sedrosken
Florida Man
I'm not quite sure if the P8Z77-WS technically qualifies as a workstation platform, but it can take 1155 Xeons and ASUS calls it such, so I consider it one. It never performed its POST that slowly.
Ryzen
- Thread starter time
- Start date
Mercutio
Fatwah on Western Digital
Virtual CPUs are only ever about 60% as fast as real ones but if I had to guess I'd say you're getting more headroom in your thermal budget to hit higher clocks if you aren't using them.
ddrueding
Fixture
I am excited to see what Intel's new HEDT Xeons look like.
sedrosken
Florida Man
I was intrigued in particular by the W3-2423. As Linus said, it's only MSRPing for 30 bucks more than the 13600K, and while that almost certainly beats it in raw CPU grunt, the extra PCI-E lanes and memory channels aren't anything to sneeze at, not at all. I expect the platform in general to be wickedly expensive, so nothing I can really put hands on, but maybe a few years down the line when it's not the new hotness anymore... it looks like an ample platform for a home server. I kind of pressed my R5-2600 box into service on that front, I know for a fact it's not at all made for it, so it does give me some slight anxiety. No more than I had with the NAS box, for sure, but I definitely find myself getting less and less comfortable running consumer platforms in that role as I get older and my budgets get a little less restrictive.
Mercutio
Fatwah on Western Digital
I have a hard time taking Linus Sebastian seriously. I know he has smart people working for him but he's also the muppet version of gamer stereotypes, and very often his non gamer content boils down to "watch what happens when we spend the GDP of Burkina Faso on hardware."
But I have been advocating workstation IO for a while. When you have dozens of CPU cores, keeping your high end rig fed with data is a bigger trick most of the time than actually bottlenecking a CPU.
Intel got out of HEDT years ago and hopefully anything they do in that space will make Threadripper more interesting as well.
sedrosken
Florida Man
I'm not exactly a fan of his either, but I figured if most of his information was flat out wrong he'd be out of business by now. I do feel like he focuses entirely too much on the high end.
Never in the history of the world did that stop everyone. Especially since there is no peer review nor quality review of the boggers.
I watch his content more for the entertainment aspect and also he will occasionally find and talk about a tangentially-related product that I'll go explore on my own because I hadn't heard of it. It's more that he raised awarness on something I was unfamiliar with and less about his expert technical review of it.
ddrueding
Fixture
That piece was written by Anthony Young, who is very knowledgeable about the things. They've even started to let him make content on even more geeky things lately. Like PiKVM (which is amazing)
After further analysis, it was observed that the maximum reported power is ~230W with SMT enabled and ~215W disabled. I also tried playing around with different lower power limits in the BIOS. Under those conditions the performance was less with SMT, but I have read that some software takes better advantage of it. For now I will go without SMT as the power limits are dependent on the air cooling and 92°C is just below the throttling at 95°C observed with SMT. It is close though to the CPU limits, so I'd estimate that maybe an extra 100MHz on liquids is not worth it and with the particular apps it's probably a wash. The differences are only low single digit anyways. At least I know what TDP is feasible now.
ddrueding
Fixture
If the CPU is throttling more with SMT enabled that could explain it.
sedrosken
Florida Man
Yikes, I knew these chips run hot, but 92C on a good air cooler is a bit much. I guess you really are expected to liquid-cool these enthusiast-grade chips, I don't know of a commonly available air cooler that does any better than a D15.
Newtun
Storage is nice, especially if it doesn't rotate
I'm considering a new "cheapo" build with a Ryzen 5700g.
I can get DDR-3600 for just a few bucks more than DDR-3200.
Would that be significantly faster, for "compute-bound" workloads?
Or would the 3600 MHz RAM speed be limited by synchronization issues with the CPU's "System Memory Specification: Up to 3200MHz"?
I can get DDR-3600 for just a few bucks more than DDR-3200.
Would that be significantly faster, for "compute-bound" workloads?
Or would the 3600 MHz RAM speed be limited by synchronization issues with the CPU's "System Memory Specification: Up to 3200MHz"?
jtr1962
Storage? I am Storage!
For the AMD APUs, faster RAM mainly helps with the GPU. I noticed higher frame rates on my system going from DDR3-1600 to DDR3-2400.
If CL is the same then there is a little improvement with speed, but it is not so sensitive. I prefer CL14 at 3200 (or C16 at 3600) for Ryzen 3. Howerver, I would not pay a while lot more over C16 3200 for a budget build.
FWIW, I built a µATX 5700G system for a family member in late 2021. The HSF bundled with the 5700G is fine. It's a nice little system that doesn't use too much power.
FWIW, I built a µATX 5700G system for a family member in late 2021. The HSF bundled with the 5700G is fine. It's a nice little system that doesn't use too much power.
