[News] Flagship AMD Ryzen Threadripper 16-core Chip Appears on GeekBench

[StyM]

https://www.techpowerup.com/234342/f...s-on-geekbench

Apparently, AMD's nomenclature of its flagship Ryzen Threadripper won't look like years-of-birth of today's gamers after all. The flagship 16-core part will bear the model name Ryzen Threadripper 1950X (and not the previously-reported 1998X). This chip was put through GeekBench 4.1.0, on an ASRock X399 Professional Gaming, paired with 16 GB of DDR4-2133 MHz memory. Whether it's dual-channel or quad-channel, is not known at this point. What is known, however, is that 2133 MHz isn't the best memory frequency for Ryzen; and paired with quad-channel DDR4-3200, one could expect the best possible performance. The 1950X was clocked at 3.40 GHz for this test, which probably is its final nominal clock speed, after all.

The Ryzen Threadripper 1950X sample scored 4,167 single-thread performance, and 24,539 points in multi-threaded performance. To put these numbers into perspective, an Intel Xeon E5-2697A v4 16-core/32-thread processor based on the "Broadwell" architecture scores 30,450 points in multi-threaded performance, even if single-thread performance is as low as 3,651 points. Perhaps the memory setup or SMT isn't optimally set for the Threadripper chip. Among the other Threadripper SKUs AMD plans to launch on July 27 are the 12-core/24-thread Ryzen Threadripper 1920X and 1920 (non-X).

[AliG]

Interesting. If I recall correctly, Ryzen scaled slightly better in multi-threaded applications than Broadwell-E did.

Either the Xeons have some extra magic juice in there, or Threadripper still needs optimization.

[Metroid]

or Threadripper still needs optimization.

This and that is about the smt once again.

[FlanK3r]

the multithread is strange, similar score I have with my Ryzen 7 CPU on air

[zanzabar]

the multithread is strange, similar score I have with my Ryzen 7 CPU on air

that is not much faster than my 5930k at stock but all core turbo. geek bench also does not run load on the cpu very hard so i wounder if fabric is a problem on amd with it.

[xlink]

This is either fake or bugged.

1600 @ 3.8GHz with 2999Mhz RAM gets 20,000

There's no way that having nearly 3x the cores(at 90% of the clocks) and QUAD channel memory generates only a 25% improvement over that. If you had perfect scaling, the improvement would be 140%, NOT 25%. I could see 60-100% scaling (infinity fabric issues) but this doesn't feel right.

http://www.overclock.net/t/1632246/w...#post_26159721

[drmrlordx]

Geekbench, pfft.