Message boards :
Number crunching :
Ryzen and Threadripper
Message board moderation
Previous · 1 . . . 25 · 26 · 27 · 28 · 29 · 30 · 31 . . . 69 · Next
Author | Message |
---|---|
Ian&Steve C. Send message Joined: 28 Sep 99 Posts: 4267 Credit: 1,282,604,591 RAC: 6,640 |
Some people just like having cool new hardware. Seti@Home classic workunits: 29,492 CPU time: 134,419 hours |
Tom M Send message Joined: 28 Nov 02 Posts: 5126 Credit: 276,046,078 RAC: 462 |
This may be obvious, but if you have so many PCIe lanes, memory lanes, and cores, wouldn't a server chock full of GPUs be the better way if you wanted to just purely crunch Seti/Boinc? I'm sure the cost is what prohibits most people from going this route, but I'm wondering if there are other reasons besides this. Inspite of the very high # of PCIe lanes, I haven't seen a EPYC motherboard with more than 7 or 8 slots. So either you would need lots of 1 to 4 expanders to run a very high count gpu system. Or the commercial gpus are running multiple gpus / card. Either way, I think you could sink $25,000 into the new Rome system and still have room for more gpus..... Tom A proud member of the OFA (Old Farts Association). |
Jord Send message Joined: 9 Jun 99 Posts: 15184 Credit: 4,362,181 RAC: 3 |
Still looking into which motherboard to get, I passed this one: the Gigabyte Aorus X570 Extreme, sitting at an eye watering €799,- And why pay that price? Just so I have 3 PCIe slots? The lower budget Gigabyte Aorus X570 Pro also has three PCIe slots and only costs €269,- while the even cheaper UD only goes for €179,-! So... what do you get for the 800 euros? Direct 16 Phases Infineon Digital VRM, Fins-Array Heatsink, NanoCarbon Baseplate, Triple PCIe 4.0 M.2 with Thermal Guards, Intel WiFi 6 802.11ax, ESS SABRE HiFi 9218, AQUANTIA® 10GbE LAN+1GbE LAN, RGB FAN COMMANDER, RGB Fusion 2.0 The Wifi6 is done with two coax connectors on the back, going out to aerials you can put in your living room. Of course, since your router isn't using a gigabit wifi connection yet, you don't actually use it. Or any of your other wifi devices in your house. And wifi6 is found on cheaper Gigabyte motherboards as well, such as the Aorus Ultra (which at €324,- isn't exactly cheap either) . It seems the only reason to pay so much money is because of the audio. As this Sabre DAC is only found on this motherboard and the MSI Godlike. Which at €777,- is about as eye watering. Who pays that much for a motherboard? |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
Only the ones that want the bragging rights to the best kit available. Nobody needs PCI Gen. 4 right now. The Ryzen 3000 cpus still work on the old X370 and X470 motherboards. Any of the middle-tier to upper tier X470 motherboards have entirely adequate power delivery for even the upcoming 3950X 16 core cpu. Look at any of buildzoid's VRM analyses. All the new X570 motherboards have overkill VRM designs that serve no purpose other than "mine is bigger than yours" bragging rights. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Jord Send message Joined: 9 Jun 99 Posts: 15184 Credit: 4,362,181 RAC: 3 |
Yes, you keep saying that. However, it wasn't what I asked. And while some of the features on it aren't used (at their full potential) yet, they will be in the near future when hardware becomes available for it. It's like saying you don't need the new AMD CPUs because the older generation still works perfectly fine. The first PCIe 4.0 SSD has arrived: https://www.rockpapershotgun.com/2019/07/18/gigabyte-aorus-nvme-gen-4-ssd-review/ As for GPUs at PCIe 4.0, you're right there, all cards out there don't even max out at PCIe 3.0: But considering graphics cards of today don’t even max out PCIE 3.0, and many of AMD’s flagship X470 motherboards from 2018 actually ran PCIE 2.0, is there really much point to the new standard? https://www.digitaltrends.com/computing/does-pcie-4-matter/ A 2080 Ti pushes the limits of what a PCIe 3.0 x8 port will allow, just – TechPowerUp’s testing proves that. So, on a board with PCIe 3.0, running it in an x16 configuration makes the most sense. But that means the graphics card alone is hogging the all 16 lanes of PCIe. On an AMD Ryzen CPU, be it first, second, or third-generation, there are only 24 lanes to go around. Four of those are set aside for PCIe and SATA drives, four more provide the link