AMD RyZen CPU Architecture for 2017

Pin compatible with Zeppelin :cool:

AMD back in early 2016 seemed to call "Zeppelin" to what is now the 2*CCX core that is present on current Ryzen CPUs.

zgFM4Ax.jpg
 
AMD back in early 2016 seemed to call "Zeppelin" to what is now the 2*CCX core that is present on current Ryzen CPUs.

zgFM4Ax.jpg
The core is still called Zeppelin, it doesn't refer to just 2*CCX but the whole 8-core chip with interconnects and memorycontrollers and everything
 
I'd be shocked if HSA were really gone, but HBM, tempting as it is for APUs, is still limited in capacity and quite expensive. AMD could theoretically make a chip with dual memory controllers and PHYs, but that would cost them some effort, time, silicon area and power. Being restricted to HBM alone would make the chip unusable for a lot of server workloads, which could be a problem.

Sent from my Archos 50 Helium Plus using Tapatalk
 
I'd be shocked if HSA were really gone, but HBM, tempting as it is for APUs, is still limited in capacity and quite expensive. AMD could theoretically make a chip with dual memory controllers and PHYs, but that would cost them some effort, time, silicon area and power. Being restricted to HBM alone would make the chip unusable for a lot of server workloads, which could be a problem.

My line of thinking is if Intel gets to charge $400 for the 15-28W dual-cores with Iris Pro and eDRAM, why wouldn't OEMs ask AMD for a similarly premium APU using HBM, knowing that it'd get substantially better performance?
 
Especially if the HBM is now being treated as a cache as they've been demonstrating recently. Should be a killer solution at low cost.
 
My line of thinking is if Intel gets to charge $400 for the 15-28W dual-cores with Iris Pro and eDRAM, why wouldn't OEMs ask AMD for a similarly premium APU using HBM, knowing that it'd get substantially better performance?

They probably dont even need HBM to do it*, but if i was AMD i would have thought it was worth the cost to put 1 HBM memory controller on a RR die connected to the GPU cache controller. In products that dont use HBM just package as normally. As HBM2 pricing changes OEM's/AMD can re-evaluate the cost/ performance trade off in there pricing stack.

*hopeflly AMD with 1/2 decent cores will force reviews etc to actually evaluate intels GPU drivers/real world performance far more then they currently do.
 
VME Broken on AMD Ryzen
That’s VME as in Virtual-8086 Mode Enhancements, introduced in the Intel Pentium Processor, and initially documented in the infamous NDA-only Appendix H.

Almost immediately since the Ryzen CPUs became available in March 2017, there have been various complaints about problems with Windows XP in a VM and with running 16-bit applications in DOS boxes in Windows VMs.

After analyzing the problem, it’s now clear what’s happening. As incredible as it is, Ryzen has buggy VME implementation; specifically, the INT instruction is known to misbehave in V86 mode with VME enabled when the given vector is redirected (i.e. it should use standard real-mode IVT and execute in V86 mode without faulting). The INT instruction simply doesn’t go where it’s supposed to go which leads to more or less immediate crashes or hangs.
 
Felix, what does that mean in the real world ?
Is it just a problem for vm's
Let's put it this way. If you had to put your OS install inside a vm, your head would explode from rage. For the rest of the world, they're really not affected.
 
I use 32 bits and 64 bits VMs on a daily basis and 64 bits oses didn't work at first 'cos the option to virtualize 64 bit OSes was disabled in BIOS. But other than that it is working fine. Gotta try with my Windows 95 vm.
 
There's plenty of bleeding edge users at the server forums that have been having nothing but grief trying to keep their new server builds stable. Like with everything Linux based, they should have waited for actual hardware support before jumping in. I hope that their issues aren't tied to non-fixable hardware and simply need better software support built into the kernels.
 
I had no idea such an enhancement to the VM86 mode existed.

If you had a 386 or better running Windows 3.0 or 3.1, the VM86 mode is what is used to run DOS command prompts (multiple of them!) from Windows.
There's still something similar today : you can install the 32bit version of Windows 10 on physical hardware and run Windows 3.1 applications. It would be fairly unlikely someone does this on a Ryzen but then your Windows 3.1 applications should crash and fail to run even though you're not using VMware, Virtualbox or Hyper-V etc.
 
I'd be shocked if HSA were really gone
AMD probably just doesn't bother to market it anymore. Its ROCm stack is built upon parts from the HSA effort anyway, and AMD can very well roll cache coherency and platform atomics for Raven Ridge in ROCm without HSA compliance at launch.

but HBM, tempting as it is for APUs, is still limited in capacity and quite expensive. AMD could theoretically make a chip with dual memory controllers and PHYs, but that would cost them some effort, time, silicon area and power.
It just costs more or less the effort of building a normal monolithic SoC, with extra cost in assembly. TBH it would just be a giant APU with a different configuration of GPU local memory. Whether it is marketable is another story.

Being restricted to HBM alone would make the chip unusable for a lot of server workloads, which could be a problem.
Why would it? The so-called "HBCC" already addressed the issue, and is working very well for Nvidia under the name Unified Virtual Memory. The HMM patch in Linux is also on its way to upstream. Let alone the fact that the APU may already access pageable system memory directly without it.
 
Last edited:
So, Ryzen can't run WinXP or previous Microsoft OSes either natively or inside a VM?
And if you're running modern 32bit Windows (from Vista to Win10) either natively or inside a VM, then the OS will work but legacy 16bit apps will not?
 
In the linked article and one of its comments, they give the workaround for disabling this thing in vmware (or leave to you the task of finding out how to do it if you use something else). That and a microcode update might either disable or fix the feature. Or a small patch for your virtualization software might work around it.
So, this is a visible hardware bug but not a terribad one.

We can be 100% sure there are many other ones like on any new CPU.
 
Last edited:
Both windows xp and server 2003 have x64 versions. Which cand run on Ryzen. Thus the issue is for existing deployments rather.

What that article seemed to imply was that this info was under NDA for a while. For me, that's the alarming bit. That sound unprofessional on AMD's part?
 
I'd expect it to be, given how it boots with a Ryzen logo :D

I'm getting really good performance on Prey at 4k Ultra but that's to be expected because I'm aiming for 60 fps and it is pretty much locked at that. Good thing they bundled that game with the Crosshair VI Hero, it's a very good game!
 
Back
Top