[中央處理器] ThreadRipper REVIEW !

samhcl

https://www.overclock3d.net/reviews/cpu_mainboard/asus_x399_rog_zenith_extreme_and_ryzen_1950x_threadripper_review/2

"The Ryzen 1950X boosts to 3.7 GHz on all cores by default but thanks to the excellent overclocking capabilities of the ASUS Zenith Extreme we were able to take all 16 cores up to 4.2 GHz. "

It said stable easily at 4Ghz (all core of cause.)




This time Asus doning great job about VRM temp.(The small fan is USEFUL )
"you can see here the ASUS VRM cooling solution absolutely gets the job done with a maximum of only 57°C, way below many of the recent VRM temperatures we've monitored."




Power draw:  379W stock (3.7) 691W OC (4.2G)




Performance  better then same price 7900X (multithread base), even stock clock already winning OC 7900x

samhcl

Single thread... AMD still lost

BIOV1

16C 超到4.2GHZ 算誇張

samhcl

16C 超到4.2GHZ 算誇張
BIOV1 發表於 2017-8-10 23:10

係，但有代價: 接近700W .

但mother broad VRM 反而不太熱，華碩有料到!

patricktang

有無長行4.2~?

abcabcabc1

3.7GHZ  OC去 4.2GHZ 要691W差唔多要多80%電，都唔少啊

samhcl

係，但有代價: 接近700W .

但mother broad VRM 反而不太熱，華碩有料到!
samhcl 發表於 2017-8-10 23:18

In the preview, the OC test seems be in that speed.

MirageKnight

無1000W牛唔好諗OC佢.

Stiggy930

睇咗Anandtech個review。贏可以贏半條街，輸可以輸半條街(尢其係行緊Non NUMA mode而個test係latency sensitive)。AMD marketing係game mode同creator mode 個解釋可以做好D。Software一日未能夠完全進入multi-thread同NUMA aware嘅時代對於AMD嘅infinite fabric 係一個負擔。Thread Ripper die2die bandwidth 高EPYC太多(ThreadRipper 1 to 1 connection between dies/EPYC 1 to 3 connection between dies)，workstation/low end server market可能會做成鬼打鬼局面。

samhcl

Gaming either, mostly not in very good result (some newer / multi core based engine can get better result)

Video encode also not as good as Intel CPU, no doubt that is the software optimize issue mostly.  (or intel's AVX instruction helps a lot too)