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!