AMD Next Horizon: Zen 2, 7nm Vega, AMD On Amazon EC2

[pmarcelll]

Gregory Stoner left AMD very recently, so ROCm seems to be without a project lead currently. I hope the leadership of AMD realizes that ROCm needs more attention and resources, especially if they want to compete seriously.

[juno]

I beg to differ, core boot has nothing to offer most of us. The lack of interest is real.

Apart from the demand for free software, which is very high here and not noticeable in the masses, there is a huge demand for having boards with firmwares that just doesn't suck as bad as the ones currently available on the market do. They are buggy, slow, forget settings, take their time to initialise bullshit nobody ever needs in that stage, don't ship AGESA updates which can be causing ongoing compatibility issues e.g. with RAM, etc. etc. They often don't ship updates promptly or at all which is of course a huge security risk.

Google has shown high demand for CoreBoot not only for their ChromeBooks but also for networking hardware. Microsoft just open-sourced a similar approach for a minimal firmware for their surface products. Companies don't trust and don't want the blob firmwares and their ring -x stuff. Google and Facebook are replacing UEFI with LinuxBoot in a huge effort. The demand for working, minimal, free, maintainable firmwares is there and can't be denied.


BTT: nice. 2019 is going to be interesting.

[wizard69]

Apart from the demand for free software, which is very high here and not noticeable in the masses, there is a huge demand for having boards with firmwares that just doesn't suck as bad as the ones currently available on the market do. They are buggy, slow, forget settings, take their time to initialise bullshit nobody ever needs in that stage, don't ship AGESA updates which can be causing ongoing compatibility issues e.g. with RAM, etc. etc. They often don't ship updates promptly or at all which is of course a huge security risk.

the problem is you just described CoreBoot, the way it is today and the way it will be in the future.

Google has shown high demand for CoreBoot not only for their ChromeBooks but also for networking hardware. Microsoft just open-sourced a similar approach for a minimal firmware for their surface products. Companies don't trust and don't want the blob firmwares and their ring -x stuff. Google and Facebook are replacing UEFI with LinuxBoot in a huge effort. The demand for working, minimal, free, maintainable firmwares is there and can't be denied.

Obviously Google needs a new solution for ChromeBooks, they are free to use whatever. However Google can't even maintain Android distros, how would you expect them to maintain CoreBoot?

BTT: nice. 2019 is going to be interesting.

By the way I'm not saying the black box is good, but rather it is a lot more difficult to maintain this sort of hardware and software over the long term than many seem to believe. Linux it self regularly purges support for older systems and has bug that crop up after years. The problem with Coreboot supporters is that many suffer from sugar Plums dancing in their heads leading them to believe that CoreBoot will bring some sort of perfect perfection to their lives. It is a fairytale that denies the technical challenges of a project like CoreBoot. personally I don't see the big draw in CoreBoot and maybe that will change in time but honestly Linux and the open source world could use that talent elsewhere.

As for 2019, yeah very interesting things should be happening in 2019, I'm kinda excited as Zen 2 could be just the nuts. I'm hoping the APU variants get lots of GPU love for very compact high performance systems. I'd love to see AMD start a compact machine form factor like Intel has done with NUC. It would be great to hear that AMD has adopted Thundrbolt/USB-C standards. There are certainly good reasons to be hopeful for AMD now that they have broken away from Global Foundries. Not that I want to see fellow NY'rs loose their jobs but Global was really hurting AMD.

I'm in the market for a new compact desktop suitable for video serving and other home duties and 2019 is the time frame. So yeah I will keep my eye open and if nothing else enjoy the new tech news.

[madscientist159]

Companies don't trust and don't want the blob firmwares and their ring -x stuff. Google and Facebook are replacing UEFI with LinuxBoot in a huge effort.

LinuxBoot is a Linux shim loader tacked on top of the proprietary, signed, "ring -x" as you call it, stack. It is NOT a libre firmware or even close to one. That's why we went with OpenPOWER, because the entire stack, not just the last stage bootloader, is fully open.

LinuxBoot is a libre cherry on top of a huge proprietary sundae. Nothing more.

[starshipeleven]

LinuxBoot is a Linux shim loader tacked on top of the proprietary, signed, "ring -x" as you call it, stack. It is NOT a libre firmware or even close to one. That's why we went with OpenPOWER, because the entire stack, not just the last stage bootloader, is fully open.

LinuxBoot is a libre cherry on top of a huge proprietary sundae. Nothing more.

Don't post spindoctored half-truths no matter how good is your goal.

LinuxBoot removes all the UEFI firmware "userspace" and leaves only the basic blobs required for platform init (nothing running in ring -x), and also disables or redirects SMM to the Linux kernel. The firmware size reduction from a UEFI firmware to a LinuxBoot one is pretty substantial. It's not just a "last stage bootloader", that's what GRUB is.

The ME or PSP still have to be dealt with separately, of course, which are the ring -x things.

While it's indeed not a libre firmware as it requires blobs for the platform init, it's still orders of magnitude more free and secure than UEFI.

[Maddo]

Don't post spindoctored half-truths no matter how good is your goal.

LinuxBoot removes all the UEFI firmware "userspace" and leaves only the basic blobs required for platform init (nothing running in ring -x), and also disables or redirects SMM to the Linux kernel. The firmware size reduction from a UEFI firmware to a LinuxBoot one is pretty substantial. It's not just a "last stage bootloader", that's what GRUB is.

The ME or PSP still have to be dealt with separately, of course, which are the ring -x things.

While it's indeed not a libre firmware as it requires blobs for the platform init, it's still orders of magnitude more free and secure than UEFI.

IMVHO you're either libre or you're not. While LinuxBoot does indeed reduce the whole "UEFI OS" to the minimum, you're trusting the remaining bits to only do hw initialization. While I'm not a big fan of conspiracy theories, "the need of trust is a weakness" and more than once we've seen proven backdoors at the firmware level.

[nomadewolf]

Given that first generation Zen had all kind of hardware problems, is it recommended to wait until Zen 2++ and avoid initial Zen 2? Zen+ is surely a lot better than initial Zen. I upgraded from 1700X to 2700X and all the hardware problems were completely gone.

What kind of hardware problems?
I have a Ryzen 1700X and never had a single issue!...

[nomadewolf]

I beg to differ, core boot has nothing to offer most of us. The lack of interest is real.

The problem with Power9 is that it costs an arm and a leg...

[L_A_G]

I do have to say that the exact details of the new Zen 2-based Epyc CPUs weren't what I expected.

My understanding was that it would be like the current Epyc CPUs with 4 dies identical with the ones used in the desktop and workstation CPUs communicating with each other over infinity fabric and with all of the memory and peripheral IO in place. However if I've understood correctly they're removing most of the IO from the CPU dies and leaving them with just the infinity fabric that they then use to communicate with a dedicated IO die that handles memory and peripheral IO along with CPU die-to-die communication.

That makes me wonder if AMD is moving away from the clever use of a single die for desktop, workstation and server use cases or if regular Ryzen and Threadripper CPUs are going to be built around separate CPU and IO dies.

[darkbasic]

World first, world first...

Supports PCI-E 4.0... you can combine Instinct MI60 with Talos II

Makes no sense because of the proprietary firmware. Why would you buy a Talos if you plan to pollute it with proprietary stuff?