Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

> They made products for customers, while simultaneously competing with said customers. It's hard to build any decent partnerships on that premise.

I once talked to an AMD engineer and asked why they didn't just build a barebones NAS chassis with some of their ryzen embedded stuff since Ryzen is quite popular but there's very few Ryzen products in that segment. Obviously QNAP and Synology have some decent demand. That's basically what they said, didn't want to compete with customers.

GPUs are sort of the opposite example where I don't really feel any particular attachment to MSI, Gigabyte, Asus, PowerColor, etc. EVGA and Sapphire are the only ones that have managed to claw together some consumer mindshare through their warranties. But the rest are essentially customizing an AMD/NVIDIA reference design PCB and a cooler that's perfectly forgettable and interchangeable with any other offering in their price class. They're not even allowed to do double VRAM configs anymore because that would cut into Radeon Pro and Quadro revenue etc. In that scenario I don't see a lot of value to having another middleman taking a 10% cut, and to me the value of products being available at MSRP, with no diffusion of blame through the supply chain, would be worth losing the partners over.

And while systems integration (NUCs, servers, etc) are clearly something where there's a lot of value from this kind of diversity, motherboards really are not. AMD and Intel have both killed off third-party chipsets like nForce, Abit, etc, and locked everything down to their one ecosystem they control, much like GPUs. It's not quite as pronounced as GPUs, and previously there was a lot of diversity in features, but PCIe 5.0 motherboards tend to reverse this, almost every board has exactly 2 PCIe slots and more or less the exact same featureset elsewhere. And the trend is towards more and more being onboard the CPU itself anyway (AMD chips the chipset is literally just an IO expander and is completely uninvolved in management tasks) and once onboard voltage regulators (FIVR, DLVR, etc) really start to take off in the next 10-15 years the motherboard is only going to become dumber and dumber. And in that world there's less and less of a need for a partner anyway - the CPU is all self-contained and locked down anyway, partners can't experiment and do cool things, they are a dumb pipe that pumps in 2V for the DLVR to step down at point-of-use. So why pay a 10% margin for their "value-add"?

It slays me that AMD is talking a big game about "open platform" and everyone is still locking down third-party chipsets and even the Platform Lock. If you want to lock the chip to the board for security, fine, but locking it to the brand doesn't do anything except ruin the secondhand market. Oh no I have to swap it out with another lenovo chip if I steal it, what exactly does that accomplish? And if you accept the conceit of the PSP there's no reason it has to be permanent anyway, you can allow the PSP to unlock it instead of permanently blowing e-fuses. And again, third-party chipsets are where a ton of innovation happened, it's better for the market if third parties can make those double-VRAM models and undercut AMD/NVIDIA's ridiculous margins on those workstation products, or shanghai a Celeron 300A into being a dual-socket system like Abit. What we have right now is tivoization in support of product segmentation, branded as a "security feature".




also to be clear, management engine is far far worse. the chipset boots the chip for Intel, Ryzen is a SOC always. For intel, it's super intimately involved with the processor bringup in ways that can't be exposed to third parties anymore. They literally can't open anything while the ME is on the chipset, but they're flailing at homeostasis let alone big rearchs of their processor's brainstem against the possibility of third-party control of the bringup.

The chipset is a pure IO expander for AMD. AMD still is doing way better at that, it's just things like X300 ("the chipset is no chipset") being restricted to industrial/embedded, or board partners not being allowed to pursue things they want that break AMD's segmentation. PCIe 4 enablement (including opt-in) on select X370/X470 boards was something partners wanted for example. And there was no technical reason for X399/TRX40 to be segmented and even WRX80 could have been shoehorned on with "everything works just not optimally" level compatibility backwards and forwards. Partners could have done that if it wasn't denied/locked out. They did it on the Socket SP3 flavor.

Partners should ideally just get the freedom to play, and if they can make something work, cool. Let's have more Asrock/Asrock Rack and ICY DOCK design shenanigans again. Clamshell VRAM cards should be sold relatively close to actual cost rather than being gated by both AMD and NVIDIA. Etc. Partners should have the ability to configure the product in any way the product could reasonably be engineered to work. If features are being explicitly segmented by product tier it should be enforced by e-fuse feature-fusing at launch and that's the deal, no taking AVX-512 out after it launched.




Consider applying for YC's Fall 2025 batch! Applications are open till Aug 4

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: