Re: [TOTALLY OFFTOPIC] Re: [?] Why should Distros be called as i386 for a 32-bit PC, and as amd64 for a 64-bit PC, when Intel Core PCs are also 64bit systems
Andrei POPESCU wrote: > On Lu, 15 mar 21, 17:21:39, Dan Ritter wrote: > > > > At last report: normal desktop Ryzens (nothing with a G suffix > > unless it also has a PRO marking) > > Do you have a reliable source for the lack of ECC support in G suffix > processors? > > And why would it work for PRO processors instead? > > I think it's unlikely AMD has 2 different cores for PRO and non-PRO, > it's more likely it either works for both or neither. https://www.asrock.com/mb/AMD/X570%20Taichi/index.asp#Specification I'm going to omit a bunch of details: AMD Ryzen series CPUs (Vermeer) support ... ECC & non-ECC, un-buffered memory* - AMD Ryzen series CPUs (Matisse) support ... ECC & non-ECC, un-buffered memory* - AMD Ryzen series APUs (Renoir) support ... ECC & non-ECC, un-buffered memory* - AMD Ryzen series CPUs (Pinnacle Ridge) support ... ECC & non-ECC, un-buffered memory* - AMD Ryzen series CPUs (Picasso) support non-ECC, un-buffered memory* * For Ryzen Series CPUs (Picasso), ECC is only supported with * PRO CPUs. The first APUs are the Raven Ridge, 2200G and 2400G, which aren't even supported on the current motherboards The next are Picassos, 3200G and 3400G, there's an explicit statement that only the PRO versions support ECC. The current ones are Renoir, 4000 series, and I haven't got a reliable source that they are ECC only on the PRO -- but I strongly suspect it. It's not the cores that differ between the PROs and non- -- it's the I/O chiplet. -dsr-
Re: [TOTALLY OFFTOPIC] Re: [?] Why should Distros be called as i386 for a 32-bit PC, and as amd64 for a 64-bit PC, when Intel Core PCs are also 64bit systems
On Lu, 15 mar 21, 17:21:39, Dan Ritter wrote: > > At last report: normal desktop Ryzens (nothing with a G suffix > unless it also has a PRO marking) Do you have a reliable source for the lack of ECC support in G suffix processors? And why would it work for PRO processors instead? I think it's unlikely AMD has 2 different cores for PRO and non-PRO, it's more likely it either works for both or neither. Kind regards, Andrei -- http://wiki.debian.org/FAQsFromDebianUser signature.asc Description: PGP signature
Re: [TOTALLY OFFTOPIC] Re: [?] Why should Distros be called as i386 for a 32-bit PC, and as amd64 for a 64-bit PC, when Intel Core PCs are also 64bit systems
Anssi Saari wrote: > Dan Ritter writes: > > As for the ECC support in Ryzen CPUs, as I understand it it's a bit of a > mess. Sure the CPUs support it but if it's not validated by motherboard > manufacturers, how do you know it actually works reliably? ... by trying it out and reporting the results to others, and reading their results and reporting your confirmation. This isn't a thing that the motherboard manufacturer can put in by accident. Anyway. If you need ECC support, you buy an EPYC server and get registered ECC support. If you would like to have ECC as a feature, you get a Ryzen board that's reported to work, and you get unbuffered ECC for one-bit correction and two-bit reporting. Then you overclock it to generate RAM errors, and it shows up in your system log. Then you bring it back down to normal speed. At last report: normal desktop Ryzens (nothing with a G suffix unless it also has a PRO marking) on any ASrock, most ASUS, and some Gigabyte motherboards will support this. To the best of my current knowledge, no MSI motherboards. -dsr-
Re: [TOTALLY OFFTOPIC] Re: [?] Why should Distros be called as i386 for a 32-bit PC, and as amd64 for a 64-bit PC, when Intel Core PCs are also 64bit systems
Dan Ritter writes: > Intel knew that their argument was bull: they owned the market > and needed ways of subdividing their CPUs to fit every price > point. Turning off ECC support was one of those ways. > That strategy started with the 80486, when they brought out a > cheap version called the 80486SX which lacked a floating point > unit. The SX has the floating point unit, it was just turned > off. Initially, yes. A panic move when AMD brought out their 40 MHz 386. It worked, got popular and later on the 486SX was manufactured separately with a smaller die and no floating point. As for the ECC support in Ryzen CPUs, as I understand it it's a bit of a mess. Sure the CPUs support it but if it's not validated by motherboard manufacturers, how do you know it actually works reliably?
Re: [TOTALLY OFFTOPIC] Re: [?] Why should Distros be called as i386 for a 32-bit PC, and as amd64 for a 64-bit PC, when Intel Core PCs are also 64bit systems
Sven Hartge wrote: > Stefan Monnier wrote: > > > From a purely technical perspective, it's hard to understand how Intel > > managed to pour so much energy into such an obviously bad idea. The > > only explanations seem all to be linked to market strategies. > > This history repeats for Intel on several fronts: > > Or the discussion about ECC for desktop devices. Intel argues "not > needed", which is, if you follow the Rowhammer issues, not true. AMD > just does it and it works. Intel knew that their argument was bull: they owned the market and needed ways of subdividing their CPUs to fit every price point. Turning off ECC support was one of those ways. That strategy started with the 80486, when they brought out a cheap version called the 80486SX which lacked a floating point unit. The SX has the floating point unit, it was just turned off. Worse: purchasing the 80487 math coprocessor to enable floating point support... the 487 was a full 486, that turned off the original. > Then there was FB-DIMM back in the 2008s. Nice idea, just, again, too > expensive and disconnected from the market in the end. Intel wanted more pricing points. > I personally am really glad that AMD got their stuff together again and > with their ZenX-Architectures showed Intel how it is done. > > What AMD now needs is a hit in the low, lower and ultra-low power > segment. They've got the low and lower parts now: 35W and 15W 4000-series APUs, from the Renoir design. Stefan and I were just talking about how you can't buy one with a normal motherboard right now because they are entirely allocated to systems integrators. AMD is selling 100% of production. They don't have any 7W or lower parts, but those things aren't very interesting compared to ARM64 architecture, where Qualcomm and Apple and any number of smaller shops are doing great things in the tablet and phone space. -dsr-
Re: [TOTALLY OFFTOPIC] Re: [?] Why should Distros be called as i386 for a 32-bit PC, and as amd64 for a 64-bit PC, when Intel Core PCs are also 64bit systems
Stefan Monnier wrote: > From a purely technical perspective, it's hard to understand how Intel > managed to pour so much energy into such an obviously bad idea. The > only explanations seem all to be linked to market strategies. This history repeats for Intel on several fronts: Look at the Netburst Pentium 4 desaster, which as scrapped as soon as the Israel division showed their improved concept based on the P3, which ran laps around the P4 while at the same time using far less power and had a bigger yield. Or the discussion about ECC for desktop devices. Intel argues "not needed", which is, if you follow the Rowhammer issues, not true. AMD just does it and it works. Then there was FB-DIMM back in the 2008s. Nice idea, just, again, too expensive and disconnected from the market in the end. And all in all the rather slow improvments on the CPU fronts, the piecemeal 5% increases sold as "big achievements" every year, while at the same time all improvements turned out to be major security problems. I personally am really glad that AMD got their stuff together again and with their ZenX-Architectures showed Intel how it is done. What AMD now needs is a hit in the low, lower and ultra-low power segment. Grüße, S° -- Sigmentation fault. Core dumped.