AMD: A Rigged Test May Sell Investors, But It Won't Sell Zen
Aug. 23, 2016 12:01 PM ET Mark Hibben S.Alpha
Summary
•AMD recently announced that its new Zen Summit Ridge “outperforms” Intel's Broadwell E.
•But the test was rigged in AMD's favor by hobbling the Intel processor clock rate.
•The announcement is just another example of clever (and successful) AMD marketing.
In an event intended to divert attention from the Intel (NASDAQ:INTC) Developer Forum, AMD (NASDAQ:AMD) held a competing event, also in San Francisco, in which it claimed that its new Zen Summit Ridge "outperformed" Intel's Broadwell E. But the comparison was rigged by artificially limiting the clock rate of Broadwell E. No reputable review site would engage in such a subterfuge. More to the point, no one will base a buying decision on such a comparison. Except for the AMD fans, of course.
Source: AMD
An Accomplishment Marred
Let me first say that I do agree that Summit Ridge is a significant technical accomplishment. Through the leadership of Mark Papermaster, AMD has finally managed to live up to a marketing promise for its products. At its 2015 Investor Day, Papermaster promised to deliver a 40% improvement in "instructions per cycle" (IPC), and Summit Ridge has delivered.
For purposes of highlighting the IPC improvement, the throttling the clock of Broadwell E to 3 MHz, the same as Summit Ridge, was perhaps sensible. However, AMD's handling of this in its press release was deliberately misleading:
“
During the event, AMD demonstrated an 8-core, 16-thread "Summit Ridge" desktop processor (featuring AMD's "Zen" core) outperforming a similarly configured 8-core, 16-thread Intel "Broadwell-E" processor when running the multi-threaded Blender rendering software with both CPUs set to the same clock speed.
SA contributor Paulo Santos was quite right to express reservations about this approach, since it amounted to hobbling Broadwell E. The only 8 core processor in the Broadwell E line is the 6900K, which has a base CPU clock of 3.2 GHz and a turbo clock of 3.7 GHz, as described in Anandtech's review of Broadwell E.
So, depending on how AMD set up the clocks in Broadwell E, peak performance could have been reduced by as much as 19%. While it's commendable that AMD achieved its promised 40% gain in IPC, maximum clock rate is important in overall processor performance. Any processor design trades maximum clock rate against a certain amount of parallelism, but in the end, it's net performance that counts.
And in the end, AMD chose not to compare true "real world" net performance, but opted for a rigged, artificial comparison that allowed it to claim that it was "outperforming" Intel. There isn't a reputable review site in the known universe that would do testing that way, or would claim that AMD was "outperforming" Intel on the basis of such a test.
When AMD's Summit Ridge actually gets into the hands of reviewers, it will be compared with Broadwell E running normally configured, and it will be compared not just in a single benchmark, but across a wide array of benchmarks that give a clearer picture of real world performance.
AMD's claim follows the same kind of obfuscation that it engaged in when it compared the Polaris RX 480 with the GTX 1080. You remember that right? AMD couldn't possibly claim that the 480 was superior to the 1080, so Raja Koduri chose to compare two 480's in CrossFire mode against a single 1080, claiming that these outperformed the single 1080 and were a better value.
Subsequently, the claim was found to be false by Mark Walton at Ars Technica, and I pointed out that due to the high power draw of the 480, using two of them in CrossFire would more or less obviate the claimed power efficiency benefit of the card.
We're seeing the same kind of obfuscation with the Zen comparison.
Management Pumping
Why engage in such shenanigans? Why not do a real world test and let the chips fall where they may? Summit Ridge is actually a considerable improvement over previous generations. While it may not come up to the performance level of Broadwell E, it could probably be priced to be competitive in the market place.
Except that the bar is always being moved higher, and Broadwell E won't be the performance standard for desktops for very long. Early next year, when AMD is just getting Summit Ridge to consumers en masse, Intel will be rolling out its 10 nm chips. As I've been saying all along, Zen won't be competing with the current generation 14 nm Intel chips. It will be competing with higher performance, completely redesigned 10 nm chips.
I believe that AMD management fully realize this, and that they're going to be disadvantaged all over again. The focus of management has become to pump up the stock in the near term. AMD CEO Lisa Su is also the largest individual shareholder, at 1.024 million shares. Large institutional investors, such as Mubadala Development Company, which owns 17.8% of AMD, also benefit from pumping up the stock. And Mubadala also owns Global Foundries outright.
The strategy of previewing products more than a year in advance of their release, which AMD did at its 2015 Investor Day, has worked brilliantly to pump up the stock. And AMD has been able to sustain and build anticipation for those products by investors and fans through events such as Capsaicin. I don't doubt that AMD will become a text book case for marketing and business classes for years to come.
At some point, AMD has to deliver on its promises. Certainly, it didn't deliver with Polaris. With Polaris, AMD completely abdicated at the high performance end of the market, and its Computer and Graphics segment lost money as a result in Q2. Meanwhile, Nvidia (NASDAQ:NVDA) posted a 25% y/y increase in GPU revenue and a huge increase in operating profit for the company.
In order to distract investors from the failure of Polaris, investors are now being tantalized with a new set of promises, this time about Zen.
Investor Takeaway
AMD fan reaction to my articles has devolved into standard boiler plate messages. Mark's wrong because the stock is up so much and continues to go up. The fans a missing a number of points.
First and foremost, I don't write about technology companies with the intention of influencing the price of the stock. Furthermore, my assessments of technical competitiveness are independent of the behavior of the stock price. While investors point to AMD's stock price for validation, I see it as meaningless, except as an indicator of how self deceived AMD investors have become.
My long term view on AMD hasn't changed. It's trying to compete in a market that is fundamentally on the decline. The x86 PC is simply on the decline. I believe that even Intel realizes this. So what real difference does it make if AMD becomes more competitive relative to Intel? It's just fighting for a larger piece of a shrinking pie.
AMD's focus is entirely on the wrong technology: x86 PCs and GPUs for x86 PCs. AMD's focus is on the wrong business model: commodity processors when the industry is shifting to custom processors produced by device makers such as Apple (NASDAQ:AAPL). The broad industry trends are all working against AMD, and I believe that AMD's management understand this.
So what's the end game for AMD's management? I think it's about cashing out with huge profits, which AMD's management and large institutional investors will certainly do, no matter what befalls the company.
But Zen is good enough to get some traction, and allow the AMD fan/investors to live a little longer on hope of a real turnaround. And I acknowledge the success that AMD's management have had in shaping public perceptions. Through clever manipulation of social media, AMD's management have manufactured the momentum play of the year. Therefore I'm changing my rating of the stock to buy, with this caveat: when you start to see a lot of insider selling, get out.
Disclosure: I am/we are long AAPL, NVDA.
I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.
