Yoooooooooooooooooo, what the fuck?Server CPUs on my desktop motherboards?, more likely than I thought.AMD caught wind that Intel was lacking on the game server space so they are finishing intel forever.
>>101460035>© 2023 AMD
https://www.youtube.com/watch?v=n1tXJ8HZcj4Too bad ECC still seems to be a clusterfuck on DDR5.
>>101460035AM5 Epycs are just half-lock chips binned for energy efficiency and are tailored for platforms with BMC support
>literally worse than their ryzen counterpartslol, home server chuds cucked yet again
>>101461932No, they are practically the same thing. You are just paying access to BMC platforms and fancy virtualization which neither option makes any sense for home environment.
>>101462754Makes sense for game servers and stuff that needs high single core performance, the base frequency on these is higher than the boost on most EPYCs and almost double that of Xeons.
>>101460053He literally said it's fixed with the latest bios.
>>101462870I mean sourcing ECC ram compatible with AM5, you need UDIMM and a bunch of luck because nobody lists them on their QVL outside server motherboards.
>>101460053Most Ryzen SKUs have ECC support. The problem is with the motherboards where you really have to dig around to get it.
>>101460035does this mean chipsets that will allow more pcie 5.0 lanes? only being able to run a single gpu and ssd at full bandwidth is pretty pathetic
ecc ram is just shitty ram with an automatic address blacklist as addresses fail its shit and there is no reason to use it
>>101464349Nope, still capped to 28 but I'm pretty sure nothing can't saturate an 5.0 x16 slot yet, most boards just run the GPU at x8 and give you a bunch of m.2 ports.>>101464354The automatic part is very important.
>>101460035smells like a shitty rebranding cash grab
>>101462870being fixed in a bios update means nothing because amd can fuck up support for ecc in an agesa update at any time like it did for am4 leaving you to choose between ecc support or a later bios with security updates.wendell has also still not qualifed, not once on either his forum or his videos, whether multi-bit ecc functions as it should; most research into amd's ecc platform support on consumer motherboards/chipsets/cpus (including the asrock rack board he's advertising in that video) only turns up answers to the effect of: single bit errors get corrected, multi-bit errors aren't dealt with.he also still seems to be misguided at dmidecode being a reliable source of information>man dmidecode>dmidecode is a tool for dumping a computer's DMI (some say SMBIOS ) table contents in a human-readable format. This table contains a description of the system's hardware components, as well as other useful pieces of information such as serial numbers and BIOS revision. Thanks to this table, you can retrieve this information without having to probe for the actual hardware. While this is a good point in terms of report speed and safeness, this also makes the presented information possibly unreliable.the only way to know it is even operating in a half-assed state "reliably" is to query amd's memory controller as mentioned in this article:https://sunshowers.io/posts/am5-ryzen-7000-ecc-ram/and see if the proper bit is set.
>>101460053so what is the difference between mine 7950x and 4564P, is it some better quality silicon, will the cores boost higher?
>>101465345They have the exact same paper specsAMD probably relabelled 7950x's to epycs earlier this year when 7950x's were really low in supply
>>101465345effectively nothing, they're the same exact cpu, same exact family/model number (family 0x19 model 61h), etconly real difference is the ihs says epyc instead of ryzen
>>101464638buy stable ram which runs with 2x the speed instead of shit slow ram made to compensate for manufacturing defects under the guise imaginary radiation causes defects because ecc fucking blacklists bad addresses permanently
>>101465896the only stable ram, is ram that runs at jedec rated speeds.everything else is an overclock which will eventually degrade your memory controller quicker than it will happen during its rated life span.doesn't matter how much you jizz in your pants over your ddr5 8400 single rank dual channel setup that can pass karhu for 24 hours, this will always be true.
>>101466087nobody respects JEDEC, I recently had 2 systems crash like crazy after I put them together, one had some good DDR5 memory and the other was ECC DDR4, turns out the motherboard was overclocking them and pushing them out of spec with default settings.
>>101465345Epycs are half-locked and their boosting profile is more conservative.They also have the pass for some of AMD's virtual management tools and support for BMC out there.>>101465282It is limited by motherboard support kiddo. You need to have a board that has the tracing for parity chip and UEFI/BIOS that knows how to handle 2-bit error code halt. That's it. The vast majority of gayming SKUs and non-professional lack these and will simply treat ECC UDIMMs as regular non-ECC DIMMs.You just need really do your homework if you want to use ECC UDIMMs.
>>101464354Wrong kiddo. ECC memory is just JEDEC-spec DIMMs that have a parity chip on them. That's it. The main catch is that you need a memory controller with support, a motherboard has the tracing for parity chip and UEFI/BIOS that knows how to handle 2-bit error halt. The memory controller support has been eased up on Intel front as of recent and AMD historical has ECC support ECC on their higher-end mainstream CPU SKUs. The main problem is motherboard support there are limited SKUs out there have support and are mostly in the professional domain.
>>101467698Gayming motherboard SKUs don't respect JEDEC by default. You have go back and ensure it doesn't do silly "auto-overclocking" non-sense. OTOH, regular plain vanilla SKUs and professional-geared SKUs do.
>>101460035Would be a big deal but you don't even get more cores. What is the point? At this point the core count is stagnating, we need 24 core and 32 core desktop parts. And no your intel cuck cores don't count as real cores.
>>101465345i would think like Epyc gets better quality die or something.. to be more "reliable" for the server usage and that would mean it could sustain higher boost frequencies longer time(like lover temps or something).. each die is different, not single CPU is the same.. so i would expect the server CPUs to be the better dies from the waffer.. but who knows, thats just what i would think.. but what you guys say it looks like it is just the same
>>101470330Best dies probably go to V-cache chips of all kinds, since having the cache on top obviously has thermal implications so they need to run at lower voltage.