|
I'd expect POWER to catch up in short order now that it's open sourced, has oodles of implementations, documentations, and IBM can essentially deploy the same business model as their recently acquired Red Hat for vendors deploying it. I have no idea if RHEL currently supports it, but if not that'll probably be fixed soon. I was really hopeful for RISC-V, but fully opening up POWER will likely eat up that market.
|
#
?
Sep 12, 2019 05:37
|
|
|
|
| # ? May 22, 2024 13:11 |
|
|
NewFatMike posted:I'd expect POWER to catch up in short order now Catch up to ARM in what dimensions? TFLOP/W? Server revenue? Application base? Unless you are using “short order” in a geological sense, I don’t see it. You’ve been able to get RHEL on POWER for a long time, fully supported, and “open sourcing” POWER9 isn’t going to make people suddenly start to manufacture it at scale on competitive processes. Everyone who wanted to manufacture POWER chips was already doing so, and weren’t impeded by it being really onerous to join OpenPOWER. I would be surprised if Google’s processor strategy people were able to keep a straight face when “threatening” Intel with their Rackspace collab toy, but maybe they were. Google is one of the few places that could actually move materially to a new architecture, because they control the whole stack. They would be a good customer, but they don’t make a market. It’s likely that the majority of POWER9 CPUs on Earth are still those inside the two DoE supercomputers, which is a market environment to which Earth economics do not really apply: even if every dollar spent on Summit went to IBM for CPUs, they paid less than $50 per 22-core CPU. That’s less than 1% of retail price. (All of the big supercomputers are built on economics like this. They’re like F1 teams for CPU manufacturers: cute press, “innovation showcase”, little business impact.) quote:I was really hopeful for RISC-V, but fully opening up POWER will likely eat up that market. This is more likely, but only because there is basically zero meaningful market for RISC-V outside of CPU architecture bloggers.
|
|
#
?
Sep 12, 2019 12:13
|
|
|
Subjunctive posted:This is more likely, but only because there is basically zero meaningful market for RISC-V outside of CPU architecture bloggers. You're going to see RISC-V used in all kinds of controllers, not being dependent on ARM is pretty valuable.
|
|
#
?
Sep 12, 2019 12:17
|
|
|
K8.0 posted:People look at the efficiency of low performance ARM CPUs and fancifully imagine that if they scaled up to perform as well as x86 CPUs, they wouldn't be facing the same performance/efficiency tradeoffs that x86 CPUs are. B-b-b-but Apple and ARM and (It's not going to happen)
|
|
#
?
Sep 12, 2019 15:37
|
|
|
It's an impossible dream but I would be really curious to see how a Linux laptop or Chromebook using the iPad Pro's chip would perform. I bet you could get some really incredible battery life and stay pretty light with a ~12" form factor.
|
|
#
?
Sep 12, 2019 15:41
|
|
|
iospace posted:B-b-b-but Apple and ARM and I have a feeling this statement won't age well  vvv because i personally do believe that Apple has a laptop ARM CPU in the works that'll compare favourably to the x86 architecture equivalents. I'm talking about 15-30W TDP, not 150W. time will tell. vvv eames fucked around with this message at 15:50 on Sep 12, 2019 |
|
#
?
Sep 12, 2019 15:42
|
|
|
eames posted:I have a feeling this statement won't age well Why?
|
|
#
?
Sep 12, 2019 15:46
|
|
|
Steve from Gamers' Nexus giving the lowdown on the new AGESA ABBA update. It's looking good.
|
|
#
?
Sep 12, 2019 16:08
|
|
|
eames posted:I have a feeling this statement won't age well I thought I read somewhere (probably on this forum) that apple is almost definitely doing what you say
|
|
#
?
