|
Anime Schoolgirl posted:the lower 16 core SKUs look like discount runt bins for cheap workstations. The Epyc boards are gonna be really really expensive and complicated. A threadripper + X399 board will be cheaper, unless you need the 128 PCI-Ex lanes (like for deep learning or trying to become rich in cosbycoins) or the 2TB of memory.
|
#
¿
Jun 20, 2017 23:26
|
|
|
|
| # ¿ May 2, 2024 14:03 |
|
|
Measly Twerp posted:Have you got a spare PSU you can try? Maybe it's failing to deliver power correctly :S This is good advice. PSUs and spinny HDs seem to be my primary fail points these days.
|
|
#
¿
Jun 21, 2017 23:25
|
|
|
Microcenter cut all their AMD motherboard bundle discounts today. They were offering $100(!) off if you bought a Ryzen 7+mobo for the last few months, but thats been cut to $50. The $50 discount on the Ryzen 5 was cut to $30, which is the same they offer for Intel bundles.
|
|
#
¿
Jun 22, 2017 02:57
|
|
|
AT has a long write up on Glofo's new 7nm DUV process and the future of AMD's chips in general. Pretty interesting if you wanna get real nerdy. http://www.anandtech.com/show/11558/globalfoundries-details-7-nm-plans-three-generations-700-mm-hvm-in-2018
|
|
#
¿
Jun 23, 2017 19:09
|
|
|
NewFatMike posted:I'm sad the R3 line are 4C/4T. Hopefully Raven Ridge will have 4C/8T SKUs. I think 4/8's are called R5's in this naming scheme, like the 1500X/1400. Thats if you even try making sense of modern computer product naming schemes. AT has their preview of the Pro models up, it basically sounds like these just have some enterprise security features enabled and a longer warranty. Roughly equivalent to the vPro line of i7/i5. http://www.anandtech.com/show/11591/amd-launches-ryzen-pro-cpus-enhanced-security-longer-warranty-better-quality
|
|
#
¿
Jun 29, 2017 16:51
|
|
|
GRINDCORE MEGGIDO posted:Is it really the best silicon? I'd have expected it to be just relabelled. All that stuffs probably going to THREADRIPPER / EPYC. It's likely the best silicon when it comes to voltage (since they are talking about extended warranties and reliability), not for clocks.
|
|
#
¿
Jun 29, 2017 19:35
|
|
|
its really hard to say if Epyc or ThreadRipper is a worse name to me. pretty neck and neck in complete terribleness.
|
|
#
¿
Jun 29, 2017 21:57
|
|
|
threadripper is a cool name for a sewing machine, its dumb for a $1000 computer chip. sounds like somethin a dude with those huge plastic dragon computer cases would have
|
|
#
¿
Jun 30, 2017 01:13
|
|
|
SamDabbers posted:Yes, that's threadripper's target market im a man of more REFINED tastes  (this came up when i GISed "bad computer cases", and that couldnt be more wrong)
|
|
#
¿
Jun 30, 2017 01:29
|
|
|
Well here's how Asus is doing it for LGA2066: To make Socket TR4 work, I imagine you would have to also put the sodimms on a daughter board. Probably wouldn't do regular dimms on a daughter board cause of weight issues.
|
|
#
¿
Jun 30, 2017 18:57
|
|
|
Ive also never benchmarked any one of the work systems ive been given over the years, and ive probably had 5x as many of those than home systems. I don't have any doubt that Ryzen has made AMD far more competitive in the custom build arena though.
|
|
#
¿
Jul 2, 2017 07:46
|
|
|
Not sure if this has been posted before but wccftech posted a leaked lineup for TR:
|
|
#
¿
Jul 6, 2017 20:31
|
|
|
It seems AMD planned from the start to use MCMs with Zen, so I imagine 2 and 4 way configs were always on the table and planned for. How they were targeted, branded, and priced was likely decided much further down the road, as it would be highly dependent on what sort of competition they were facing at roll out (something they would be unable to peg when design started out in 2012). Basically I doubt 2 way was a last second idea. That doesn't mean it will be a good consumer product, we will just have to wait and see on that.
