2070 SUPER JetStream review
MSI GeForce RTX 2070 SUPER Gaming X TRIO review
Palit GeForce RTX 2060 SUPER JetStream review
AMD Zen 2: Production yields for Ryzen 3000 Dies at 70 percent
Whenever you fab a wafer with CPU/GPU dies (or whatever they are fabbing), there are always defects. So if a wafer has 200 CPUs dies and 50 are functional, your yield is 25%. Well, the yield of the Zen 2-cores should be good, roughly 70 percent of the chips are fully functional.
When AMD started original ZEN at 14nm, that was number was 80%, so they are not that far off in this early stage of fabrication. in a report from bitsandchips that number is 70 percent, which is not perfect, but good. The site relies on "unspecified sources" though, but has been proven reliable in the past.
The small die size of the Zen 2-core is estimated to be 88 mm² and seems to be the helping factor opposed to designing big monolithic chips. For example, Intel's XCC-Die with 28 cores, according to chatter, for a fully functional chip shows only 35 percent in yields. If you keep that thought in mind, the upcoming Threadripper also would yield at 70% as it uses the same CPU dies added together in a chiplet design. So that's twice the yield of Intel.
Unfortunately, there's no official word on clock frequencies and prices just yet, we can't wait to hear the official announcements, which are expected at the AMD keynote starting at Computex late in May.
AMD Zen 2 "Rome" 64-core processor Photo up close - 11/07/2018 09:29 AM
So if you have been following yesterdays' announcement, you will have noticed that AMD has talked about Zen2 and actually announced a new processor based on it. Last week Intel announced a 48-core pr...
FMA4 instruction set hidden, but is working on AMD Zen processors - 10/15/2018 11:06 AM
In an interesting find, it has been discovered that AMD processors based on ZEN architecture actually support the latest iteration of FMA, the FMA4-instruction set. The theory is that the FMA3 supplem...
AMD Zen 2 nm architecture starts with EPYC CPUs - Talks 5nm also - 07/27/2018 07:50 AM
The second quarter results for AMD ended strong, during the Q2 presentations the company also talked about upcoming products a bit. That includes the first 7nm processors based on the Zen 2 archite...
Confirmed: AMD Zen designer Jim Keller joins Intel to lead silicon engineering department - 04/27/2018 09:27 AM
Yesterday we posted that CPU architecture designer Jim Keller might make a move to Intel. Word actually is that he was fired at Tesla. Intel decided to not delay any announcements and issued a press ...
AMD Zen Designer Keller Is Moving to intel - 04/26/2018 09:22 AM
Jim Keller, ex AMD designer and Chief Architect of the ZEN Architecture left AMD roughly a year ago to join Tesla. Now, it seems he is leaving Tesla where he worked on AI, and has reached his final...
Clawedge
Senior Member
Posts: 2400
Joined: 2006-09-02
Senior Member
Posts: 2400
Joined: 2006-09-02
#5662070 Posted on: 04/22/2019 10:30 AM
My body is ready!!!
And here is Intel
My body is ready!!!
And here is Intel
Clae
Junior Member
Posts: 1
Joined: 2019-04-22
Junior Member
Posts: 1
Joined: 2019-04-22
#5662109 Posted on: 04/22/2019 01:51 PM
Just wanted to write that if threadripper uses 2 Zen2 dies, then wouldn't the yield be 0.7*0.7? Still better than Intel.
Just wanted to write that if threadripper uses 2 Zen2 dies, then wouldn't the yield be 0.7*0.7? Still better than Intel.
djmcave
Junior Member
Posts: 7
Joined: 2016-08-10
Junior Member
Posts: 7
Joined: 2016-08-10
#5662128 Posted on: 04/22/2019 03:03 PM
I'm assuming they can test them out before assembling the dies together.. so you would only assemble healthy dies, so only the 0.7 yield matters.
Just wanted to write that if threadripper uses 2 Zen2 dies, then wouldn't the yield be 0.7*0.7? Still better than Intel.