Sep 12, 2019 16:14
|
|
|
Subjunctive posted:Catch up to ARM in what dimensions? TFLOP/W? Server revenue? Application base? Application base is what I was thinking. quote:This is more likely, but only because there is basically zero meaningful market for RISC-V outside of CPU architecture bloggers. I thought Western Digital had transitioned to using RISC-V for their memory controllers, but I have been super misremembering their SweRV thing. Comedy option would be Apple going back to POWER CPU implementations now that it's fully open sourced and they can sit on even later hoards of dollars.
|
|
#
?
Sep 12, 2019 16:19
|
|
|
Statutory Ape posted:I thought I read somewhere (probably on this forum) that apple is almost definitely doing what you say Apple uses a threat to move to ARM to keep favorable pricing with Intel. That's it.
|
|
#
?
Sep 12, 2019 16:27
|
|
|
Pittsburgh Fentanyl Cloud posted:Apple uses a threat to move to ARM to keep favorable pricing with Intel. That's it. They also pull this same stunt with AMD, as well. As long as Intel is willing to bend over for Apple, they won't change.
|
|
#
?
Sep 12, 2019 16:49
|
|
|
Idk why Intel wouldn't be willing to do that , or really any company that can afford to
|
|
#
?
Sep 12, 2019 16:58
|
|
|
HalloKitty posted:Steve from Gamers' Nexus giving the lowdown on the new AGESA ABBA update. 10% increase sometimes on 0.1% lows seems pretty good. I think they also mentioned this update has better fixes for high idle voltages/temps too
|
|
#
?
Sep 12, 2019 17:53
|
|
|
HalloKitty posted:Steve from Gamers' Nexus giving the lowdown on the new AGESA ABBA update. I can't watch the video at the moment, did they mention how the update achieved the performance uplift? I get the impression that they simply extended max turbo frequency upwards with the same voltage scaling at the cost of long term reliability. Admittedly nobody will care when the CPUs start becoming unstable in a decade or so.
|
|
#
?
Sep 12, 2019 19:19
|
|
|
eames posted:I can't watch the video at the moment, did they mention how the update achieved the performance uplift? https://community.amd.com/community/gaming/blog/2019/09/10/ryzen-community-update-bios-updates-for-boost-and-idle-plus-a-new-sdk
|
|
#
?
Sep 12, 2019 21:03
|
|
|
Well that was pretty quick. Gigabyte has the new ABBA bios out now. I think I'll give it a few days before I update though.
|
|
#
?
Sep 12, 2019 21:21
|
|
|
Most recent BIOS for my board is beta from a few days ago with Update AMD ComboPI1.0.0.3abb Based on ABB being the previous one I assume this is what MSI is calling AMD ComboPI and I'm waiting for one with ABBA (so a month or two).
|
|
#
?
Sep 12, 2019 21:28
|
|
|
Golluk posted:Well that was pretty quick. Gigabyte has the new ABBA bios out now. I think I'll give it a few days before I update though. Nice, I'll try this out tomorrow.
|
|
#
?
Sep 12, 2019 23:48
|
|
|
Even MSI somehow got a beta ABBA bios out two days ago after being super slow with ABB.
|
|
#
?
Sep 13, 2019 00:23
|
|
|
Hello AMD thread, you may remember me from last week where my 3700x-based build was constantly rebooting and wouldn't get past BIOS. I just got in a new NVMe hard drive and motherboard (this time an X570 Aorus Pro Wifi, because I wanted to get something going faster than the Newegg RMA process would go.) Installed both, had a thumb drive in with Windows install, booted up. Same issue, kind of, froze a couple seconds into the initial Windows install menu, basically before I could even hit "next" on the first window. I tried loading it into BIOS to see if it would sit stable there, but now it would sometimes even freeze and require a reboot there, with no pattern as to when. This is after a lot of random reboots on its own in general too. I tried swapping out the power supply with that of my existing PC and still had the same issue. Made triple plus sure all the cables were seated and plugged in as well. When I did get in BIOS, it was showing that it saw all the RAM and hard drives within the case (2x 16gb RAM, NVMe drive, platter drive), so as far as I can tell, that's working. Oh yeah, tried to update the BIOS too and it froze solid at 72% on there. When I rebooted it, the BIOS version went back to the default that came with the board. At this point the only components I can think of that might be causing this that I haven't swapped out yet are the processor itself and the RAM. Which would you all think is the likely culprit? I'm still within the RMA period, so I'm very willing to toss more parts back. This is bad comedy. Edit: swapped RAM with my old PC and same error happening. Guess that narrows it down to the processor? The Rat fucked around with this message at 04:46 on Sep 13, 2019 |
|
#
?
