|
I'm pretty annoyed by AMD's support, there's a well documented KVM related bug that needs to be fixed. Their PR guide forwarded my mail to tech support person, he escalated my ticket twice after sending me the usual copy & paste responses and then they went silent again. At this point I suspect the bug is in the silicone because they won't even acknowledge it. 
|
#
¿
Jun 15, 2017 00:00
|
|
|
|
| # ¿ May 17, 2024 20:02 |
|
|
SwissArmyDruid posted:...silicon, eames. Silicone Valley is in SoCal. Whoops, well yeah, that. 
|
|
#
¿
Jun 15, 2017 00:04
|
|
|
redeyes posted:Huh, so like maybe %5-10? yeah, sounds about right
|
|
#
¿
Jun 15, 2017 15:13
|
|
|
heise.de reports that Skylake-SP (and Skylake-X) uses a mesh (like the Xeon Phi) instead of a ringbus machine translated link: http://translate.google.com/transla...n&langpair=auto e: i just realised that this is the wrong topic but I'll just leave it here because of infinity fabric 
eames fucked around with this message at 17:17 on Jun 15, 2017 |
|
#
¿
Jun 15, 2017 17:03
|
|
|
The guys who leaked the first Ryzen benchmarks in december claim say there's a B2 stepping (focusing on SoC/PCIe controller errata) on the way. https://twitter.com/CPCHardware/status/876193860946468865 It's just a single tweet so  but perhaps that'll fix the virtualisation issues that they refuse to acknowledge.
|
|
#
¿
Jun 18, 2017 15:27
|
|
|
I fully expect TR/Epyc processors to launch with B2 steeping cores if such a thing already exists.
|
|
#
¿
Jun 18, 2017 22:21
|
|
|
Combat Pretzel posted:Zen+ is scheduled for when? Beginning 2018? The B2 stepping ain't it, right? There's no roadmap but rumors suggest 12 months after the Ryzen launch, so Q1/2018 although I feel like that would be quite ambitious. B2 stepping is most likely going to be a minor bugfix/errata revision and won't be advertised separately.
|
|
#
¿
Jun 20, 2017 17:03
|
|
|
Server launch happening right now at http://www.anandtech.com/show/11562 I don't think there's a livestream. e: NDA lifted; overview at http://www.anandtech.com/show/11551/amds-future-in-servers-new-7000-series-cpus-launched-and-epyc-analysis eames fucked around with this message at 21:02 on Jun 20, 2017 |
|
#
¿
Jun 20, 2017 20:58
|
|
|
Can't help but be impressed to see this announced by a company that was on the brink of bankruptcy about a year ago    source: https://twitter.com/TheRegister/status/877265375439749120 eames fucked around with this message at 21:54 on Jun 20, 2017 |
|
#
¿
Jun 20, 2017 21:33
|
|
|
well tom's hardware wrote this but I couldn't find a slide to confirm it so tomshardware posted:AMD provided some basic benchmarks, seen in the slides above, that compare its processors to the nearest Intel comparables. The price and performance breakdown chart is perhaps the most interesting, as it indicates much higher performance (as measured by SPECint_rate_base2006), at every price point. It bears mentioning that Intel publicly posts its SPEC benchmark data, and AMD's endnotes indicates that it reduced the scores used for these calculations by 46%. AMD justified this adjustment because they feel the Intel C++ compiler provides an unfair advantage in the benchmark. There is a notable advantage to the compiler, but most predict it is in the 20% range, so AMD's adjustments appear aggressive. We should take these price and performance comparisons with the necessary skepticism and instead rely upon third-party data as it emerges.
|
|
#
¿
Jun 20, 2017 22:01
|
|
|
Anime Schoolgirl posted:the lower 16 core SKUs look like discount runt bins for cheap workstations. so that's where the R3 dies went...
|
|
#
¿
Jun 20, 2017 22:16
|
|
|
lDDQD posted:There's no B2 stepping, it's just techpowerup rumor mongering again. somebody found this:  so there's hope that Threadripper will be B2 and fix annoying errata like the NPT GPU pass-through performance bug.
|
|
#
¿
Jun 22, 2017 09:01
|
|
|
have you tried updating the BIOS/microcode?
