-However AMD is not tied down by the WSA as much as before, since they e.g. can start producing the chips at Samsung and then later move production to GF.
That seems counter intuitive why would amds shrinking volume not give them better leverage when the whole wsa thing has been about making sure amd had enough manufacturing capacity.I can't really see where all these conclusions of yours are coming from. Here are my two cents:
- Apple indeed got a second-source for their chips, but that second source is Samsung, not Globalfoundries, because their first source is still (and will be for the foreseeable future) TSMC. Globalfoundries got a total of 0 Apple orders allocated to them.
- That Samsung might be manufacturing chips for AMD is much more likely of a settlement between Samsung and Globalfoundries than a sudden relief from the stringent conditions on the WSA. The trend on the WSA is for it to become more and more stringent, not lax: It comprised part of the CPU, then all CPUs, then finally all CPUs and part of the GPU line up.
- Globalfoundries wasn't shy to make AMD swallow a bill to backport a low volume part like Jaguar to their 28nm process.
- Nothing was reported at AMD that tilted the WSA balance towards them, quite the opposite, the shrinking volumes gives even more bargaining power to Globalfoundries.
That seems counter intuitive why would amds shrinking volume not give them better leverage when the whole wsa thing has been about making sure amd had enough manufacturing capacity.
That seems counter intuitive why would amds shrinking volume not give them better leverage when the whole wsa thing has been about making sure amd had enough manufacturing capacity.
We have no idea how this settlement is arranged, we can only speculate. But the main point is that AMD now can have their chips produced at both Samsung and GF. And Samsung already has the process up and running. So in that respect AMD is not tied down by the WSA as much as before. AMD's chip launches will not be delayed by GF potentially now having their 14 nm process ready in time.
Yes the sizes changed but nothing else did, amd never stated the 1mb l2 latency for ex so we only have some very old numbers. But the whole mostly write through thus mostly inclusive cache policy the wcc and the poor latencies never got fixed. Davr kanter on rwt has said that as he understands it the cmt design creates massive layout issues for the l2 and thats a big reason for its poor performance. He also said the actual l2 arrays are very fast, so hopefully we are goign to see a ~10 cycle l2 for Zen.Actually, there were some tweaks to L1 and L2 cache in Carrizo that flew under my radar. The L1 data cache doubled in size without reduction in efficiency while the l2 size was cut in half. The l2 is faster in Carrizo than it is in Kaveri . . .
.
Its doesn't work that way. They use different tools etc. So you cant just transfer it from one to the other. You need chip redesign to move it.
Copy exact refers to the way the manufacturing is set up at the GloFo facility, it has no impact on the process itself, other than how lack of using it makes it harder to replicate good yields.Its not a copy exact setup.
Yes the sizes changed but nothing else did, amd never stated the 1mb l2 latency for ex so we only have some very old numbers. But the whole mostly write through thus mostly inclusive cache policy the wcc and the poor latencies never got fixed.
Davr kanter on rwt has said that as he understands it the cmt design creates massive layout issues for the l2 and thats a big reason for its poor performance. He also said the actual l2 arrays are very fast, so hopefully we are goign to see a ~10 cycle l2 for Zen.
This is incorrect. The process and tools used at GloFo are exactly the same as those at Samsung. If you could transport masks, you could use the same masks at both facilities to produce the same chips. (Since you can´t, you just have to pay extra for duplicating your masks.)
Copy exact refers to the way the manufacturing is set up at the GloFo facility, it has no impact on the process itself, other than how lack of using it makes it harder to replicate good yields.
Industry source a.k.a. rumor - "AMD, Samsung Electronics, and GF had finalized on such decision when Samsung Electronics and GF were signing on contracts on common license for 14-nano processing last year"
http://english.etnews.com/news/article.html?id=20151222200002
If that so... expect Samsung helping AMD also on the HBM2 production.
Hoping thag HBM1 enters as standard too since it might help nVIDIA and AMD a lot.
HBM1 and HBM2 are part of the same standard. (JEDEC Standard no. 235A)
There are parts defined with 1.0Gbps, 1.6Gbps and 2.0Gbs. SKHynix and
Samsung are both producing the 2.0Gbps 8GByte versions (Samsung will
demonstrate a 2.4Gbps version next month with 307.2 GByte/s. at ISSCC)
No: http://isscc.org/doc/2016/ISSCC2016AdvanceProgram.pdfHey Hans, do you know if we will see a Zen briefing at ISSCC 2016?
4.2 Increasing the Performance of a 28nm x86-64 Microprocessor
Through System Power Management
Surely the console wafers by themselves should be enough to fulfil the WSA obligations, and it's not like the silicon has to be on the cutting edge.
Now we have:That would be odd, because it doesn't match shipments of PS4/Xbox One. If anything 2014 was a better year. Also backing up financials. So they had to have an explosion in sales in Q4 to make this possible.
Now it looks like the same number of units sold according to your chart.Sony said:Tokyo, January 5, 2016 Sony Computer Entertainment Inc. (SCE) today announced that the PlayStation®4 (PS4) computer entertainment system has sold through more than 5.7 million units*1 during the 2015 holiday season*2. PS4 has now cumulatively sold through more than 35.9 million units globally as of January 3, 2016.
Now we have:
Now it looks like the same number of units sold according to your chart.
Apple's A10 can not be exclusively produced by TSMC, which slashed its
capex by 30% due to lost opportunities in the mobile sector... To late now.
10nm is the next opportunity, the next race.
We estimate our foundry market segment share of 16, 14-nanometer node increases from about 40% in 2015 to above 70% in 2016 exceeding the previous prediction we made in mid-2014.