Sep 13, 2019 04:31
|
|
|
GRINDCORE MEGGIDO posted:Nice, I'll try this out tomorrow. Went ahead of updated from F3 to F5a on my Aorus Elite. Used their app center and bios update tool to do it from windows. Pretty painless, and haven't noticed any issues. Stock bios settings other than setting XMP and 1800 Fclk, and two cores maxed at 4192mhz, the rest around ~4150. It seems strange to me though, that for a 65W TDP chip, its running the fan far harder than my old i5-2500k (95W TDP OC'd even), both on Hyper 212 evo coolers. Just sitting on the desktop, and its jumping from 37 up to 48c, then back down as the fan ramps up. It's tempting me to just get an AIO for the "thermal load" capacity.
|
|
#
?
Sep 13, 2019 21:13
|
|
|
So I've had this problem on a number of Naples/Epyc servers where they will intermittently throw a NMI fault 2d off various CPU cores when running rhel 7.6. Finally after months of chasing things down and working with support, build 3.10.0-957 seems to have stabilized things. Just in time for 7.7 to come out which seems to regressed and its now worse than ever. But at least the kernel now knows that those faults are CPU C6 park issues because of some goofy issue with how the linux interacts with power management, so maybe there is hope that they'll figure it out. Apparently this has been going on since the release of zen1.
BangersInMyKnickers fucked around with this message at 22:21 on Sep 13, 2019 |
|
#
?
Sep 13, 2019 21:32
|
|
|
The Rat posted:Edit: swapped RAM with my old PC and same error happening. Guess that narrows it down to the processor? CPU deaths are so rare that -- in general -- diagnosing them is like the old saw in computer science: "No, it's not a compiler bug. It's never a compiler bug." Except that very, very occasionally, it actually is a compiler bug. You just have to impress upon the newbies the overwhelmingly more probable case that the problem is something they have done. In this case it sounds like you've done the due diligence in checking every other subsystem you can. Also, anecdotally, I've bought 2 dead AMD CPUs in the past 3 months myself: one near the bottom of the range (2200GE) and one at the very top (3900X). The 2200GE was just plain dead; the 3900X exhibited freakish, randomized, won't-POST/will-POST/boots-halfway/won't-POST behavior. Swap it.
|
|
#
?
Sep 13, 2019 22:11
|
|
|
Golluk posted:It seems strange to me though, that for a 65W TDP chip, its running the fan far harder than my old i5-2500k (95W TDP OC'd even), both on Hyper 212 evo coolers. Just sitting on the desktop, and its jumping from 37 up to 48c, then back down as the fan ramps up. It's tempting me to just get an AIO for the "thermal load" capacity. I see the same behavior (saw-tooth CPU temp at 'idle', 3600 w/ a NH-D15) playing around with power plans also seems to effect it - sometimes. I'm think that it's just the nature of beast when you have 4-8 cores in a little chiplet without much surface area to dissipate the heat, so I don't know if a AIO would help any in that regards. You might want to play around with your fan curves/hysteresis settings. I had to set mine flat 40% to 45C and then a linear ramp to 100% at 80C. That keeps my fans quiet until I'm under sustained load (and quiet there, too.) Do you temps look good otherwise?
|
|
#
?