|
|
#
¿
Jun 23, 2017 21:59
|
|
|
Yeah that's the problem. A couple of people sent a support requests to AMD regarding this issue, the requests were escalated and when they got "high" enough eventually blackholed, so no response given, neither positive or negative. At this point I am fairly certain that it will need a new stepping to fix this which is why I'm holding off Ryzen for now. If Epyc/Threadripper doesn't fix the problem then it's highly unlikely that we'll see it fixed before Zen+. It's really puzzling how they overlooked this for a system so perfect for virtualisation but launching a completely new architecture is no small feat so 
|
|
#
¿
Jun 25, 2017 19:58
|
|
|
Hmm, that makes Ryzen very tempting if passing through the entire SATA controller really solves all performance issues. His idea of using a physical drive to dual boot as physical and virtual machine is pretty neat too. This reddit thread has some *very* recent info on the topic, looks like one Ryzen use could in fact confirm that it's a KVM bug and a AMD dev will look into it tomorrow.  https://www.reddit.com/r/VFIO/comments/6iomlh/apparently_the_kvm_amdnpt1_performance_bug_is/ eames fucked around with this message at 23:08 on Jun 25, 2017 |
|
#
¿
Jun 25, 2017 23:05
|
|
|
well why not posted:Jay switched over too, but, naturally, he did an overkill hardline watercooling rig in a Phanteks case which doesn't really have the airflow and as such doesn't really use his build because it runs very hot. I never understood why people mount their watercooling radiators to suck air into the case. It seems like the equivalent of having a 200W space heater blowing into the case. Top mounted radiators blowing air out on the other hand make a lot of sense.
|
|
#
¿
Jun 27, 2017 14:22
|
|
|
welp, learned something new today.
|
|
#
¿
Jun 27, 2017 16:07
|
|
|
found this datasheet which may be useful https://docs.google.com/spreadsheets/d/151hj_fgVGRk6mfVQz5eAOcFK1olCvU-ww8N6e6D6424/edit#gid=27884076 e: to be honest these 1700 builds are looking so cheap that I'm really tempted to build one just to tinker with the new architecture. I just put together a whole system sans GPU without really paying attention to the price turned out to be cheaper than the 7900X CPU alone.
eames fucked around with this message at 23:28 on Jun 28, 2017 |
|
#
¿
Jun 28, 2017 23:17
|
|
|
I have a hard time understanding the way Ryzen XFR/Turbo interacts with a load that i.e. maxes out one core and has medium load on four other cores. Would a 1800X fall back to 3.6 Ghz base frequency on all cores in this case? Would a "manually overclocked" 1700 at 3.7 Ghz (most seem to do that) outperform it in that scenario?
|
|
#
¿
Jun 29, 2017 18:24
|
|
|
AFAIK it silently corrects single bit errors but doesn't report the kernel or halt the system when a two bit error occurs. I just ordered a 1700 system because I'm tired of waiting for Coffee Lake and need something to play around with; NPT performance bug seems to be fixable via software because people were able to get around it by using Xen as a hypervisor.
|
|
#
¿
Jun 29, 2017 19:35
|
|
|
Looks like Ryzen really likes RAM speed for  loads.https://www.youtube.com/watch?v=RZS2XHcQdqA&t=128s Not the most reputable source but anandtech posters seem to back this up, increased infinity fabric speed (= higher RAM speed) increases Zen draw call performance *a lot*, so much so that a 4 Ghz/3600 Mhz CL16 Ryzen system suddenly becomes competitive compared to a 5 Ghz 7700K. If AMD can address this with Zen+ things are looking pretty good.
|
|
#
¿
Jun 30, 2017 10:28
|
|
|
some say he found a job at the Intel HEDT assembly line after MSI fired him for this video
|
|
#
¿
Jun 30, 2017 18:02
|
|
|
The main thing to watch out for with B350 boards are the VRMs. Some cheap boards can barely handle a stock 1800X. https://www.youtube.com/watch?v=ZGrxhf_xZWI&t=1551s eames fucked around with this message at 21:36 on Jun 30, 2017 |
|
#
¿
Jun 30, 2017 21:34
|
|
|
Something to keep in mind is that AMD claims AM4 will last until 2020. There's a chance we'll see 8 core CPUs running at 5+ Ghz towards the end of the platform, that's one of the reasons why I ordered a X370 board with pretty oversized VRMs for today's CPUs.
