|
Mark Larson posted:You mean I should keep the same timings and just try to bump up the speed? Mark Larson posted:Also how do I know if I have the B die stuff? Sorry for the late reply. Munkeymon posted:Is it weird or is it just not what Intel does? It seems it'll require some degree of software support by the OS to get working well with some stuff given that it seems to work better with Linux than Windows right now but once they get that worked out it could be justifiable for those who are more CPU than memory bandwidth constrained. Otherwise I'd just stick with a 12C/24T 2920X or a 16C/32T 2950X Threadripper2 if you need more threads but don't want to deal with NUMA or bandwidth shenanigans right now. For the money ($649 and $899 MSRP's) they seem awfully impressive. PC LOAD LETTER fucked around with this message at 16:18 on Aug 16, 2018 |
#
?
Aug 16, 2018 16:09
|
|
|
|
| # ? Apr 27, 2024 01:41 |
|
|
Can someone briefly describe to me what happens when you're given more virtual threads in a workload than real cores and/or HT threads, and how much relative time it takes to solve that problem versus simply having enough cores for the entire workload?
|
|
#
?
Aug 17, 2018 20:38
|
|
|
Sidesaddle Cavalry posted:Can someone briefly describe to me what happens when you're given more virtual threads in a workload than real cores and/or HT threads, and how much relative time it takes to solve that problem versus simply having enough cores for the entire workload? Same thing that happened on single-core systems, the OS scheduler is responsible for timesharing them on the available cores. Switching thread contexts does incur some performance overhead, so you don't want to go totally nuclear because you'll end up increasing the amount of time lost to thread switching. One of the things SMT/HT does is allow one thread to keep running while the other is switched out, so non-SMT/non-HT processors will be hit more heavily by this overhead than SMT/HT.
|
|
#
?
Aug 17, 2018 20:51
|
|
|
Okay so two answers here, the first is for a thread over-commit situation where no HT is involved: Basically you get CPU wait time, where threads are waiting for available cycles on the core to execute. The resource scheduler in the OS/hypervisor is responsible for scheduling out that load and allowing for some amount of equatable share of access to the cores or shuffling the thread over to a different core where there is availability assuming all worker threads have the same priority. Stuff can get crowded out entirely if something has a higher priority that another thread. You can keep assigning physical cores so long as there are worker threads are there to use them but not all workloads/software is able to scale wide like this well because the result of one thread will impact the calculation on another so you have to keep them synchronized. Now throw HT in to that: For the sake of discussion, each core can perform the following tasks in a cycle: read from register, write to register, and Do Math. You can Do Math for a thread until you've read the register, and you probably can't do subsequent operations until that result is written to the register, so you end up with one one of the three available functions of a core being used each cycle (this is a gross simplification). With HT enabled, you can use these periods of downtime for other threads so in the same cycle that Thread A is Doing Math on the core, Thread B can be reading from the register to stage data so it can be Doing Math on that core in the subsequent cycle. This is your execution pipeline and you're more efficiently packing it with commands to get more work done in the same number of cycles. In reality, the holes are not that big and software/hardware has already been optimized to minimize these pipeline gaps through various means so with HT enable you will see a 0-15% increase in throughput on CPU-bound processes. Assuming you're on a HT-enabled platform, don't bother assigning more than 2x the number of worker threads to physical cores available assuming the cores will be saturated. BangersInMyKnickers fucked around with this message at 21:00 on Aug 17, 2018 |
|
#
?
Aug 17, 2018 20:57
|
|
|
TheJeffers posted:The 2990WX is crazy (like a fox, because it works well most of the time). Intel probably would never build something like this because Intel architectures seem to be about providing consistent performance and not chancing cases where even subpar code would perform badly for any reason except not scaling well with thread counts. AMD is apparently more than happy to provide a part that has an equal chance of scaling like mad or making programs and operating systems fall flat on their faces. If they tried to sell this thing into the server market it would be laughed out of the room but high-end desktops are a little more wild-West and AMD has nothing to lose by trying. TR is basically a 4-socket board with memory on two of the sockets. It's not particularly unusual a design, and the off-core penalty is lower because IF beats socket-to-socket QPI. Microsoft probably has NUMA gated to server SKUs so yeah, it will gut performance due to bad allocation. Epyc has a controller for each core complex so there's no starved cores in the datacenter.
|
|
#
?
