Qualcomm starts RISC-V joint venture

[Cmaier]

Qualcomm—one of Arm’s biggest customers—starts a RISC-V joint venture

The joint venture will first target automotive, with IoT and mobile coming later.

arstechnica.com

Hard to understand why Qualcomm would want to dabble in RISC-V when Nuvia‘s Arm designs are so awesome, and right around the corner any year now.

And not at all ironic that a licensing entity wants to avoid taking licenses.

 

[Jimmyjames]

I heard their new super-duper chip will beat H2 chip in my AirPods.

 

[Aaronage]

In the Arm space, every design I can think of that didn't originate from Arm or Apple has been mediocre. All of the in-house designs from Nvidia, Samsung, Qualcomm etc. were a disappointment for one reason or another. My takeaway from this (as an outsider to the industry) is that uarch design is really f*cking hard and the chances of success are extremely remote!

Good luck to this new joint venture, but history forces me to view it with scepticism (especially given it's Qualcomm).

 

[Agent47]

Do we expect Nuvia to turn out as vaporware?
Do we expect Microsoft (or any other player) to come up with a halfway serious Apple Silicon contender?

 

[Cmaier]

Do we expect Nuvia to turn out as vaporware?
Do we expect Microsoft (or any other player) to come up with a halfway serious Apple Silicon contender?

I’m sure that Nuvia will turn up at some point. I don’t expect it to match what Apple has by the time it comes out, but it will be good enough to be competitive with mainstream intel-based chips, I’d guess.

The reason I make fun of it is that people have been claiming for years that Nuvia was an M-series killer, but it’s been forever since qualcomm purchased them and still there’s nothing. Which tells me that they had nothing.

I worked with Manu for years, and he knows how to design blocks, so I’m sure eventually there will be something. But we have no reason to expect it to be the miracle that folks on Macrumors and elsewhere proclaimed it to be back when qualcomm made the purchase.

I don’t expect anybody else to come up with a contender any time soon.

 

[Agent47]

I see - as expected. Somewhat disappointing, though.

 

[mr_roboto]

I’m sure that Nuvia will turn up at some point. I don’t expect it to match what Apple has by the time it comes out, but it will be good enough to be competitive with mainstream intel-based chips, I’d guess.

The reason I make fun of it is that people have been claiming for years that Nuvia was an M-series killer, but it’s been forever since qualcomm purchased them and still there’s nothing. Which tells me that they had nothing.

I think a lot of it was that Nuvia got Gerard Williams and he brought some people along with him, and that got press since Williams was Apple's lead CPU architect for quite a while.

However, people are a bit too receptive to Great Man mythologizing. I've seen your thoughts on the following before, so none of what I'm about to write is really addressed to you, I think we're in broad agreement. What matters is whether Williams was able to set up as effective an organization at Nuvia/QC as he had at Apple. He might not have been able to! He couldn't (and didn't) poach everyone, and he couldn't take all the infrastructure Apple has built up for designing and validating CPUs and the SoCs built from them.

And even assuming he has been able to set up an organization and methodology capable of producing similar results, it takes time to do all that, and then some more time to get the first design out the door. And maybe even more time if you've been disrupted by getting absorbed by Qualcomm. (I don't know how much Qualcomm management messes with acquisitions...)

 

[Agent47]

I - being little more than a layman when it comes to CPU / processor design - am just surptised that nothing came out if Nuvia yet. It's been a while since Qualcomm sucked them up, my (unqualified) expectation was that they at least should have have announced a product by now.

 

[dada_dave]

I - being little more than a layman when it comes to CPU / processor design - am just surptised that nothing came out if Nuvia yet. It's been a while since Qualcomm sucked them up, my (unqualified) expectation was that they at least should have have announced a product by now.

The other issue is that those who left Apple to start Nuvia apparently did so to build server chips. What Qualcomm wants them to build isn’t what they had started the company to do. I have no idea how big an engineering factor that is, but in terms of goals …

 

[Cmaier]

The other issue is that those who left Apple to start Nuvia apparently did so to build server chips. What Qualcomm wants them to build isn’t what they had started the company to do. I have no idea how big an engineering factor that is, but in terms of goals …

It may well be that their “high performance cores” weren’t high enough performance for PCs. For servers you can get away with lower performance cores, as long as you have a ton of them and they are energy efficient. It’s not that easy to match Apple’s IPC and perf/watt.

 

[dada_dave]

It may well be that their “high performance cores” weren’t high enough performance for PCs. For servers you can get away with lower performance cores, as long as you have a ton of them and they are energy efficient. It’s not that easy to match Apple’s IPC and perf/watt.

And if they had a fabric, that would’ve probably had to change too I imagine. And if not, then it would still need to be designed. Supposedly if I remember correctly, the SOC will be getting all new fabric, a new Qualcomm designed GPU (presumably a different team but still - going to allow dGPUs too interestingly), etc …

 

[dada_dave]

In the Arm space, every design I can think of that didn't originate from Arm or Apple has been mediocre. All of the in-house designs from Nvidia, Samsung, Qualcomm etc. were a disappointment for one reason or another. My takeaway from this (as an outsider to the industry) is that uarch design is really f*cking hard and the chances of success are extremely remote!

Good luck to this new joint venture, but history forces me to view it with scepticism (especially given it's Qualcomm).

It’s also not jus uarch as I mentioned in the last post, everything around the cores also has to designed in house or gotten somewhere else if you aren’t going to go with ARM’s solutions. It’s a pretty massive undertaking.

 

[Andropov]

Found this post on Mastodon (related to Qualcomm’s Snapdragon 8+ Gen 1 benchmarking), interesting if it's not a software issue after all:

Hector Martin (@marcan@treehouse.systems)

Attached: 1 image Wait, 20 *milliseconds*? Either their kernel scheduler config is completely out of whack, or ARM/Qualcomm really screwed this one up. Apple cores can boost to max in around 50 *micro* seconds. 20 milliseconds is just broken. That's more than one frame, and that's how you get...

social.treehouse.systems

 

[dada_dave]

Found this post on Mastodon (related to Qualcomm’s Snapdragon 8+ Gen 1 benchmarking), interesting if it's not a software issue after all:

Hector Martin (@marcan@treehouse.systems)

Attached: 1 image Wait, 20 *milliseconds*? Either their kernel scheduler config is completely out of whack, or ARM/Qualcomm really screwed this one up. Apple cores can boost to max in around 50 *micro* seconds. 20 milliseconds is just broken. That's more than one frame, and that's how you get...

social.treehouse.systems

Seems like from the thread that they’re settling on software with Dougall even noting that if you do similar tests on macOS it will look like it takes 90ms to ramp up on a Max.