|
|
#
¿
Jun 30, 2017 23:28
|
|
|
3peat posted:Which one? I chose the Gigabyte X370 Gaming 5 because it has a couple of features that are useful for virtualization.* The VRMs are good but not as good as the best boards (Asrock/Asus) but that's ok as I value stability over ~extreme 4.1 Ghz overclocks~. * working IOMMU groups and extra LAN/SATA controllers that can be passed through separately
|
|
#
¿
Jul 1, 2017 09:57
|
|
|
I received my R7 1700 yesterday and the first AMD experience in over a decade is not so pleasant, although it isn't AMD's fault. I ordered 16GB of GSkill Trident-Z RGB RAM with it because that was the only Samsung B-die RAM in stock. Long story short, my attempt at turning off the RGB lights bricked both RAM sticks. The RGB lights are controlled by an SPD setting and the Gskill software has a problem with the Ryzen platform's SMBus, so it (sometimes?) writes garbage to the SPD and bricks RAM sticks. According to other people with this problem Gskill does not honor warranty for sticks with rewritten SPD. Lovely! Fortunately there is a solution. http://softnology.biz sells software for $16 that can flash SPDs and has a database of common RAM sticks. Since GSkill doesn't write things like serial number or manufacturing date into their modules it should be possible restore them to factory conditon. The only problem is that I have to find an Intel DDR4 system to do this because this Ryzen system straight up doesn't recognise the sticks anymore, not even in the BIOS. TLDR: Don't buy dumb RGB RAM for your Ryzen systems. Actually don't buy dumb RGB RAM period.
|
|
#
¿
Jul 4, 2017 08:31
|
|
|
Visidan posted:What the gently caress. I always see people recommending this exact RAM too. The comments of this Youtube video show that not just my system, sadly I didn't see it earlier or I would have ordered different B-die RAM elsewhere. https://www.youtube.com/watch?v=ARfK-YLaPZo I've got a spare PC-2133 stick here so the system boots but neither of the Gskill sticks shows up anywhere, not even in the SPD software. Intel systems are supposedly better at recognising sticks with corrupt SPD so I'm going to try that with a friend's system later today. The "Taiphoon Burner" licenses are system-specific so I get to buy a license for his system. 
|
|
#
¿
Jul 4, 2017 12:38
|
|
|
The Trident-Z had significantly better timings than the more expensive non-RGB part, so it seemed like a no-brainer because my Define S has no window. Anyway, we were able to revive one stick on a Intel board. Second one is still completely dead. Fingers crossed for RMA. Other observations: The 1700 stock cooler has a vapor-chamber core, not a solid copper core like most Intel coolers. Cools much better than I expected. Think twice before ordering some Noctua monster unless you need more than 3.8 Ghz. IOMMU groups are fine but turns out that my board does not have the additional separate SATA controller. There's no isolation between on-die controllers, so I can't just pass through one of the two internal SATA controllers. I can pass through the NVME drive though so that's no problem. VM CPU performance is still abysmal without NPT but at least the KVM maintainers acknowledged the bug. UnRAID is buggy and requires C-States to be disabled in the BIOS. The system pulls almost 110W idle at 3.7 Ghz and "auto" vCore. (Seasonic Titanium PSU)  I'm going to keep ~25W Haswell system as a NAS and will be using this AMD box with a OS on bare metal until stuff gets fixed.
|
|
#
¿
Jul 4, 2017 16:04
|
|
|
SoftNum posted:Do you have a link to this or the bug report or something? Yes, here's the thread: http://www.spinics.net/lists/kvm/thrd4.html#149446 To quote the KVM Maintainer: "Nested AMD needs some care. It's known, but time has been lacking..." SourKraut posted:The Wraith Spire does? I thought only the Wraith Max cooler did. That's what I thought until I took off the fan to rotate the fan frame (  )Here's a vid I found that shows the soldered filling port. https://www.youtube.com/watch?v=p8qeLXjiVms&t=139s Bonus: watch the dude thread steel screws into the aluminum heatsink while it is mounted on his motherboard and blow the shavings into his RAM slots. 
