Zen的新八卦，好消息壞消息都有，信不信隨人
http://carbonite.co.za/f20/amd-zen-141214/
Apologies if there's another thread about AMD's upcoming CPU architecture and
of course desktop CPUs, if there is, delete or perhaps merge this info.
This is the current state of the retail CPUs, which have been improving by
the month.
There are some errata issues present in the current testing samples, similar
in a way to the TLB bug of the Phenom. The workaround right now is done via
the BIOS. The workaround however, strips around 30 ~ 40% of the CPU
performance.
Zen的測試版有類似當初飛龍的TLB Bug，AMD已經有解決方案。但目前解決的方案會讓Zen
的性能表現下降30%~40%
The CPUs are well behind schedule and every day there's real progress and bug
fixing being done. Unlike with INTEL's E0 CPUs which make it to the wild that
are almost completely final silicon. AMD's samples will continue to get bug
fixes right up until retail spec sampling to partners.
Zen目前進度落後中
In August Clock speeds were 3.8GHz, right now 4.2GHz overclocking is
possible, with LN2 5GHz is doable. Again this will change of course, but it
is just the current silicon that is behaving like this.
8月時Zen只能上3.8Ghz 目前Zen可以超上4.2Ghz了。如果用液態氮應該能上5Ghz，但要是
體質好的晶片
AM4/ZEN uses an SOC design, that means even CMOS/BIOS configuration is on
package (not necessarily on silicon, I can't confirm this) so it is possible
to clear the "BIOS" and still have old value applied 30 minutes later. How
this will be addressed remains to be seen. Perhaps it won't be the same
scenario for final silicon
Operating voltages (nominal) are 1.3v and all the way up to 1.5v should be
fine it seems for AIO cooling. Frequency scaling isn't a strong point but
again that may have everything to do with the process at this point rather
than an inherent design limitation.
Performance is particularly strong at this point vs. INTEL's latest
offerings. Single thread performance is matching Haswell-E and of course
multi-threading performance as well. Tests that are memory bandwidth
dependent may go to the INTEL platform simply as a result of having more
memory channels, but I can't confirm that right now and have no info on that.
The important thing here is that the 16Thread/8-Core CPU is minimum 5960X
performance if not better actually. (Based on Cinebench R15) with the error
fix disabled.
Zen單核表現和多核表現都和 Haswwll-E差不多 8c16t表現不輸給5960X (Cinebench
R15) 前提是bug要除完
Can't speak to how well the IMC is working as current samples are locked to
low DRAM frequencies (2133MHz and lower) and of course this has an impact on
performance.
As stated in the beginning, every week is progress and AMD is working at an
unprecedented rate to get these ready by March.
3月.....
You're unlikely to see any high end boards for the CPUs prior to launch or at
launch, simply because no vendors can commit to too much right now as plenty
is changing at a rapid rate.
--
重返榮耀？農藥？
--
http://carbonite.co.za/f20/amd-zen-141214/
Apologies if there's another thread about AMD's upcoming CPU architecture and
of course desktop CPUs, if there is, delete or perhaps merge this info.
This is the current state of the retail CPUs, which have been improving by
the month.
There are some errata issues present in the current testing samples, similar
in a way to the TLB bug of the Phenom. The workaround right now is done via
the BIOS. The workaround however, strips around 30 ~ 40% of the CPU
performance.
Zen的測試版有類似當初飛龍的TLB Bug，AMD已經有解決方案。但目前解決的方案會讓Zen
的性能表現下降30%~40%
The CPUs are well behind schedule and every day there's real progress and bug
fixing being done. Unlike with INTEL's E0 CPUs which make it to the wild that
are almost completely final silicon. AMD's samples will continue to get bug
fixes right up until retail spec sampling to partners.
Zen目前進度落後中
In August Clock speeds were 3.8GHz, right now 4.2GHz overclocking is
possible, with LN2 5GHz is doable. Again this will change of course, but it
is just the current silicon that is behaving like this.
8月時Zen只能上3.8Ghz 目前Zen可以超上4.2Ghz了。如果用液態氮應該能上5Ghz，但要是
體質好的晶片
AM4/ZEN uses an SOC design, that means even CMOS/BIOS configuration is on
package (not necessarily on silicon, I can't confirm this) so it is possible
to clear the "BIOS" and still have old value applied 30 minutes later. How
this will be addressed remains to be seen. Perhaps it won't be the same
scenario for final silicon
Operating voltages (nominal) are 1.3v and all the way up to 1.5v should be
fine it seems for AIO cooling. Frequency scaling isn't a strong point but
again that may have everything to do with the process at this point rather
than an inherent design limitation.
Performance is particularly strong at this point vs. INTEL's latest
offerings. Single thread performance is matching Haswell-E and of course
multi-threading performance as well. Tests that are memory bandwidth
dependent may go to the INTEL platform simply as a result of having more
memory channels, but I can't confirm that right now and have no info on that.
The important thing here is that the 16Thread/8-Core CPU is minimum 5960X
performance if not better actually. (Based on Cinebench R15) with the error
fix disabled.
Zen單核表現和多核表現都和 Haswwll-E差不多 8c16t表現不輸給5960X (Cinebench
R15) 前提是bug要除完
Can't speak to how well the IMC is working as current samples are locked to
low DRAM frequencies (2133MHz and lower) and of course this has an impact on
performance.
As stated in the beginning, every week is progress and AMD is working at an
unprecedented rate to get these ready by March.
3月.....
You're unlikely to see any high end boards for the CPUs prior to launch or at
launch, simply because no vendors can commit to too much right now as plenty
is changing at a rapid rate.
--
重返榮耀？農藥？
--
All Comments