Aug 17, 2018 22:44
|
|
|
Alright, I think I got most of that. I'm kinda coming to a conclusion on what hardware the developer of an application had in mind when writing it for the x86 architecture (surprise, the developer was Capcom and the workload was the PC port of Monster Hunter: World) and it definitely seems like the original game must have really taken advantage of the peculiarities of Jaguar's 8 cores on PS4 with not much effort in looking at the low-level on non-console architectures. The MT Framework engine for the port full-stop throws 32 different threads at PCs with some thread priorities way out of wack, like rendering threads designated higher than disk i/o threads  This is causing a bunch of issues involving hitching and crashes and possibly even network disconnections(!) because of the short amount of time the developers had (since this spring) to bring the game to Steam. Overtaxed consumer PC processors, especially ones lower on thread capacity and clock speed look like the biggest victims.basically i'm thinking of getting a 8c16t CFL refresh aka whiskey lake or ryzen or threadripper in NUMA mode for mhw
|
|
#
?
Aug 17, 2018 23:31
|
|
|
Sidesaddle Cavalry posted:basically i'm thinking of getting a 8c16t CFL refresh aka whiskey lake or ryzen or threadripper in NUMA mode for mhw Whiskey Lake is specifically about mobile chips, the desktop chips are still considered part of Coffee Lake. For the most part, Coffee Lake is the objectively correct choice if you're primarily gaming and the 8C16T version will be the best non-HEDT chips out for productivity as well. TR is great for productivity stuff but difficult to recommend on its merits for gaming - it's not any worse than an 1800X but it's not any better either (although you can stream on one set of cores while gaming on the other). Zen2 will definitely overtake Coffee Lake again next year in productivity, but gaming is harder to predict. Coffee Lake will probably retain the gaming crown, or at least tie up with Zen2, but anything's possible. Ryzen's not bad for gaming though, particularly if you're only doing 60 Hz. If you do go with Ryzen, get a 2000 series, they do better in gaming because of the lower latency. The 2600X and 2700X are the value leaders, XFR2 is worth it this time around. And you should really be buying decent RAM for all of these platforms now, like at minimum 3000 CAS-15 or 3200 CAS-16. Oh, also, dunno if it helps you but I hear volumetric lighting really tanks performance in MHW. Probably moreso on the GPU side, but might be worth turning it off and seeing if it helps performance any. Paul MaudDib fucked around with this message at 00:04 on Aug 18, 2018 |
|
#
?
Aug 18, 2018 00:01
|
|
|
Yes volumetric fog is bad for computers and also actually makes everything look worse, it's turned off on a lot of MH:W tests out there and I do it too. Biggest change as it turns out is maximum Level of Detail and its distance, which provide a great boost when reduced but also makes some models very, very simple. Anyways, duly noted on unofficial nomenclature and Ryzen/memory choices. It would be a complete platform change with memory for me and I've got a 144Hz 1440p monitor, so it's not going to be an easy change and nVidia might get involved, too. I can certainly wait until Zen 2 to find out how well the IPC gains and power efficiency come along with the 7nm process!
|
|
#
?
Aug 18, 2018 00:21
|
|
|
Is there a reason why my ryzen 1800x running on an asrock b450 itx/ac gaming motherboard is never able to hit the two core turboclock speed of 4ghz? It seems like no matter how much load you put on the chip it never clocks beyond 3.6 ghz.
|
|
#
?
Aug 18, 2018 15:32
|
|
|
CAPS LOCK BROKEN posted:Is there a reason why my ryzen 1800x running on an asrock b450 itx/ac gaming motherboard is never able to hit the two core turboclock speed of 4ghz? It seems like no matter how much load you put on the chip it never clocks beyond 3.6 ghz. How is your CPU temperature doing? Try running just 1 or 2 threads of prime 95.
|
|
#
?
Aug 18, 2018 16:30
|
|
|
kaschei posted:How is your CPU temperature doing? Try running just 1 or 2 threads of prime 95. I flattened and reinstalled windows 10 and that seems to have fixed it. Just dropped it straight in a Windows 10 install that's been on the ivy bridge chipset for more than 3 years now.
|
|
#
?