Sep 13, 2019 23:23
|
|
Boys who love boys
|
monsterzero posted:I see the same behavior (saw-tooth CPU temp at 'idle', 3600 w/ a NH-D15) playing around with power plans also seems to effect it - sometimes. I'm think that it's just the nature of beast when you have 4-8 cores in a little chiplet without much surface area to dissipate the heat, so I don't know if a AIO would help any in that regards. I did something similar, it helps a bit keeping the sound constant, rather than the rising falling pitch of the fan. I may also try swapping the fans. I think this new one is far louder than my old 212. They seem normal otherwise, just the fan being louder to do it. Golluk fucked around with this message at 23:48 on Sep 13, 2019 |
|
#
?
Sep 13, 2019 23:45
|
|
|
Golluk posted:It seems strange to me though, that for a 65W TDP chip, its running the fan far harder than my old i5-2500k (95W TDP OC'd even), both on Hyper 212 evo coolers. Just sitting on the desktop, and its jumping from 37 up to 48c, then back down as the fan ramps up. It's tempting me to just get an AIO for the "thermal load" capacity. monsterzero posted:I see the same behavior (saw-tooth CPU temp at 'idle', 3600 w/ a NH-D15) playing around with power plans also seems to effect it - sometimes. I'm think that it's just the nature of beast when you have 4-8 cores in a little chiplet without much surface area to dissipate the heat, so I don't know if a AIO would help any in that regards. Watercooling would only work, if the fans are run based on the water temperature. There's maybe AIOs out there that do that stock, IDK. If you'd plug the fans into the mainboard headers, they'd still react directly to CPU temps and the sawtoothing.
|
|
#
?
Sep 14, 2019 00:59
|
|
|
Combat Pretzel posted:The sawtooth pattern is Precision Boost doing its thing. You'd either need to disable it, or hammer the fan curve into shape accordingly. Latter would be annoying I'd figure, especially if you value silence and want the fans running as slowly as possible, you'll have a steep ramp at the end of the curve, and it'll make even more of a ruckus when the sawtoothing strays into that part (say during warmer days). All AIOs control their fans based on water temp by default.
|
|
#
?
Sep 14, 2019 01:05
|
|
|
Combat Pretzel posted:The sawtooth pattern is Precision Boost doing its thing. You'd either need to disable it, or hammer the fan curve into shape accordingly. Latter would be annoying I'd figure, especially if you value silence and want the fans running as slowly as possible, you'll have a steep ramp at the end of the curve, and it'll make even more of a ruckus when the sawtoothing strays into that part (say during warmer days). Yup. That's why power plans (and AGESA versions) affect the behavior. No ruckus here. I just had to get it through my head that there's no benefit to my CPU being 35C instead of 35-45C, and to not even bother ramping up the fans before that point. It's not lot like you're worried about boost clocks at idle, and any real load it going to push a core up into and keep it at least the 50s. You also need to keep in mind you're probably looking at the temp of hottest core, so the overall chip temp, wattage and cooling system efficiency is probably not that different than your old platform.
|
|
#
?
Sep 14, 2019 01:29
|
|
|
What is different is thermal density. Zen2 is dissipating similar heat to previous generation CPUs, but doing it on much smaller cores. This means that the on-die temp of the hottest core will spike more quickly than older CPUs.
|
|
#
?
Sep 14, 2019 01:56
|
|
|
K8.0 posted:What is different is thermal density. Zen2 is dissipating similar heat to previous generation CPUs, but doing it on much smaller cores. This means that the on-die temp of the hottest core will spike more quickly than older CPUs. It's only two more cores, but I suppose allot of the chip is taken up by larger memory sizes as well. It's still a bit crazy when I have it set at 60% fan, seeing the temp drop to 32c, then spike to 40c every 4-5 seconds. Opening chrome it shot up to 63.
|
|
#
?
Sep 14, 2019 06:21
|
|
|
I think the offset and asymmetric position of the Die may also have something to do with it. https://youtu.be/C5gTffl98Xc?t=3m40s depending on the orientation some direct heatpipe coolers are suboptimal for Ryzen 2 because none of the heatpipes make contact with the hotspot.
|
|
#
?