I'm assuming they can test them out before assembling the dies together.. so you would only assemble healthy dies, so only the 0.7 yield matters.
Kool64
Senior Member
Posts: 207
Joined: 2006-10-21
Senior Member
Posts: 207
Joined: 2006-10-21
#5662129 Posted on: 04/22/2019 03:04 PM
It will be like Christmas in June/July when these chips are released. I can't wait.
It will be like Christmas in June/July when these chips are released. I can't wait.
FrostNixon
Senior Member
Posts: 181
Joined: 2016-04-15
Senior Member
Posts: 181
Joined: 2016-04-15
#5662132 Posted on: 04/22/2019 03:09 PM
Those CPUs can't come soon enough, the R5 2400g is not made for what I do to it...
Those CPUs can't come soon enough, the R5 2400g is not made for what I do to it...
icedman
Senior Member
Posts: 886
Joined: 2013-02-22
Senior Member
Posts: 886
Joined: 2013-02-22
#5662139 Posted on: 04/22/2019 03:39 PM
Even at 70% wouldnt they still yield more good chips per wafer as theoretically on 7nm you are already making more chips on each wafer so 70% should still be a great number
Even at 70% wouldnt they still yield more good chips per wafer as theoretically on 7nm you are already making more chips on each wafer so 70% should still be a great number
Undying
Senior Member
Posts: 11240
Joined: 2008-08-28
Senior Member
Posts: 11240
Joined: 2008-08-28
#5662142 Posted on: 04/22/2019 03:44 PM
Still couple months away i guess. Have some time save up for an upgrade.
Still couple months away i guess. Have some time save up for an upgrade.
Fox2232
Senior Member
Posts: 9380
Joined: 2012-07-20
Senior Member
Posts: 9380
Joined: 2012-07-20
#5662146 Posted on: 04/22/2019 04:01 PM
Those 70% are fully functional. (As of no locking of CPU cores needed or plain dead die.)
=> fully working 8C/16T
Even at 70% wouldnt they still yield more good chips per wafer as theoretically on 7nm you are already making more chips on each wafer so 70% should still be a great number
Those 70% are fully functional. (As of no locking of CPU cores needed or plain dead die.)
=> fully working 8C/16T
JamesSneed
Senior Member
Posts: 546
Joined: 2017-02-14
Senior Member
Posts: 546
Joined: 2017-02-14
#5662162 Posted on: 04/22/2019 05:24 PM
Those 70% are fully functional. (As of no locking of CPU cores needed or plain dead die.)
=> fully working 8C/16T
Exactly which we can also assume they have much better than 70% yields for 4-6 core chiplets. If true this is a good sign for early 7nm production.
Those 70% are fully functional. (As of no locking of CPU cores needed or plain dead die.)
=> fully working 8C/16T
Exactly which we can also assume they have much better than 70% yields for 4-6 core chiplets. If true this is a good sign for early 7nm production.