Aug 18, 2018 20:43
|
|
|
Yeah you pretty much always should format/reinstall windows if you switch mobos/platforms. Even if windows will actually boot up and work for the most part there are always so many weird issues that persist and just go away with a fresh reinstall. Its not even much of a time waster these days if you've got a SSD and install from a USB3.whatever flash drive. If you're willing to use stuff like NTlite to build your own install images you can speed it up even further and get pretty much everything set up in a half hour or so.
|
|
#
?
Aug 19, 2018 09:09
|
|
|
Some news: https://twitter.com/wccftechdotcom/status/1034124920328863744
|
|
#
?
Aug 27, 2018 19:03
|
|
|
I scrolled down to the comment section on WCCF the first time ever. What in the gently caress.
|
|
#
?
Aug 27, 2018 20:08
|
|
|
That seems to have caused some investor panic. 
|
|
#
?
Aug 27, 2018 20:45
|
|
|
Combat Pretzel posted:I scrolled down to the comment section on WCCF the first time ever. What in the gently caress. Really not a good scene in there.
|
|
#
?
Aug 27, 2018 20:49
|
|
|
https://www.anandtech.com/show/13277/globalfoundries-stops-all-7nm-development This is the actual what the gently caress.
|
|
#
?
Aug 27, 2018 21:10
|
|
|
Khorne posted:https://www.anandtech.com/show/13277/globalfoundries-stops-all-7nm-development This was kind of known though. https://www.anandtech.com/show/13122/amd-rome-epyc-cpus-to-be-fabbed-by-tsmc AMD “Rome” EPYC CPUs to Be Fabbed By TSMC - from the end of July.
|
|
#
?
Aug 27, 2018 21:17
|
|
|
Khorne posted:https://www.anandtech.com/show/13277/globalfoundries-stops-all-7nm-development quote:The CTO stressed that the decision was made not based on technical issues that the company faced, but on a careful consideration of business opportunities the company had with its 7LP platform as well as financial concerns. AMD had been saying they would use both TSMC and Glofo, but it looks pretty clear that AMD decided to dump Glofo altogether. The end of process improvements is gettin messy.
|
|
#
?
Aug 27, 2018 21:18
|
|
|
Khorne posted:https://www.anandtech.com/show/13277/globalfoundries-stops-all-7nm-development  HOLY poo poo
|
|
#
?
Aug 27, 2018 21:18
|
|
|
Yeah was going to post that, so it basically means AMD is going all in on TSMC, and GloFo will continue to fill out AMD orders for I guess semicustom parts? Lol, this basically means AMD is finally free of the GloFo albatross.
|
|
#
?
Aug 27, 2018 21:23
|
|
|
pixaal posted:This was kind of known though. It's probably not a big deal for AMD, because they also publicly said they had planned to use both fabs and go with whoever was ready. It's still a pretty big deal and huge surprise given the time and money invested into a 7nm process by GloFo. Process improvements cost billions of dollars. Khorne fucked around with this message at 21:26 on Aug 27, 2018 |
|
#
?
Aug 27, 2018 21:24
|
|
|
Khorne posted:I'd argue it wasn't, TSMC's process was supposed to be ready earlier. AMD fabbed Vega and Rome Epycs there. GloFo was still, publicly at least, supposed to be used for other projects. For GloFo oh yeah, I can't imagine it's actually a good idea to dump 7nm unless AMD has it locked up so tight they can't use it with anyone else.
|
|
#
?
Aug 27, 2018 21:31
|
|
|
Khorne posted:https://www.anandtech.com/show/13277/globalfoundries-stops-all-7nm-development 
|
|
#
?
Aug 27, 2018 21:37
|
|
|
Khorne posted:https://www.anandtech.com/show/13277/globalfoundries-stops-all-7nm-development This seems similar to some of the shady stuff that happens in the pharma industry where companies are acquired for their patents, all R&D is cut, and prices are jacked up. Like the comment by the CTO in the article "we'll just acquire IP" rather than spend on R&D
|
|
#
?
Aug 27, 2018 21:38
|
|
|
It is kjind of weird how this pretty much coincides with Anderson leaving, related? Did AMD receive samples form GloFo, the associated price, and give a hearty laugh? And Anderson wanted to still dual source?
|
|
#
?
Aug 27, 2018 21:40
|
|
|
So if they aren't going to develop their own 7nm process, does this still leave the door open for them to license it from Samsung, one wonders.
|
|
#
?