with the chipset (which has a number of lanes of its own and shares them between storage drives and USB ports), and 16 dedicated for graphics cards. .. With PCIe 4.0, however, there’s no need to run a graphics card with the full 16 lanes, because x8 mode will give more than enough for even a 2080 Ti. That means unlocking an extra eight lanes (or more in the case of lower-power GPUs) that can be used for a variety of additional purposes. .. More likely and more useful, is that those lanes could be used for a variety of add-in cards. If you want the fastest storage possible, PCIe 4.0 NVME drives are at the forefront and some have already been announced with sustained read/write speeds as high as five GB/s. That’s 10 times the speed of your average SATA III SSD. .. Faster networking is also a more viable option with PCIe 4.0. 10-gigabit Ethernet could be possible on just one dedicated lane, making it far more accessible for the average person. Additional USB or Thunderbolt 3 ports could also benefit from those free lanes, allowing for more accessories and peripherals. Those Thunderbolt 3 ports could come in extra handy in the laptop space. With additional bandwidth and dedicated lanes that don’t have to rely on the chipset’s more limited connection to the CPU, we could see external graphics cards compete directly with onboard GPUs for the first time. That would make external enclosures like Razer’s Core X Chroma, a real alternative for laptop owners who want gaming power at home and a more portable, longer battery-life device during their work day. And then there's... https://www.digitaltrends.com/computing/pci-express-6-coming-2021/ 😱 |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
If you have legitimate storage needs, like a RAID array or massive database that is constantly being read and written, then certainly update to X570 for the PCIe Gen. 4 lanes. But I posted on the assumption we were speaking of appropriate SETI or BOINC crunching hardware. The highest production comes from multiple gpus and the X370/X470/X570 platforms are limited to native 3 cards. If you want to run multiple cards, you need to purchase an HEDT platform like X99, X299 or X399 which gives you normally four double wide slot positions. Or go the mining PCIe splitter route which still needs PCIe lanes. The HEDT platforms always provide more than the consumer platforms. You don't need to spend €799 for one. There are much more sensible and economic choices. You point out another flaw in your thinking. The X570 and PCIe Gen 4 chipsets are in reality a interim solution. Both PCIe Gen. 5 and DDR5 standards are already of out of committee and approved and will likely be seen in motherboards at the end of next year or early 2021. If your argument is that a X570 purchase "future proofs" you, yes for about a year. Since these new X570 motherboards are not cheap as the previous AMD motherboards, maybe waiting out one development cycle would be a smart choice. An X470 motherboard still has a lot of functional life left in it. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Bill Send message Joined: 30 Nov 05 Posts: 282 Credit: 6,916,194 RAC: 60 |
The highest production comes from multiple gpus and the X370/X470/X570 platforms are limited to native 3 cards.Are you saying that strictly because of the number of x16 PCIe slots, or something else? The reason I ask is because the MSI Meg Godlike has four x16 slots...I assumed all four could be fitted with GPUs. Seti@home classic: 1,456 results, 1.613 years CPU time |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
The highest production comes from multiple gpus and the X370/X470/X570 platforms are limited to native 3 cards.Are you saying that strictly because of the number of x16 PCIe slots, or something else? The reason I ask is because the MSI Meg Godlike has four x16 slots...I assumed all four could be fitted with GPUs. OK, that mobo is the exception to the rule. 95% of all X370/X470/X570 boards only have three usable slots for double wide cards. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Bill Send message Joined: 30 Nov 05 Posts: 282 Credit: 6,916,194 RAC: 60 |
I wasn’t trying to be snarky. Of course, that’s a pretty penny just for one more slot, Seti@home classic: 1,456 results, 1.613 years CPU time |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