Sep 14, 2019 06:22
|
|
|
Golluk posted:It's only two more cores, but I suppose allot of the chip is taken up by larger memory sizes as well. It's still a bit crazy when I have it set at 60% fan, seeing the temp drop to 32c, then spike to 40c every 4-5 seconds. Opening chrome it shot up to 63. It's not about number of cores. Zen2 is made on a smaller process than previous CPUs. There's a smaller physical area producing the same amount of heat, and thus less area to transfer heat through as well as less heat capacity. That means temps will spike faster. The an behavior is something that's avoidable, but the temperature volatility is not.
|
|
#
?
Sep 14, 2019 06:42
|
|
|
AMD confirms "not enough public pressure" to force them to port RIS from GCN 4.0 and 6.0 back to 5.0. Fiji support is left as an exercise to the viewer.
|
|
#
?
Sep 14, 2019 11:23
|
|
|
GutBomb posted:All AIOs control their fans based on water temp by default.
|
|
#
?
Sep 14, 2019 13:14
|
|
|
Combat Pretzel posted:Really? What the hell. Makes me wonder why EKWB and the like aren't selling temp probes and fan controllers for their custom loops. Increasing fan speed when the core temp changes instead of the fluid won’t do you anything except get you a fluttering fan. If the heat hasn’t moved out of the core in to the medium, then increasing air movement on the radiator does fuckall. You’d be better off increasing the circulating pump speeds
|
|
#
?
Sep 14, 2019 13:54
|
|
|
What? No, I'm annoyed that water temp based fan control isn't shoved in your face with custom loops. If you get kits from vendors, it doesn't have any of that. You need to specifically go out there and get inline sensors and external fan controllers. 10K probe inputs aren't standard on high-end mainboards yet.
|
|
#
?
Sep 14, 2019 15:55
|
|
|
Asus motherboards have headers for external temperature sensors and you can hook up a water temp probe to it. It would be great if that was standard across all manufacturers.
|
|
#
?
Sep 14, 2019 16:57
|
|
|
Finally got around to installing ryzen master, and seeing what this 3600 can do. Stock/PBO/Auto OC seem to behave virtually the same in cinebench r15. All cores at 3960mhz, Vcpu at 1.36, and a cpu temp of 75c (all fans maxed), getting a score around 1575. Switching over to manual, I set all cores to 4200, and lowered Vcpu until it crashed at 1.3v. Worked at 1.325, with a score of 1670, at 74c. Highest I got was 4250 all core at 1.3875v, with a score of 1690, at 81c. I tried 4275 at up to 1.45v, but only managed one pass of 1700 at 85c. I did notice the vcores drop to 1.4, so I may just need more aggressive LLC to go higher. I'll just stick to 4200 AC at 1.35v. One odd thing though, is when I set that in bios, Ryzen master shows Vcpu only peaking at 1.1v. But if I enable manual, and set it to 1.35v, the temp is the same.
|
|
#
?
Sep 14, 2019 23:37
|
|
|
|
| # ? May 22, 2024 13:11 |
|
|
My gigabyte board has two 10k thermristor headers which I use with an XSPC g1/4 plug that fits into where the plugs for the alternate coolant line routes in my GPU block are. BIOS had some stupid limitations like the CPU fan header can only be set to monitor off the CPU highest core temperature, so I had to plug all the fans in to aux/case/pump headers, but otherwise it works. It takes a couple minutes for the fans to begin to spin up significantly when I dump a solid 375w of power into the loop, and around one minute to ramp back down once I go back to idle (because they don't even start speeding up till it passes 34C, and long idle is 27-28C). Before that I used to use speedfan with custom curves which behaved roughly the same once I got it dialed in even basing it off the actual CPU/GPU temps, but since it is no longer updated and wouldn't work on newer boards I had to find a different solution. Otherwise Corsair makes a fan controller that can do it, but an aquacomputer is the ultimate high end + brute force solution (completely independent programmable computerized multi-channel fan controller).
|
|
#
?
Sep 14, 2019 23:54
|
|