Aug 27, 2018 21:43
|
|
|
SwissArmyDruid posted:So if they aren't going to develop their own 7nm process, does this still leave the door open for them to license it from Samsung, one wonders. GlobalFoundries™; A subsidiary of Samsung©
|
|
#
?
Aug 27, 2018 21:44
|
|
|
Would not be the WEIRDEST thing to happen, but they'd have to buy it from the Abu Dhabis, first.
|
|
#
?
Aug 27, 2018 21:47
|
|
|
Why in the world does AMD still have a wafer agreement with GF, did GloFo find out that the secret ingredient for Ryzen is actually people or what
|
|
#
?
Aug 27, 2018 21:47
|
|
|
Paul MaudDib posted:Why in the world does AMD still have a wafer agreement with GF, did GloFo find out that the secret ingredient for Ryzen is actually people or what AMD will probably renegotiate the WSA for Polaris 12, Vega 10 and Semicustom. IIRC, it all terminates in 2020 anyway.
|
|
#
?
Aug 27, 2018 21:57
|
|
|
There is probably no path to profitability for GloFo, but the next gen consoles are no doubt going to be 7nm so what exactly will GloFo be producing? Chipsets?
|
|
#
?
Aug 27, 2018 22:00
|
|
|
Risky Bisquick posted:There is probably no path to profitability for GloFo, but the next gen consoles are no doubt going to be 7nm so what exactly will GloFo be producing? Chipsets? I don't think Semicustom will be entirely 7nm consoles, but yeah I guess? Speaking of chipsets, AMD just needs to buy ASMedia already.
|
|
#
?
Aug 27, 2018 22:07
|
|
|
They need to get the BoM down given the die sizes they will need on the graphics portion of the SoC, so 7nm seems... likely. Especially so given nothing has yet leaked about ps5/xbox two xXx
|
|
#
?
Aug 27, 2018 22:10
|
|
|
Risky Bisquick posted:They need to get the BoM down given the die sizes they will need on the graphics portion of the SoC, so 7nm seems... likely. Especially so given nothing has yet leaked about ps5/xbox two xXx They may have done the math out and figured that early 7nm is so expensive via quadruple imaging and associated processing and yields that it makes more sense to just eat the extra die area and heat costs.
|
|
#
?
Aug 27, 2018 22:22
|
|
|
EmpyreanFlux posted:AMD will probably renegotiate the WSA for Polaris 12, Vega 10 and Semicustom. IIRC, it all terminates in 2020 anyway. quote:Because of the strategy shift, GF will cut 5% of its staff as well as renegotiate its WSA and IP-related deals with AMD and IBM. In a bid to understand more what is going on, we sat down with Gary Patton, CTO of GlobalFoundries.
|
|
#
?
Aug 27, 2018 22:27
|
|
|
Risky Bisquick posted:They need to get the BoM down given the die sizes they will need on the graphics portion of the SoC, so 7nm seems... likely. Especially so given nothing has yet leaked about ps5/xbox two xXx Oh I'm not saying that the PS5 and XboX Triple X won't be 7nm completely, rather AMD does have other semicustom designs already on 14/12nm, such as the Subor Z+. Those'll likely continue to be produced until 2020 probably.
|
|
#
?
Aug 27, 2018 22:30
|
|
|
Navi HAS to be on 7nm. From AMD's standpoint, this is going to be non-negotiable, especially if the rumours are that the 20XX series of Nvidia cards is going to be shortened because they had to push back the release date so much because of 10XX series oversupply, because whatever comes after for Nvidia will most certainly also be on 7nm. And if whatever winds up in the PS5 and/or XBOX 2020 is an APU of the non-MCM variety, then they too, will be 7nm as well, because of how AMD plans on scaling Navi upwards into discrete GPUs. SwissArmyDruid fucked around with this message at 01:25 on Aug 28, 2018 |
|
#
?
Aug 27, 2018 22:36
|
|
|
EmpyreanFlux posted:GlobalFoundries™; A subsidiary of ^ seems more likely
|
|
#
?
Aug 27, 2018 22:48
|
|
|
|
| # ? Apr 27, 2024 01:41 |
|
|
Welp. Be interesting to see the new agreement they create with AMD until 2020.
|
|
#
?
Aug 28, 2018 00:58
|
|
