How did Qualcomm snatch defeat from the jaws of victory with their Oryon/Houma SoC? Stick with SemiAccurate for this one, it will blow your mind.
Author’s Note: If you notice a bit of an attitude in this story, we will just say there is a long back story to it. The facts have been verified, most with multiple sources, and we are sure of them. The snark is because of Qualcomm’s behavior at past events. They were condescending and dared us to go around them. Game on. Part 1.
SemiAccurate has been documenting Qualcomm’s woes in the laptop space. Their ‘performance’ to date has been significantly behind their phone chips which cost them the prime socket in the space and burned their bridges with Microsoft. While performance of the core is quite solid, power remains an open question.
In this last article you may have noticed that we hinted that the rest of the SoC had some potential ‘issues’. SemiAccurate has spent months digging into this and now we have confirmed enough to write. We realize this is going to be a long one but stick with it, the end is worth it. Get the popcorn out and prepare the facepalm.
The Good:
Lets start out with the good, and that is the core. The Nuvia core is said to be perfect on A0, a frankly astounding feat for a clean sheet core from a new team. If anyone can think of another clean sheet core that good on A0, please email the author, we would love to hear about it. So far so really amazing.
The uncore took a little heat from SemiAccurate a few weeks ago. We pointed out that there were some problems with it mid-year and Qualcomm spun the silicon to cure these. The fix didn’t work and the purported release date is approaching fast. What we were trying to confirm is the nature of the bugs and that took longer than expected to confirm but the news for the SoC is actually good. The uncore is actually in decent shape and the spins are done more to cure external issues than silicon ones. If that makes you scratch your head, don’t worry, we did too.
Power Play:
Things started to go off the rails early for Oryon, not for design reasons but for idiotic corporate greed reasons. Qualcomm wants to sell more of their PMICs and Qryon seemed like a good vehicle to do it so they force bundled the two products. This is hardly unprecedented behavior although it is market distorting and ensures that the end result will be less than it could be. More on this in a bit but just realize the PMICs are not picked for merit, they were chosen because Qualcomm makes more money that way, quality of the final product be damned. Remember the crack above about Microsoft?
Back to the story, there was an issue with said PMICs. Basically they are cell phone PMICs and Oryon is not a cell phone SoC. Why is this an issue? You might be aware that cell phones and laptops require differing wattages, and not by a little. Those cell phone PMICs are optimized for phone currents, running them at laptop levels is far less efficient than engineers would prefer. But it can be made to work. The sane answer is to use a less expensive off the shelf PMIC that would result in a more efficient end product. Since that would also have the effect of making Qualcomm less money, efficiency be damned, Qualcomm PMICs it is. But…
Let The Stupid Begin:
One thing we didn’t mention above is that those PMICs, in addition to running outside their optimal efficiency range, can’t actually handle the current needs of a laptop SoC. Whoops. The solution? Run a bunch of them in parallel, we are hearing 4-6 are recommended. For Qualcomm this is a win/win, they get to not only force bundle PMICS but they get to sell far more of them than they do on a phone. How could they lose?
Since you asked, you probably don’t realize how inefficient this solution is. If basically blows out the already tenuous power situation for Oryon, Qualcomm is intentionally forcing an awful power solution because they make it and more importantly they sell it. A much less expensive off the shelf PMIC would result in a quantifiably more efficient laptop for the end user. While we have no legal training, Qualcomm has an exclusive on ARM laptops until the end of 2024 and is leveraging that to force bundle other products. What was that board game with that property named Boardwalk called again? Applicable?
Let The Stupid Accelerate:
So running multiple unsuitable PMICs together will work and will make more money on paper, more on this later, even though the end user suffers. AMD and Intel are probably smiling broadly at this point because one of their potential 2024 laptop competitors just took themselves out of real contention. But the solution will work which means all good, right?
Not really. You see those PMICs are as we keep saying, cell phone parts, specifically high end cell phone parts. Because of this, they need a .6mm pitch HDI PCB which is common in high end cell phones so no worry on availability. It is however priced for high end cell phones, IE very expensive per unit area. Very very expensive compared to laptop and desktop PCBs but a high end cell phone doesn’t use much area, they are mostly battery inside, and they can carry that cost well.
Now scale that up to a laptop which is not mostly battery inside. Laptops have a large multiple of the board area of a cell phone, think more than 10x rather than a percentage. So a very expensive cell phone spec board just blew out costs for Oryon laptops. Whoops. Some OEMs SemiAccurate talked to were a tad peeved by this because it is entirely unnecessary, it is mandated solely by the force bundled PMICs. Allowing a suitable PMIC would also allow for a much cheaper PCB too but as you might guess, Qualcomm took a different path.
Could It Get Worse?:
If you are thinking that this self-inflicted predicament is a problem, we are just getting started. Those .6mm pitch PCB are, as we keep saying, cell phone PCBs. Like cell phone PMICs, they are not designed to carry the currents that laptops require mainly because they have thinner copper layers, 1/4oz to be exact. Luckily there is an easy solution to the issue, add more layers to the PCB.