schmidtbag
Senior Member
Posts: 4132
Joined: 2012-11-10
Senior Member
Posts: 4132
Joined: 2012-11-10
#5662167 Posted on: 04/22/2019 05:33 PM
In terms of total quantity, yes, there should be more functional dies per 7nm wafer. Not only is this because the transistors are roughly half the size, but you have to consider the wafers are circular. When you decrease the area of each die, you can use up more of the entire wafer. For example:
Even at 70% wouldnt they still yield more good chips per wafer as theoretically on 7nm you are already making more chips on each wafer so 70% should still be a great number
In terms of total quantity, yes, there should be more functional dies per 7nm wafer. Not only is this because the transistors are roughly half the size, but you have to consider the wafers are circular. When you decrease the area of each die, you can use up more of the entire wafer. For example:
chispy
Senior Member
Posts: 8759
Joined: 2006-10-29
Senior Member
Posts: 8759
Joined: 2006-10-29
#5662172 Posted on: 04/22/2019 05:47 PM
Looking good , cannot wait to get my hands on this new cpus
Looking good , cannot wait to get my hands on this new cpus
Aura89
Senior Member
Posts: 7484
Joined: 2008-07-31
Senior Member
Posts: 7484
Joined: 2008-07-31
#5662182 Posted on: 04/22/2019 05:58 PM
Depends on the size of the actual chip, which, i'll be honest, is a little confusing as of yet
7nm itself doesn't make anything better if they increase the size of the die
What is confusing is the fact that this article states a core for zen 2 being estimated at 88 mm²
This is confusing because i hope that would be the module (Zeppelin?), not core, as the core of Zen core is 7 mm²
SOOO...i'm going to assume they mean the module/zeppelin, in which case, it appears it'll be quite a bit smaller, since current Zen ones are 212.97 mm², or 190, some conflicting information out there
Even at 70% wouldnt they still yield more good chips per wafer as theoretically on 7nm you are already making more chips on each wafer so 70% should still be a great number
Depends on the size of the actual chip, which, i'll be honest, is a little confusing as of yet
7nm itself doesn't make anything better if they increase the size of the die
What is confusing is the fact that this article states a core for zen 2 being estimated at 88 mm²
This is confusing because i hope that would be the module (Zeppelin?), not core, as the core of Zen core is 7 mm²
SOOO...i'm going to assume they mean the module/zeppelin, in which case, it appears it'll be quite a bit smaller, since current Zen ones are 212.97 mm², or 190, some conflicting information out there
schmidtbag
Senior Member
Posts: 4132
Joined: 2012-11-10
Senior Member
Posts: 4132
Joined: 2012-11-10
#5662196 Posted on: 04/22/2019 07:01 PM
Depends on the size of the actual chip, which, i'll be honest, is a little confusing as of yet
7nm itself doesn't make anything better if they increase the size of the die
I've considered pointing this out myself, but I'm sure the overall core-per-core architecture size is roughly the same as last gen parts (if they were the same transistor size). There's likely some new features here and there, while maybe some might be offloaded onto the central hub for the chiplet design. Either way, we're going from 14nm or 7nm, so even if each die were significantly larger (like by 25%, for example), the transistors are ostensibly half the size, so you're still going to fit more dies per wafer. As for whether that means the 10% decrease in yield will make up for the additional dies, I'm not sure and I don't care to do the math to figure that out haha.
Depends on the size of the actual chip, which, i'll be honest, is a little confusing as of yet
7nm itself doesn't make anything better if they increase the size of the die
I've considered pointing this out myself, but I'm sure the overall core-per-core architecture size is roughly the same as last gen parts (if they were the same transistor size). There's likely some new features here and there, while maybe some might be offloaded onto the central hub for the chiplet design. Either way, we're going from 14nm or 7nm, so even if each die were significantly larger (like by 25%, for example), the transistors are ostensibly half the size, so you're still going to fit more dies per wafer. As for whether that means the 10% decrease in yield will make up for the additional dies, I'm not sure and I don't care to do the math to figure that out haha.
marian4q
Junior Member
Posts: 14
Joined: 2015-03-25
Junior Member
Posts: 14
Joined: 2015-03-25
#5662218 Posted on: 04/22/2019 08:35 PM
I remember AdoredTV analisis on yields and die size and how this would enable AMD be ahead in that area. AS I thinkk AMD small resource, this IS the way to go, the only way and the smartest.
I remember AdoredTV analisis on yields and die size and how this would enable AMD be ahead in that area. AS I thinkk AMD small resource, this IS the way to go, the only way and the smartest.
Click here to post a comment for this news story on the message forum.
Senior Member
Posts: 9380
Joined: 2012-07-20
"roughly 70 percent of the chips are fully functional" ...
This means that from remaining 30% with some form of defect, some are salvaged to lower core count products and some are dead. Not bad at all for price competitiveness.