Last edited:
ddrueding
Fixture
For desktops I'll be sticking with Intel/NVidia, but my current home server is AMD, and
r.
Mercutio
Fatwah on Western Digital
Small servers are not necessarily loud. I have Lenovo RD550s servers at home, in my office and in the datacenter and they're all shockingly quiet except right when they start up and put the 20krpm fans on full blast for about 15 seconds. Even with the CPUs pegged at 100% for 10 minutes, they don't get nearly as loud as they are on boot.
But then Ryzen 7000s are notoriously hot and that's a 1U enclosure and IMO there's no way that thing isn't in leafblower mode full time unless the included heat sink is an 18" wide slab of finned copper.
Also, every Ryzen 7000 has ECC RAM, so I'm not exactly sure what the dude in dd's video is so excited about. IPMI is good but standard on anything from Asrock Rack, and I'd expect a 10GbE port on any high end motherboard. It still has SATA rather than SAS or U.2 and I didn't see an external SAS or Thunderbolt connector, so that's a REAL mixed message of a server. It's I/O limited for storage, PCIe and RAM (max 128GB), doesn't have the internal volume for any kind of advanced GPU and it only supports one PSU. I guess it would be decent for a home VM server, but I can't think of much else where someone would want a system with modest RAM, no GPU and such limited I/O and THAT many fast CPU cores.
But then Ryzen 7000s are notoriously hot and that's a 1U enclosure and IMO there's no way that thing isn't in leafblower mode full time unless the included heat sink is an 18" wide slab of finned copper.
Also, every Ryzen 7000 has ECC RAM, so I'm not exactly sure what the dude in dd's video is so excited about. IPMI is good but standard on anything from Asrock Rack, and I'd expect a 10GbE port on any high end motherboard. It still has SATA rather than SAS or U.2 and I didn't see an external SAS or Thunderbolt connector, so that's a REAL mixed message of a server. It's I/O limited for storage, PCIe and RAM (max 128GB), doesn't have the internal volume for any kind of advanced GPU and it only supports one PSU. I guess it would be decent for a home VM server, but I can't think of much else where someone would want a system with modest RAM, no GPU and such limited I/O and THAT many fast CPU cores.
Mercutio
Fatwah on Western Digital
Apparently, Intel has new fab processes that put it on par with TSMC "2nm" scale production. If that's the case, I fully expect Intel to absolutely leapfrog AMD in a generation or two.
I put the "2nm" in quotes because everything now is apparently a set of compounding rounding errors and the reality is that the gates are actually way bigger than that. It's like the Performance Rating values of CPUs we got 20 years ago. 2nm is the wavelength of a goddamn X-ray. Pretty sure they aren't etching wafers with an X-ray yet.
I put the "2nm" in quotes because everything now is apparently a set of compounding rounding errors and the reality is that the gates are actually way bigger than that. It's like the Performance Rating values of CPUs we got 20 years ago. 2nm is the wavelength of a goddamn X-ray. Pretty sure they aren't etching wafers with an X-ray yet.
Yeah, even I know that is strange and I was thinking about the power. Maybe one of the non-X version at 65W, but still what are people supposed to use them for at home? David will need another heat sink out the Windows or another weird contraption. I know some small businesses still have small servers somewhere.
Maybe my next build in 3 years will be Windows Part 12 and need it.
sdbardwick
Storage is cool
Actually, I think the opposite might be true.
Say you have a 65W intel cpu with tmax of 80C, and a 65W AMD cpu with a tmax of 95C. The same power must be dissipated 65W, but the HSF must be larger and/or with a faster fan to keep the intel below 80C, while the more tolerant AMD can get by with a less massive and/or slower fan, as it can tolerate 95C. (Plus, IIRC, and I might not, the greater the delta between the heat source and the dissipating medium, the greater the rate of heat flow from hot to cold, resulting in a more efficient heatsink. )
Mercutio
Fatwah on Western Digital
It's going to depend on which CPU you stick in that thing. The -Xs and X3Ds have much higher TDPs than the baseline non-suffixed versions of the CPUs. There are 65W 8c/16t Ryzen 7 series CPUs, but the faster parts all have 120W+ TDPs, high enough to be of concern.
13th gen Intel isn't exactly winning friends and influencing people in the TDP department, either, but the overall lesson here is that you want to be really careful in cooling contemporary CPUs.
13th gen Intel isn't exactly winning friends and influencing people in the TDP department, either, but the overall lesson here is that you want to be really careful in cooling contemporary CPUs.
ddrueding
Fixture
They mentioned in the video that the entire server has a 400W PSU and will not support the higher TDP chips. But for a barebones I can hang on the rack in the closet it seems quite competitive in the <2k USD segment.