Yes Qualcomm’s force bundled PMICs require four more board layers to handle laptop currents needed for Oryon. If you thought a high end tight pitch cell phone PCB scaled up to laptop areas was expensive, four more layers blow out the BoM. Remember that yield for boards scales down with layers, and not linearly. When OEMs heard this, they went ballistic. Really. We confirmed the love they felt, it is real and heart warming. Not really, you don’t want to mention the name Qualcomm in the offices of some design teams right about now. That said, like parallel PMICs, it will work. But…
Of Course It Could:
At this point OEMs SemiAccurate talked to had enough of the Qualcomm mandated stupidity, or so they thought, and said they would buy the PMICs and just not use them. Yes, several OEMs were going to buy the PMICs and throw them away and put a cheap and more efficient off the shelf solution in it’s place. The PCB costs would be a fraction of what they were with the Qualcomm solution, the power delivery would be more efficient, and the net result would be a cheaper and provably better device for the end user.
As you might have guessed, Qualcomm would have none of this. Flagrantly violating the first rule of holes, they kept digging much to the detriment of the OEMs. Qualcomm told them that the SoC to PMIC protocols were proprietary and, purportedly, baked into the SoC. We will state up front that we haven’t been able to independently verify this last claim but have both heard the proprietary bit from multiple recipients of the speech and have circumstantial evidence that backs up the silicon side. More on this later.
So OEMs wanted a better, cheaper solution, and were willing to pay Qualcomm for the force bundled PMICs on top of this but not use them. Of course Qualcomm blocked them. To say that OEMs were a tad peeved at this point is understating things, incandescent is too nice a word as well. And still Qualcomm held firm so several OEMs basically threatened to shelve the entire mess.
Qulitay* Management Decisions:
So what could Qualcomm do at this point to placate the rightfully aggrieved customers? They are forcing bundles of PMICs which blow out efficiency, raise PCB costs beyond what is tenable, and flat out blocking OEMs from fixing the problems on their own. At this point money was the only way to ensure a device that hits the market, much less is in the correct price range to sell against faster and more efficient x86 solutions.
And that is exactly what Qualcomm did, shoveled money at the problem. Yup they are giving customers money to ‘offset’ the costs that the PMICs mandated, ’tis to laugh. As it turns out, the money Qualcomm is giving the OEMs to offset this mess is said to be MORE than the cost of the PMICs so they are net losing money on the deal. If that M-word we mentioned above is being abused to force bundle PMICs, what do you call it when you give the OEM more money than the cost to of an item to ‘buy’ products? Not being a lawyer and not being close to understanding the nuances here, it sure looks like a word that starts with D and rhymes with ‘mumping’ to us. But wait there’s more!
This Gets Expensive:
Remember those respins we mentioned in previous articles? The cores are fine, the uncore is said to be pretty solid, and yet Qualcomm keeps respinning it. Don’t forget a full respin on cutting edge TSMC nodes are well north of $10 million with metal layer spins costing far less but not trivial amounts. The latest respin SemiAccurate heard about was noted for not having fixed the intended problems. Also don’t forget that a wafer takes 4-6 months to move through the fab on a modern process, fixes aren’t timely in this space but a metal layer only spin is much quicker.
So when we said we couldn’t confirm the hardened status of the PMIC to SoC protocols on Oryon, we can say that multiple people have pointed to the silicon spins being done to alleviate platform problems. To us this says it is something to do with the PMICs but several sources have strongly suggested the woes go much deeper than that. We’ll update this if we get more confirmation but for now the root causes of the respins is a little opaque.
What Do We Have?:
In the end we have a flaming mess brought on by abject idiocy and greed. Unfortunately this is not the first time the industry has done such things but it is a pretty stark example of doubling down on a failed premise. Why anyone involved in this still has a job is beyond us but, well this story wouldn’t have been possible if Qualcomm had competent people who answered questions instead of condescendingly mocking the questioner and dared them to work around company messaging.
For the tech we have forced bundling of unsuitable, expensive PMICs. This forced the use of multiple unsuitable, expensive PMICs in parallel which blew out the efficiency of the device. That in turn forced a stunningly expensive tight pitch HDI PCB which wasn’t able to carry the current. More layers fixed that but made the cost so untenable that most OEMs balked.
Rather than allow them to fix the problem and still pay Qualcomm for the PMICs, Qualcomm shut the OEMs down with, “It’s proprietary”. When customers decided to walk, Qualcomm threw money at them, multiple sources saying that it is more than the cost of the PMICs so the financial ‘win’ of the forced bundle just evaporated and then some. Respins aren’t adding to Qualcomm’s margins, nor is the spectre of missing delivery deadlines. It is a mess.
How Qualcomm allowed this rank incompetence and greed to override sound technical decisions is beyond us. Unfortunately it is far from the only case like it at the company but those are topics for another day. Qualcomm may be able to hit the intended clocks we revealed in earlier articles for press demos and golden samples for benchmarks, but when the real world runs headlong into the forced inefficiencies, don’t expect it to end well for the products or the users of those products. The performance numbers can be hit but it is unlikely to be at an acceptable efficiency level. And it is all self-inflicted.S|A
Charlie Demerjian
Latest posts by Charlie Demerjian (see all)
- Microsoft Hobbles Intel Once Again - Sep 20, 2024
- What is really going on with Intel’s 18a process? - Sep 9, 2024
- Industry pioneer Mike Magee has passed away - Aug 12, 2024
- What is Qualcomm launching at IFA this year? - Aug 9, 2024
- SemiAccurate is back up - Aug 7, 2024