Less than a month left until the launch of Microsoft’s Surface range, the first commercial products with Qualcomm’s new Snapdragon X Elite processor on board, and the bomb explodes.
Charlie Demerjian, founder of Semiaccurate and a longtime journalist who used to work for The Enquirer, is certain that Qualcomm is cheating on the data released regarding the Snapdragon X Elite. Semiaccurate was the first to write about the switch from Intel to Apple Silicon, about NVIDIA’s Fermi processor problems, about faulty graphics processors in Sony’s computer, and was often able to anticipate NVIDIA’s and AMD’s lineups. In short, there is no doubt that in the world of semiconductors he has the right sources and knows who to talk to.
The accusations launched by Demerjian are very serious: on the one hand the finger is pointed at Qualcomm’s lack of technical information relating to the new ARM SoCs for Windows, on the other at the fact that the press and Youtubers benchmarks are given that are not at all truthful and that Qualcomm is somehow cheating. It is necessary to clarify this aspect, however, because the scam would be linked to the fact that notebook manufacturers are not able to replicate the same results with production machines.
There are some aspects of the accusation article that don’t add up, the first is obviously the wrong name of the processor, Pro instead of Plus, and the second is linked to some information that Qualcomm allegedly refused to give. It is true that the American company remained very mysterious about some details, especially at the Snapdragon Summit, however during the European briefings we were given details that according to Demerjian Qualcomm would have refused to give to the Americans. For example, we have the cluster configuration of Snapdragon cluster. This is information that Qualcomm gave us upon explicit request, but which is actually missing from the official data sheets.
The part relating to the benchmarks is instead more interesting: we have repeatedly shown how fast the Snapdragon the reference design, we were able to run the benchmarks on both battery and power, and we were able to re-download Geekbench and other benchmarks from the internet to ensure that the pre-installed version had not been optimized or modified in any way. The data declared by Qualcomm are aligned with those we were able to verify on the computer made by Qualcomm. The only aspect that still remains unknown, and about which we too have raised several suspicions, is the actual consumption of the individual Oryon cores, given that Qualcomm refused to give and which we were unable to obtain.
Demerjian However, it does not say that Qualcomm’s data is distorted, but that what is seen on the reference designs produced by the company is not reflected in the products that will arrive on the market. According to the founder of Semiaccurate two of the main OEMs, with a very capable engineering department, they would have found themselves in serious difficulty for two reasons. The first would be the impossibility of obtaining the same results, the second a terrible state of development of Windows on ARM in terms of stability and performance. According to Demerjian, a source at Qualcomm also confirmed this: they are perfectly aware that the processor is strong only in certain specific conditions which are fine on a reference design but which then cannot be applied, also for cost reasons, to production computers . Initially Qualcomm would have justified the mediocre performance with poor optimization of the drivers, the dissipation system and Windows, but as time passed the situation did not change and one manufacturer even declared that the San Diego company is passing off as miraculous a processor that has the same performance as a Celeron.
The sources are rightly protected, and we don’t know who these top-level OEMs are who are complaining about the Qualcomm SoC, what we know is that a Lenovo with Snapdragon Qualcomm. The same thing goes for the Surface Laptop, with the benchmarks popped up just yesterday and aligned in performance.
It is also true that in recent months benchmarks of a Samsung notebook with Snapdragon X Elite had appeared and the performance was decidedly lower, 50% slower than the data released by the company.
There is another piece to add to this puzzle: last year, when the existence of Snapdragon by Oryon, initially designed for data centers, It required unique and incredibly expensive core power managementwith a dedicated power stage for each cluster, something that in today’s notebook world would be unworkable if you want to provide a product at a competitive price.
It cannot be ruled out that Qualcomm’s reference design was structured without thinking about costs at all, it only serves to show performance and how a computer should be made according to Qualcomm, and that the reality is the one described by Demerjian, therefore a more problematic Soc than expected to use and with several problems. This would explain the absence of a development platform and the impossibility for us to have finished products in our hands before the actual launch.
Microsoft should be the first to launch Surface in a month, and Qualcomm told us that for product tests we will have to wait for samples from the various OEMs which will be distributed to the press before sales or at least at the same time. According to Semiaccurate, it’s all part of the strategy: making sure that everyone talks about the performance of the Snapdragon, while allowing them to touch on the only notebooks that really offer the much-vaunted performance, only to then launch products on the market that are far from their promises. On the notebook front, it promises to be an explosive month.
Qualcomm, contacted by us, responded in this way: “We stand firm in our performance claims and can’t wait for consumers to get their hands on the Snapdragon X Elite and X Plus devices”. Qualcomm is too big and exposes itself too much on these numbers, it would be a debacle if they weren’t truthful.
For our part, as we have written, we are confident that we have achieved the results that Qualcomm advertised, but we did it on the only machines we had access to. We have always said that there is certainly something that Qualcomm hasn’t said, and that the way they have always treated the “consumption” topic is a bit vague, they always talk about the entire platform and never about core consumption, but not we never gave it too much weight because in the end each OEM implements a different consumption profile so it is difficult to give a value that can then be replicated.
We are waiting for the first samples, which should arrive in a few weeks.
