AMD Threadripper 3990X Scores Another Win: We Test New SPECWorkstation 3 Update
#1
Information 
Quote:
[Image: r8ZwPTg5RCoyWyocG5twjJ-970-80.jpg]

Calling all speedfreaks

The SPECwpc subcommittee has announced a new version of the SPECworkstation 3 benchmark for workstations to accommodate processors with more than 64 threads. That move certainly benefits the only workstation-class processor with more than 64 threads in a single socket: AMD's monstrous 64-core 128-thread Threadripper 3990X. We decided to put the new release to the test to see the full threaded might of the 3990X in action, and the results are impressive.

Upsetting the semiconductor industry is hard, particularly when you're fighting an entrenched and much-larger rival, and sometimes things get broken when you're redefining an industry. In AMD's case, those broken things consist of operating systems and applications that weren't tuned to extract the full performance of its fledgling first-gen Zen architecture, let alone the core-heavy designs of Zen 2. The 64-core 128-thread Threadripper 3990X is the best example because it more than doubles the number of cores available with Intel's halo workstation parts and sets an entirely new bar for single-socket systems.

The 3990X exposes just how unprepared applications and software are for a 128-thread beast: Windows 10 splits cores up into 'processor groups' of 64 threads apiece, so some applications and benchmarks that aren't tuned to span across the groups don't benefit from the increased thread count.

Aside from the obvious performance loss in unoptimized applications, there are other reasons this is important: The problem has the knock-on effect of impacting industry benchmarks and applications that are used to quantify performance for OEMs, ODMs, and end customers. These benchmark results, like the ones developed and maintained by the non-profit Standard Performance Evaluation Corporation (SPEC), are submitted by companies that buy the benchmark ($5,000 for the benchmark in question), and then submit the results to SPEC. The committee charges a $500 to $1,000 fee to verify each result and then publish it in its database.

These benchmarks are verified and maintained by an official group comprised of industry leaders, so the test results are used to guide the development dollars and purchasing decisions for OxMs, like Dell, HP, and Lenovo (among others), and corporate customers alike. As a result, positioning a product in the best and most accurate light possible is critical for sales to professional organizations that comprise the target market for the Threadripper 3990X.

We test with the SPECworkstation 3 suite for our own tests of workstation-class processors (but we don't submit them to the official body), and noted in our recent review that many of the SPEC benchmarks we typically use didn't scale correctly across processor groups. In fact, most of the benchmarks only operated at half of the potential performance, and some results were unusable. That could present an issue for AMD with some professional customers because the benchmarks aren't representative of the 3990X's true performance and potential.

However, the alterations to the existing SPECworkstation 3 benchmark allow the benchmarks to scale correctly across different processor groups (and multi-socket systems) as of the new version V3.0.4, which has an enhanced version of the multi-threading code. Results from the update are comparable to older 3.0 versions of the benchmark, which is important to ensure the new tests can be published to the existing database. It's also great because we can plug-and-play the new benchmark for some direct comparisons, while also providing updated performance results.

The SPECworkstation 3 update is available as either a new complete download or as a patch, and is available today.

So we tested today. After debugging why we couldn't extract the utmost performance of the 3990X in our first round of SPECworkstation 3 tests, it's truly a wonderful sight to unleash the beast and finally see all 128 of those threads utilized in the subtests that can now leverage them. Let's see what that looks like.
...
Continue Reading
[-] The following 1 user says Thank You to harlan4096 for this post:
  • silversurfer
Reply


Forum Jump:


Users browsing this thread: 1 Guest(s)
[-]
Welcome
You have to register before you can post on our site.

Username/Email:


Password:





[-]
Recent Posts
QOwnNotes
26.4.0  LanguageT...Kool — 08:39
Surfshark VPN : Award-winning VPN servi...
Surfshark Apps Ver...jasonX — 04:41
Surfshark VPN : Award-winning VPN servi...
Surfshark launches...jasonX — 03:43
ESET 19.1.12.0
Changes in 19.1.12...harlan4096 — 14:49
Vivaldi 7.9 Build 3970.47
Vivaldi 7.9 Build ...harlan4096 — 07:31

[-]
Birthdays
Today's Birthdays
avatar (41)alapesihy
Upcoming Birthdays
avatar (45)wapedDow
avatar (49)oapedDow
avatar (42)Sanchowogy
avatar (46)MeighGoask
avatar (47)creatralGuelm
avatar (38)procnipsut
avatar (44)accenwibly
avatar (41)ahyvily
avatar (38)urumahiz
avatar (44)techlignub
avatar (43)Stevenmam
avatar (50)onlinbah
avatar (50)fuspeukChark
avatar (44)werriewWaiNg
avatar (38)Freemanleo
avatar (48)cticigges
avatar (50)ecoFit
avatar (44)soccejeS
avatar (43)cdoubapKit
avatar (38)lystraPonia
avatar (31)smith8395john
avatar (51)steakelask
avatar (45)Termoplenka
avatar (43)bycoPaist
avatar (49)pieloKat
avatar (43)ilyagNeexy
avatar (51)donitascene
avatar (51)burntLaw
avatar (41)MrDoorsskibheeds
avatar (51)Toligo
avatar (46)Rodneykak
avatar (49)tradeSmode
avatar (39)vemedProkbior
avatar (38)RobertUtelt
avatar (46)JamesZic
avatar (43)Sanfordbup
avatar (38)Der.Reisende
avatar (36)Kiran78

[-]
Online Staff
There are no staff members currently online.

>