Mercutio
Fatwah on Western Digital
It just seems like such a weird conglomeration of hardware. "I want something small, but not ITX small. I really wanna slide it in between my 48 port switch and the Drobo!" USB 3.2 is respectable for external IO, but not what I think of for rackmount gear. At least put U.2 on those front bays!
ddrueding
Fixture
20C lower temps on Ryzen 7000
The ZEN 4 X3D CPUs are a disaster due to EXPO settings overvolting and literally frying the CPUs. Ryzen BURnout ASUS.
ddrueding
Fixture
It seems that all the motherboards have firmware updates that better protect the CPUs now (just be sure to update to the latest BIOS). No word on whether it affects performance or not, but if it is giving it a lower power target I can't see how it couldn't.
Chewy509
Wotty wot wot.
From what I seen on Reddit, no, it affects all 7000' series chips. (there are reports of 7600X and 7700X - non-3D models seeing the same failure).
The root cause appears to be too much voltage sent to the IO Die (aka VSOC), where the IO Die on the 7000's should receive no more than 1.3V. (Remember the 7000 series chips are made up of 2-3 components, 1-2 CCD's with actual CPU cores and L2/L3 caches and an IO Die that has the memory controller, gpu, pci chipsets, etc).
Some motherboards, have been sending 1.4V+ when EXPO is enabled to assist in memory stability, where the 1.4V over time will kill the IO Die... (good dies will handle the extra, lesser quality but still in-spec dies will die quicker). IRC the expected voltage for the IO Die is 1.0-1.1V, and design docs says to not exceed 1.3V for significant periods of time.
Buildzoid has an excellent video on a Gigabyte board and the crap UEFI it has, and his measuring voltages shows that what the UEFI says the setting may be, may in fact be wrong. (remember the core chipset/cpu components of the UEFI are same amongst all vendors, so not one vendor is to blame). Steve from GN shows the same by on an Asus board.
Steve from GN managed to kill 3x CPUs in there own testing, and their reporting IMHO passes the sniff test...
I suspect, someone f**ked when sending out documentation, and not validating the UEFI modules for the new CPU where in fact doing what they were meant to be doing. Whilst the root problem may have originated with AMD (they supply the firmware), the motherboard vendors should have caught this in the QA stage. This appears to be a failure of QA/QC on the UEFI firmware side, and a failure of the board vendors in their QA on a new platform.
If I update the BIOS the RAM will crash down to slow speeds and performance will suck, is that right? I guess the question is whether to leave it alone until failure and then replace with Intel or not. I don't see VSOC listed in my BIOS; maybe it has another name?
Mercutio
Fatwah on Western Digital
Apparently you REALLY need to look at the language on the license agreement of your AM5 motherboard firmware updates. Asus's newest BIOS versions demand that you agree to void your warranty before you're allowed to download the latest version.
My BIOS shows 1.36V with EXPO actuated, so I guess that is excessive for the AMD CPU.
The RAM slows to 48000 with the EXPO deactivated. Voltage on the RAMs goes from 1.4 to 1.1.
It's not clear what the reduction on the CPU does to the RAM.
The RAM slows to 48000 with the EXPO deactivated. Voltage on the RAMs goes from 1.4 to 1.1.
It's not clear what the reduction on the CPU does to the RAM.
Is that because it is a BEta BIOS?
ddrueding
Fixture
It seems that that is the boilerplate language on all their beta BIOSes, they're in the process of removing that language from these BIOSes, and say they'll honor all warranty claims related to this issue. So that is something?
My MIS motherboard had a BEta BIOS in late April, but I was not aware that the AMD cluster impacted the regular X CPUs.
Today I anxiously proceeded to update that BIOS. Fortunately the computer is still working at DDR5-6000 speed on the EXPOs. From what I can see the voltage was reduced from 1.36 to 1.31. The RAMs are still 1.4V and the timings are unchanged.
The bad part is that AMD takes forever to boot now.
I guess with less voltage the RAMs are barely synchronizing with the IUD and CPU.
I can hardly imagine any "normal" person waiting that long in the 2020s. I assume Intel DDR5 boots faster and I know that MAC is right away.
Today I anxiously proceeded to update that BIOS. Fortunately the computer is still working at DDR5-6000 speed on the EXPOs. From what I can see the voltage was reduced from 1.36 to 1.31. The RAMs are still 1.4V and the timings are unchanged.
The bad part is that AMD takes forever to boot now.
I guess with less voltage the RAMs are barely synchronizing with the IUD and CPU. I can hardly imagine any "normal" person waiting that long in the 2020s. I assume Intel DDR5 boots faster and I know that MAC is right away.