eames fucked around with this message at 18:20 on Jul 4, 2017 |
|
#
¿
Jul 4, 2017 18:18
|
|
|
Since we're talking cases and windows... I'm that that thrilled with the NZXT S340. It looks good, cable management is excellent and the magnetic rubber puck for the headphone is a great idea but thermals and acoustics are pretty bad. The HDDs aren't decoupled and ventilated so they'll get seriously hot during maintenance operations like parity checks or zfs scrubbing. In retrospect I should have bought the Thermaltake F51, a shameless Define R5 copy but larger and with a silly amount of fans options.
|
|
#
¿
Jul 5, 2017 19:19
|
|
|
PerrineClostermann posted:I have the sudden desire for a 2x200mm radiator. 2x420mm evidently works too https://imgur.com/WObHeu6
|
|
#
¿
Jul 5, 2017 20:13
|
|
|
FWIW some people were able to sell the 1700 boxed cooler for $40 because it works well on the 1600X and some people like LED backlit AMD logos.
|
|
#
¿
Jul 12, 2017 12:11
|
|
|
to be honest i think Intel is becoming really worried because they have no response to Zen+ except for their current CPUs +10% clockspeed +25% power consumption
|
|
#
¿
Jul 12, 2017 16:38
|
|
|
new price and specs for the glued together HEDT CPUs Threadripper 1950X 16C/32T 3.4/4.0 Ghz 999 USD Threadripper 1920X 12C/24T 3.5/4.0 Ghz 799 USD https://videocardz.com/newz/amd-ryzen-threadripper-1950x-to-cost-999-usd I'd say 4.0 Ghz on all 16 cores should be possible with good watercooling and decent boards/vrms/heatsink. It's a huge chip and soldered so cooling it should be a lot easier than with SL-X. Intel's 18 core will probably cost $2000 and won't clock a whole lot higher unless they fix the X299 issues (VRMs, TIM, etc). Man how I wish KVM would fix that NPT issue. eames fucked around with this message at 10:24 on Jul 13, 2017 |
|
#
¿
Jul 13, 2017 10:18
|
|
|
If my little 1700 is any indicator then an underclocked/-volted TR 1950X should run 16 cores at 3.0 Ghz at less than 100W.  It's a shame games don't scale well with threads (yet) or that'd be my go-to platform.
|
|
#
¿
Jul 13, 2017 10:40
|
|
|
Maxwell Adams posted:Is there even a watercooling block on the market that could bolt onto that socket and actually cover the whole chip? Not that I know of. I'm sure they are being worked on. I wonder if it is going to be a challenge to keep all four Dies at the same temperature, probably not unless the coolant flowrate is too low. Combat Pretzel posted:If that's 4GHz boost, then I'm in. The IPC should be equivalent to my 4GHz Haswell-E, but I'd gain 10 cores and ECC. Four core 4 GHz boost seems realistic to me, double that of a 1800X and one core per Die.
|
|
#
¿
Jul 13, 2017 15:10
|
|
|
I just wish there was some decent higher frequency ECC RAM, PC-3200 would be perfect. AMD published a video confirming the previously leaked prices and specs: https://www.youtube.com/watch?v=J3pJ_--nf5E Early August. videocardz commenter posted:Cinebench scores comparison:  the scaling in cinebench seems very close to 100% eames fucked around with this message at 15:18 on Jul 13, 2017 |
|
#
¿
Jul 13, 2017 15:15
|
|
|
Combat Pretzel posted:Wait, I thought boost is all-core where as XFR gets you another 100MHz on a single (favored) core or something like that? No, the 1800X for example only boosts over 4.0 with two cores. Look at the drop-off where it says "Number of Active Cores > 2". XFR uses various parameters like voltage and thermal headroom for even higher frequencies during boost. At least that's my understanding... eames fucked around with this message at 15:41 on Jul 13, 2017 |
|
#
¿
Jul 13, 2017 15:38
|
|
|
I think Amdahl's Law describes this problem: https://en.wikipedia.org/wiki/Amdahl%27s_law and a quick look at steam's hardware survey shows dual/quadcore processors at 94%, granted that includes a lot of laptops but still. I think ideally you'd want a game engine that's well threaded to begin with and then uses extra cores to scale up/down features that can be parallelised depending on the available core count.
|
|
#
¿
Jul 13, 2017 21:13
|
|
|
|
| # ¿ May 17, 2024 20:02 |
|
|
AFAIK the APU line is going to be called Athlon, wendell dropped a hint months ago.
|
|
#
¿
Jul 14, 2017 20:04
|
|