Which CAD Workstation Buyers Can Benefit from Threadripper?31 Dec, 2017 By: Alex Herrera
Herrera on Hardware: An analysis of AMD’s Ryzen Threadripper illustrates why a workstation CPU is not a one-size-fits-all proposition.
Ever since AMD disclosed the first hints about its Zen processor architecture, I’ve been left wondering — and writing — about when AMD might exploit it to make a serious push into the market for workstation CPUs again. In a recent column, I made the case that AMD’s window of opportunity is now wide open, supported by not just one but several Zen-derived products that are now well positioned to launch that offensive. Ryzen 7, Ryzen Threadripper, Ryzen Pro, and Epyc are all viable candidates for workstation applications, from entry-level single-socket (1S) models all the way up to max-configured dual-socket (2S) monsters.
Threadripper: More Cores than the Workstation Norm
The last time that AMD mounted a serious challenge to Intel in workstations was in the previous decade, in the form of the Opteron-branded Hammer, a processor that truly disrupted high-performance computing markets for servers and workstations. Launched in 2003, Hammer introduced two key innovations — 64-bit instruction set extensions and direct-attach memory — which allowed Opteron to leapfrog Intel, especially in the dual-socket configurations common in workstations and servers. On the back of Hammer, AMD began taking meaningful share from Intel, peaking in 2006–2008 before fading from a combination of poor execution by AMD and a righting of the ship by Intel.
This time, AMD may follow a similar playbook for its Zen campaign, by focusing on technology and products that OEMs have not emphasized with Intel’s Core and Xeon CPUs. Where Hammer innovated in memory architecture and 64-bit extensions, Zen may leverage its aptitude in high-density, multi-core processors.
Cue Threadripper, the first Zen CPU making noise in high-performance client-computing markets, including not just gaming rigs but workstation platforms. Comprising two eight-core Ryzen 7–type dies in a single multichip package (with intra-package interconnect à la its Epyc siblings), Threadripper populates as many as 16 cores — well beyond the typical quad-core CPUs powering the vast majority of workstations CAD professionals buy and use today. In the past few months, several vendors have launched Threadripper workstations, including Boxx, Velocity Micro, Puget Systems, and Maingear. (Boxx has also hinted at a 1S workstation based on Epyc.)
Threadripper Workstation vs. Max-GHz Intel Core i7 Machine
I was able to get my hands on a Threadripper workstation, thanks to the folks at Boxx in Austin, Texas. The timing of Boxx’s Threadripper-based Apexx 4 6301 release proved serendipitous, since the company had also recently launched the similarly configured Apexx S3, built around Intel’s new six-core Core i7-8700K, overclocked to a staggering 4.8 GHz. As a result, I had the chance to assess both machines in the context of workstation computing, for CAD usage in particular but also for other high-performance, visually intensive segments of the market.
In addition, contrasting the two machines made for an ideal opportunity to explore an often misleading — or at least confusing — debate about which is a better workstation CPU: the fastest single-thread processor, or the one that can process the most threads in parallel? The two processors driving my Boxx workstation test units represent opposite ends of the CPU spectrum. On one hand, I’ve got a latest-generation (Coffee Lake) Intel Core i7 with a mainstream number of cores (six, although four is even more typical) running at a bleeding-edge frequency, and on the other I have a Threadripper running at mainstream frequency but populating a bleeding-edge number of cores.
To compare the two CPUs, I configured both machines with the same memory (footprint and speed), the same storage (a solid-state drive [SSD] on PCI Express), and the same graphics processing unit (an Nvidia Quadro 4000 GPU). Memory subsystem differences (e.g., cache sizes and memory channels) remained, but since those are a function of the CPU more than the system, I had virtually identical system specs.
I benchmarked both systems on the SPECwpc 2.1 benchmark, which does an excellent job of stressing all of a workstation’s components in a real-world environment. SPECwpc aggregates results from a slew of individual tests into groups that represent typical workloads for a range of workstation applications; the Product Development group is the most appropriate for CAD folks.
SPECwpc reports a composite score for each application group, but instead of focusing on the summary composites, I instead cataloged individual tests according to thread count, then compared the scores — and scores per dollar — of the Threadripper-powered Apexx 4 6301 normalized to that of the Core i7-based Apexx S3. Charting performance by thread count lets the relative strengths of the two CPUs stand out, as well as how much bang for the buck each delivers as a function of thread count. And just as you might expect, the 4.8-GHz six-core Core i7 consistently provided better performance per dollar on single-thread tests, while the 3.4-GHz 16-core Threadripper left the Core i7 in the dust on 32-thread workloads. The middle ground of eight-thread workloads turned out to be a mixed bag, though the under-core’d Core i7 lagged more than it led. No doubt, AMD Threadripper proved itself a strong contender for workstation duty — and the more threads made available, the bigger its advantage.