I wasn’t trying to be snarky. Of course, that’s a pretty penny just for one more slot, I know. I was just commenting that the mobo has all the bells and whistles plus the kitchen sink and the vendor is pricing it accordingly. One of the common comments about this recent launch is that all the mobo vendors are pricing their products more like the Intel mobo mainstream products that have been priced fairly high for a long while now. AMD doesn't feel that they need to be seen as the "value leader" anymore and more like the premium platform now. So all the X570 boards are a lot more expensive than previous generations. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
ML1 Send message Joined: 25 Nov 01 Posts: 21088 Credit: 7,508,002 RAC: 20 |
Tom's Hardware has been getting all excited about the AMD EPYC: AMD Unveils 7nm EPYC Rome Processors, up to 64 Cores and 128 Threads... AMD's launch of its 7nm processors marked the first time in the company's history that it had wrested the process lead from Intel, an advantage that can't be overstated... AMD EPYC Rome SKU List and Block Diagram Posted Gigabyte has put up the product page for the brand's latest MZ32-AR0 motherboard which contains the SKU list and block diagram for AMD's EPYC (Rome) processors that are set to debut later today... Is there any need for bitmining motherboards/mainboards with capability like that! Spectacular stuff!! Happy fast crunchin', Martin See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
Interesting post by The Stilt over at OCN. Seems they are detuning the SMU in the later BIOS' to boost more and pull back on the temp inflection points. The original limits for Ryzen 3000 SKUs were: Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
Just a FYI datum. I changed the kernel on all my hosts to the 5.0.0.23 kernel today via the HWE stack. WOW! What a difference that made in the Threadripper host. I never had that host overcommitted on cpu usage yet always had a 4-9 minute difference in cpu_time versus run_time. Even reducing the cpu count down to only a quarter of the cores never closed the gap. Changing to the 5.0 kernel now has the cpu_time equaling the run_time for the most part even at 65% core usage which is where I like to run all my AMD hosts. The difference is all the improvements in the cpu thread scheduler that is in this kernel. It has the fixes for Threadripper in it that was always lacking in the stock kernels that came with Ubuntu 18.04. So if you are already up on the latest Ubuntu 18.04.3 LTS release lsb_release -a Then go ahead and update to both the 5.0 kernel and the newer xorg.xserver platform which is also speedier. sudo apt install --install-recommends linux-generic-hwe-18.04 xserver-xorg-hwe-18.04 You can read about it here: https://itsfoss.com/ubuntu-hwe-kernel/ Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
jsm Send message Joined: 1 Oct 16 Posts: 124 Credit: 51,135,572 RAC: 298 |
And if you are on 19.04? jsm |
Ian&Steve C. Send message Joined: 28 Sep 99 Posts: 4267 Credit: 1,282,604,591 RAC: 6,640 |
And if you are on 19.04? you already have the new kernel Seti@Home classic workunits: 29,492 CPU time: 134,419 hours |
Tom M Send message Joined: 28 Nov 02 Posts: 5126 Credit: 276,046,078 RAC: 462 |
Just a FYI datum. I changed the kernel on all my hosts to the 5.0.0.23 kernel today via the HWE stack. WOW! What a difference that made in the Threadripper host. I never had that host overcommitted on cpu usage yet always had a 4-9 minute difference in cpu_time versus run_time. Even reducing the cpu count down to only a quarter of the cores never closed the gap. Changing to the 5.0 kernel now has the cpu_time equaling the run_time for the most part even at 65% core usage which is where I like to run all my AMD hosts. I know you were talking about the Threadripper. Do you have any opinion on the Am4 socket series on this? Tom A proud member of the OFA (Old Farts Association). |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
