Thursday, August 10th 2017

AMD Ryzen Threadripper "Summit Ridge" Dies are Heavily Binned

AMD revealed that the pair of 8-core "Summit Ridge" dies that make up the Ryzen Threadripper multi-chip module are heavily binned. AMD hand-selects the top-5% highest performing "Summit Ridge" dies for Ryzen Threadripper manufacturing, which makes these chips of a higher grade than even what AMD sets aside for Ryzen 7-series socket AM4 chips.

AMD requires the highest grade "Summit Ridge" dies to use in Threadripper chips, to keep electrical leakage to the minimum, so the chips can run as cool as possible, with the least power-draw. Choosing the best dies could also ensure that Threadripper chips have the highest overclocking-headroom taking into account other electrical and thermal constraints. A 7-series chip such as the 1800X could still achieve higher clocks than a Threadripper chip, in that sense.
Add your own comment

60 Comments on AMD Ryzen Threadripper "Summit Ridge" Dies are Heavily Binned

#1
drade
"which makes them chips."

I think you mean "which makes their chips?"
Posted on Reply
#2
infrared
We've been expecting this from the beginning, makes sense. It's good to have it confirmed by AMD.
Posted on Reply
#3
chaosmassive
how many hours left till NDA for benchmark result end?
Posted on Reply
#4
Dbiggs9
Ryzen chips cost is .44/$
Posted on Reply
#6
XiGMAKiD
AMD and chips in the same sentence remind me of this
Posted on Reply
#7
qubit
Overclocked quantum bit
I can't wait for some enthusiast to unlock the hidden 16 cores and unleash a monster! Intel's reaction to that will be very interesting to see.

I'm also pretty much on tenterhooks for the reviews. We've been waiting for this kind of competition for more than a decade now.
Posted on Reply
#9
justimber
reviews now trickling in...seems that initial reviews are confirming the price/performance crown for AMD...and let's see what Intel will offer to ebb the AMD TR wave.

mighty time to build a new HEDT rig :D
Posted on Reply
#10
Slizzo
qubit said:
I can't wait for some enthusiast to unlock the hidden 16 cores and unleash a monster! Intel's reaction to that will be very interesting to see.

I'm also pretty much on tenterhooks for the reviews. We've been waiting for this kind of competition for more than a decade now.
Hidden 16 cores? Two dies are blank silicon, two are actual dies with cores.
Posted on Reply
#11
qubit
Overclocked quantum bit
Slizzo said:
Hidden 16 cores? Two dies are blank silicon, two are actual dies with cores.
How do you know they're blank silicon and not disabled cores?
Posted on Reply
#12
notb
qubit said:
I can't wait for some enthusiast to unlock the hidden 16 cores and unleash a monster! Intel's reaction to that will be very interesting to see.
There will be no "unlocking". And even if it happened, Intel wouldn't have to react.
I'm also pretty much on tenterhooks for the reviews. We've been waiting for this kind of competition for more than a decade now.
You can't be serious.
We're still talking about a very tiny niche.
The reason why Intel seems to have given up on X299 is... they don't care. AMD can have the "PC Enthusiast" segment.

Now... the workstation segment is something entirely different. It will be interesting to see how Intel reacts with workstation-oriented Xeons.
BTW: does Threadripper have security features from Ryzen PRO? Because if not, ECC might not be enough to make it successful in business.
Posted on Reply
#13
jabbadap
qubit said:
How do you know they're blank silicon and not disabled cores?
Even if they are real cores, you can't use them with x399. Platform lacks memory channels needed for them(Every ryzen chip inside threadripper needs two memory channels).
Posted on Reply
#14
cdawall
where the hell are my stars
infrared said:
We've been expecting this from the beginning, makes sense. It's good to have it confirmed by AMD.
Yea I called this day one Lol. Mainstream always gets the lowest bins...
Posted on Reply
#15
R0H1T
justimber said:
reviews now trickling in...seems that initial reviews are confirming the price/performance crown for AMD...and let's see what Intel will offer to ebb the AMD TR wave.

mighty time to build a new HEDT rig :D
perf/$ as well as perf/W in most cases.
cdawall said:
Yea I called this day one Lol. Mainstream always gets the lowest bins...
I wouldn;t call the 1700 (non X) a low binned part.
Posted on Reply
#16
qubit
Overclocked quantum bit
notb said:
There will be no "unlocking". And even if it happened, Intel wouldn't have to react.

You can't be serious.
We're still talking about a very tiny niche.
The reason why Intel seems to have given up on X299 is... they don't care. AMD can have the "PC Enthusiast" segment.

Now... the workstation segment is something entirely different. It will be interesting to see how Intel reacts with workstation-oriented Xeons.
BTW: does Threadripper have security features from Ryzen PRO? Because if not, ECC might not be enough to make it successful in business.
Yes, I can be serious and I really don't like your tone towards me. Always unpleasant so how about you lighten up a bit eh? No need for unpleasantness regardless of who's technical point is right or wrong.

