Zen and the Art of CPU Design

Beyond the Blue HoRyzen

We are on the cusp of AMD releasing their first iteration of the Zen architecture.  Zen was built from the ground up to be competitive in IPC while offering SMT performance that should rival Intel.  AMD stepped away from CMT and decided that SMT was the way to go.  All indications point to this being the correct decision when it comes to a design standpoint.  While CMT was interesting, it could never match the performance of a traditional CPU core or a SMT enabled core.  The front end seemed to be too much of a bottleneck to adequately feed the two integer and single FPU/SIMD unit each module held.

I will not go into an in-depth architectural analysis of Zen in this editorial.  That will be saved for a later article.  Instead I am looking at the performance leaks we have seen and what kind of environment AMD is releasing this chip into.

From what AMD has given us, and what other unofficial leaks have posted, the Ryzen processor is meeting expectations in line with single thread and multi-thread performance.  AMD’s benchmarks likely are best-case scenarios for Ryzen, but they are still indicative of the performance that we can expect.  We also discussed with David Kanter the in-depth architectural features of the chip and what our expectations are before 3rd party benchmarks are released.

All indications point to AMD having released a truly competitive part at a price that Intel has yet to match.  The Ryzen 1800X is an eight core, 16 thread part that compares well to the i7 6800K and 6900K parts from Intel.  The 1700X and 1700 are also up for preorder and seem to perform very well for the price as compared to Intel parts at those price ranges.  This is a big jump forward for AMD.  It is a much needed one.  If they did not have a Zen, they would again be relegated as a second source supplier with poor margins and just managing to survive in a landscape dominated by Intel.

Ryzen seems to hit performance numbers for a fraction of the price of Intel’s high end Broadwell-E parts, yet provides double the threaded performance of the Skylake and Kaby Lake based Intel 4 core/8 thread parts in the $250 to $350 range.  These are areas that AMD has not been competitive with in a long, long time.  It is hard to remember the last time AMD had a processor on the desktop side that could command a $499 price tag.

So how does Ryzen fit into this market as compared to previous new products over the past 20 years that AMD has released.  First off I will say that this does not appear to be an Athlon 64 moment.  2003 was a slam dunk for AMD and they were able to introduce parts that not only competed with Intel, but essentially ran away from those parts.  It was THE gaming processor to have and excelled at nearly every other workload thrown at it.  It lead to many positive quarters for the company and a dramatic rise in desktop and enthusiast marketshare.

It also is not a Athlon K7 moment either.  AMD was able to offer measurably better overall performance as compared to Intel’s Pentium II/!!!.  It was not as big of a splash as the Athlon 64, but it too lead to many positive quarters for the company and some very competitive iterations of the product.

Ryzen is most certainly not the second coming of Bulldozer.  It meets its performance estimates and in fact exceeds what they were aiming for.  It does not run overly hot and does not suck a tremendous amount of juice.

Perhaps Ryzen is something new entirely for AMD that has not been seen before.  It competes with Intel at the high end for a fraction of the cost, yet we see areas where Intel will still hold an effective lead in performance.  Intel still has better overall IPC and that will benefit many applications such as lightly threaded games and workloads.  Intel also has a clockspeed advantage with their new Kaby Lake CPUs with the top end 7700K going from 4.2 to 4.5 GHz boost.  Compare that to the max 4 GHz boost and slightly lower IPC of the 1800X and we can imagine quite a few scenarios where the Intel part will have a significant advantage.

This is not to say that AMD will not be competitive against these 4 core/8 thread parts, and certainly AMD will have an advantage over those 4c/4t parts that approach $250.  My point here is that workloads these days are far more varied than what we have seen in the past.  Some things are heavily threaded, some things are focused on single or dual threads, and there are workloads in between.  For example there are games that will utilize two to four heavy threads while throwing off a handful of lesser important threads.  These games might see an advantage on the 7700K due to higher IPC and core clock, but other games that will leverage more than 4 threads heavily will see a boost on the AMD part.

The important idea to come away from here with is that AMD does truly have a competitive part for the marketplace at a very reasonable price considering what the user gets.  The first three SKUs are 8c/16t parts.  This has been unheard of for $499 and below.  Intel is charging $1000 for their 8c/16t part.  AMD has their Ryzen 1700 at $329.  This is a tremendous deal for somebody looking to address that number of threads yet not break the bank.

AMD does not appear to have an overwhelming part here, but it is impressive in what it can do.  Performance and user experience can be two totally different things and we will certainly learn more in the next few days where things fall exactly.  If there is one area that AMD is suspect in is that of the Southbridge and I/O performance.  AMD’s track record on chipsets has been inconsistent.  There have been some good ones and there have been some disappointments.  AMD looks to have contracted out much of the design of the chipsets to ASMedia.  We do not yet know how well that has worked out.  My hope is that it will be solid.  Worst case scenario is that AMD will suffer a 20% performance hit in I/O operations.

Closing Thoughts

AMD has done a very admirable job of pushing out an aggressive design that challenges Intel at the top end of their performance stack.  The Zen architecture is very forward looking and can be iterated upon in the future to provide increased performance and extra features.  A more modern foundation in terms of chipset design and feature set allows AMD to again more adequately compete with Intel in important OEM circles.  NVME and U.2 support are very big as well as USB 3.1 Gen 1 and Gen 2.

The shot across Intel’s bow has certainly gotten a lot of attention.  I am not sure if Intel expected AMD to pull off Ryzen as well as it has considering the obstacles in front of the company.  AMD does not have the R&D resources of Intel, and neither do they have a cutting edge manufacturing arm that works arm in arm with CPU design.  Instead AMD relies upon GLOBALFOUNDRIES which has not had the greatest track record on cutting edge process nodes since 32nm PD-SOI.  GF seems to have provided a very solid 14nm FinFET process that utilizes high density libraries to allow AMD to have a reasonably sized die given 8 cores and 20 MB of cache.

I think Ryzen will be a success for the company and they will start clawing back some marketshare while increasing their margins and ASPs.  Throughout the year we will see the company flesh out their offerings to include 4c/8t and 4c/4c products.  We have yet to see how they implement their APUs with the new architecture, but the addition of a new memory controller and the Infinity Fabric makes that product very interesting.

We will find out in a few days how Ryzen holds up against the best that Intel has to offer.  Until that time, my gut says that AMD does have a winner and a good seller on their hands.  There is a lot of pent up demand from AMD enthusiasts and so far the pre-order on Ryzen parts has been robust to say the least.  We can only hope that review results will convince others to jump on the bandwagon and help support AMD.  As the last five years have proven, Intel does not innovate aggressively unless there is actual competition in the marketplace.