Actually its helped all my systems since 4 of 5 are AMD based. Saw a significant change in the 3900X system too similar to the Theadripper system. Probably because to the Linux scheduler a dual chiplet 3900X looks a lot like a dual die Threadripper. It has taken a while for the Linux kernels to catch up to AMD Ryzen and Threadripper since they are new to the ecosystem. It's not like Intel where each iteration of design was just more of the same just numbered differently. The Intel microcode hasn't really changed in ten years. Ryzen and Threadripper are only a couple years old now. The only changes for Intel recently have been the huge amount of security mitigations necessary for Intel cpus. You have to remember the kernels packaged into Ubuntu 18.04 were already locked in a year before Ryzen ever appeared on the scene. Only the more recent kernels have had code to handle Ryzen properly. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Jord Send message Joined: 9 Jun 99 Posts: 15184 Credit: 4,362,181 RAC: 3 |
Overclocking your Ryzen gen 3 chip explained by Anthony of LTT: https://www.youtube.com/watch?v=7w1EGPZUESU |
ML1 Send message Joined: 25 Nov 01 Posts: 21088 Credit: 7,508,002 RAC: 20 |
AMD storms ahead: Alleged 32-Core AMD Threadripper 3000-Series Castle Peak CPU Benchmarked ... The obscure 32-core, 64-thread Threadripper processor appeared in the Geekbench 4 database ... very close to the final product. Sadly, Geekbench 4 didn't register the processor's boost clocks so, for now, we only know that it has a 3.6 GHz base clock, which is mighty impressive for a 32-core chip. [64 threads.] The chip's cache configuration consists of 2MB of L1 cache, 16MB of L2 cache and 128MB of L3 cache... Way to go!!! So why have we been expensively stagnating with Intel for far far so long?... Happy fast crunchin'! Martin See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
If you want to see all the reported core clocks for all 64 threads, all you have to do is take the geekbench4 url and add .gb4 to it. Looks like all the cores were running at around 4150 - 4200Mhz. https://browser.geekbench.com/v4/cpu/14281644.gb4 The other entry is here: https://browser.geekbench.com/v4/cpu/14281648.gb4 and has ever better core clocks. "processor_frequency": { "minimum": 4112, "maximum": 4363, "median": 4321, "mean": 4279, "stddev": 79.18823125532559, "processor_frequency": { "minimum": 4154, "maximum": 4201, "median": 4192, "mean": 4187, "stddev": 13.40320477794806, "frequencies": [ 4187.0395800432789, 4185.6576932304688, 4187.5578694325313, 4165.6323748615914, 4162.6909499185813, 4174.5110948277352, 4190.6242911297095, 4192.8529550955091, 4195.6077944155531, 4193.9224443924677, 4187.0343206597527, 4181.1459751787106, 4190.6664387502869, 4193.0220816702586, 4201.0365721658882, 4199.2918398227157, 4199.9581600168094, 4198.1207364410557, 4197.6636140137489, 4198.6964223243676, 4197.7783258210584, 4198.7708182405013, 4201.8144947641604, 4199.4898795653489, 4200.3203332298926, 4198.5659713953373, 4200.4920036283847, 4194.4758668943477, 4199.4315061592843, 4199.219541454464, 4199.4350332072418, 4192.3934638405844, 4169.8666918637809, 4167.0357965876474, 4178.3121449035307, 4175.3440754915564, 4179.4320452769571, 4180.251089289829, 4193.3039310043832, 4199.2689156803171, 4198.7371458286489, 4196.8956821707679, 4196.8939207781705, 4195.3432361333835, 4163.1945624016234, 4184.6628090771019, 4174.0608415282322, 4178.9262432890619, 4164.9892172594145, 4180.5925688963744, 4178.0689650446875, 4168.5030659548474, 4157.7228850256151, 4175.3144387550674, 4170.6002377909426, 4171.4699404918947, 4185.8192313781055, 4184.4589862763132, 4168.1270741902854, 4159.0031834258061, 4187.2205108819371, 4178.1108604894516, 4183.7979245684664, 4179.1012341303849, 4161.3112324984095, 4190.3866997796604, 4175.312695430729, 4164.0629128277988, 4165.5801445123061, 4177.7996323118541, 4184.8083338951437, 4160.0589330588691, 4188.3116869518462, 4162.4097355433778, 4182.6336965267528, 4171.0375664498415, 4154.8342181676016, 4184.432721895655, 4192.4038338023793, 4200.1384449634224, 4200.9128583641223, 4199.3874185597042, 4200.4104893154781, 4190.0243063310008, 4198.6576387689929, 4200.5361144242834, 4200.5272921909882, 4197.3845257543117, 4198.2074493832615, 4198.9032212851807, 4200.9905095423392, 4200.5378788753897, 4200.5431722376015, 4200.318745388312, 4199.7867096321643, 4200.3557953379996, 4199.0975215624703, 4200.0516522352191, 4199.8061117510442, 4199.3519728009651, 4199.5782363577218, 4198.4970472180185, 4198.6664531504812, 4198.4653181783842, 4197.8012336918046 No worries about the core clocks being inadequate for the upcoming Castle Peak Threadrippers using Zen 2 chiplets. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.