I think Intel will care about the gaming market, even a niche one as it's all about prestige and image, which helps to bring in the sales. Intel won't invest big money on bringing out a new platform they "don't care about", that's just silly. More likely they're not sure how to react to the Threadripper threat which is why we have these anomalies with Skylake-X and the next generation will be much better.

I've since read the review below linked to by Fluffmeister above which shows AMD caning Intel in productivity performance, so Intel very much has something to worry about there. This appears to be an official review too, so likely the others will be similar. Personally, I just wish that Ryzen's IPC was better than Intel's, which would then deliver an all-round killer product. The way things are going, that might actually happen one day.

How can you say for certain that AMD won't allow their CPUs to be unlocked? On what authority are you basing that on, or is that just your opinion? I'm saying maybe/hopefully and that's the best we can do at the moment.

https://www.techspot.com/review/1465-amd-ryzen-threadripper-1950x-1920x

jabbadap said:
Even if they are real cores, you can't use them with x399. Platform lacks memory channels needed for them(Every ryzen chip inside threadripper needs two memory channels).
Ok that's a fair point. So EPYC must have twice the memory channels then?
Posted on Reply
#17
cdawall
where the hell are my stars
qubit said:
Ok that's a fair point. So EPYC must have twice the memory channels then?
Yes it is 8 channels.
Posted on Reply
#19
jabbadap
qubit said:

Ok that's a fair point. So EPYC must have twice the memory channels then?
Like cdawall said epyc has 8 memory channels(even on 8-core epyc). Well I said wrong about it needing two for every chip, it can work single channel too. But x399 platform's has no physical routing for disabled memory channels. So it would need epyc motherboard for getting those work if they are real disabled cores and you can somehow activate them.
Posted on Reply
#20
Frick
Fishfaced Nincompoop
qubit said:

I'm also pretty much on tenterhooks for the reviews. We've been waiting for this kind of competition for more than a decade now.
Massively faster than 7900X in most multithreaded apps (depending on the program), about Ryzen speeds in games (except Civ6 which doesn't like many cores at all, so AMD has a feature to turn some of the cores off, in software, which is neat) and single threaded stuff. So as expected.
Posted on Reply
#21
cdawall
where the hell are my stars
jabbadap said:
Like cdawall said epyc has 8 memory channels(even on 8-core epyc). Well I said wrong about it needing two for every chip, it can work single channel too. But x399 platform's has no physical routing for disabled memory channels. So it would need epyc motherboard for getting those work if they are real disabled cores and you can somehow activate them.
The only way threadripper is getting 32 cores on the current board is for amd to release 16 core dies. Until then they are at their limit on this socket. It was never wired for 4 dies.
Posted on Reply
#22
jabbadap
cdawall said:
The only way threadripper is getting 32 cores on the current board is for amd to release 16 core dies. Until then they are at their limit on this socket. It was never wired for 4 dies.
Yeah that was what I said. Epyc server motherboards have physically same socket(socket sp3 has same layout as socket TR4), but it's not compatible with TR -processors(maybe because disabled cores/or software limitation).
Posted on Reply
#23
uuuaaaaaa
jabbadap said:
Like cdawall said epyc has 8 memory channels(even on 8-core epyc). Well I said wrong about it needing two for every chip, it can work single channel too. But x399 platform's has no physical routing for disabled memory channels. So it would need epyc motherboard for getting those work if they are real disabled cores and you can somehow activate them.
Those 2 dies are not disabled, they are blank pieces of silicon, just spacers, amd themselves confirmed this. According to the die layout it wouldn't make much sense being failed epyc cpus.
Posted on Reply
#24
jabbadap
uuuaaaaaa said:
Those 2 dies are not disabled, they are blank pieces of silicon, just spacers, amd themselves confirmed this. According to the die layout it wouldn't make much sense being failed epyc cpus.
Yeah of course, that should be common knowledge by now. I think that processor Der8auer delidded might have been ES sample made from 32 core epyc processor, thus all the theories of unlocking cores. But yeah how to recognize blank piece of silicon by only eyes from real deal(Well microscope, or enough MPix on camera to get die shot should do the trick).
Posted on Reply
#25
cdawall
where the hell are my stars
jabbadap said:
Yeah of course, that should be common knowledge by now. I think that processor Der8auer delidded might have been ES sample made from 32 core epyc processor, thus all the theories of unlocking cores. But yeah how to recognize blank piece of silicon by only eyes from real deal(Well microscope, or enough MPix on camera to get die shot should do the trick).
Unless you can microsolder all the missing transistors in place you weren't unlocking anything with his cpu.
Posted on Reply
Add your own comment