|
|
#
¿
Jul 6, 2017 22:39
|
|
|
incoherent posted:Could they throw out a last min 18 core threadripper or has that left the station for this cycle? TR is two Ryzen dies on one package, so the best you will get is 16 cores. I doubt X399 boards will support the power delivery for 4 dies, so if you want more than 16 cores, you'll have to go with Epyc. E: Beaten Cygni fucked around with this message at 06:10 on Jul 7, 2017 |
|
#
¿
Jul 7, 2017 06:04
|
|
|
PC LOAD LETTER posted:My WAG is Navi will by default be a mid range targeted product as a single "sweet spot" (200-300mm I believe is the sweet spot for yields but that is old information) die and they'll scavenge dies or MCM packages as necessary for the low end versions. Its going to be the high end targeted version of Navi that could be really interesting. GPUs are theoretically more suited to multiple die packages too due to the parallel workload. Epyc's latency for going between the dies is fairly atrocious, and AMDs way of addressing that was essentially working around that reality (just like Intel did with the Pentium D). GPUs wouldn't deal with that quite as much. I mean heck, the concept you're describing for Navi is basically VSA-100.  The problem Navi faces is the same one AMD has faced for years: Nvidia isn't just sitting around waiting for them to catch up. Volta is already at the reticle limit of a 12nm process, so I imagine Nvidia is as fully aware of the manufacturing constraints and future performance needs as AMD is.
|
|
#
¿
Jul 7, 2017 22:29
|
|
|
Combat Pretzel posted:So about MCM latencies of TR/EPYC, I suppose it would have been better to have a memory controller as separate entity on the IF and have the CCX groups be autonomous? My understanding from the ServeTheHome coverage is that hitting infinity fabric is a bad thing, so adding that to every single memory call probably wouldn't be great for latency, I imagine. As it is, 2 socket Epyc has four options when stuff is in memory: Local to the core (great latency, great bandwidth), One hop on IF away through another core on the package (worse latency, same bandwidth), One hop on IF away through a core on the other package (worse latency, worse bandwidth), Two hops on IF away through a core on the other package (worst latency, worse bandwidth). An on package memory controller would basically cut the top and bottom cases off, and make all the calls the middle two. If NUMA aware OS didn't exist, that might be worth it. But with NUMA, and most calls being the first two options, the current solution on Epyc is probably much more efficient. But I'm just an armchair dork, that could all be wrong.
|
|
#
¿
Jul 9, 2017 23:10
|
|
|
Combat Pretzel posted:NUMA is nice for workloads that can wait a while. But on things like gaming where you need to get frames out as fast as possible, if for some reason there's no thread scheduling capacities for the CCX that handles the memory region most of the thread's poo poo is allocated in, there's a problem. Yeah, honestly, I don't expect TR single client gaming performance to be that notable. There are barely any games that even use more than 4 threads, let alone 32 threads, hence the R5 1600X and R7 1800X being basically identical in games. So I imagine a NUMA aware OS will just stick all the processes/data on one die+memory bank and call it a day... so its all back to clockspeeds as per usual. The rumored top of the line TR 1950X is 3.4ghz base, so i imagine gaming will be basically identical to other much cheaper mainstream Ryzen chips in that clock range. Multi-client or virtualized stuff, and of course rendering/content production, will probably be TR's strong suit. For just straight gaming though, the Coffee Lake stuff coming next month will probably be the hot ticket until Cannonlake/Zen2.
|
|
#
¿
Jul 10, 2017 06:38
|
|
|
TR will likely be dealing with NUMA, because we have 2 separate memory controllers each with a dual channel connection to their local DIMM banks, and a higher latency connection via IF between them. Would be advantageous for the OS to know that. Someone asked AT's editor about it: https://twitter.com/RyanSmithAT/status/870598439993720832
|
|
#
¿
Jul 10, 2017 18:52
|
|
|
incoherent posted:AMD Ryzen 7 1700X Processor - 299 on amazon prime day. It looks to be the lowest ever. This is the same price its been at Microcenter for a while. Plus they kick in a $50 motherboard credit. Thats if you have a local microcenter though.
|
|
#
¿
Jul 12, 2017 01:27
|
|
|
Paul MaudDib posted:Do they still? It's not showing on my local Microcenter anymore. I thought they rolled back some (perhaps all?) of the discounts. Still showing as available at $299 with the $50 motherboard kickback at the Microcenter in the LA area, with a couple board choices in stock. Dunno if thats around anywhere else though. Kinda kicking myself for not getting one when it was $299+$100 kickback a few weeks ago.
|
|
#
¿
Jul 12, 2017 02:12
|
|
|
Second Sun posted:That's the most childish bullshit. Really hope the big data centres all tell intel to go gently caress themselves. lmao gettin worked up about a tiny rear end joke in marketing slides
|
|
#
¿
Jul 12, 2017 16:50
|
|
|
NewFatMike posted:R3 hits July 27th, but still no APU information. I had thought for sure R3 was going to have integrated graphics, but it looks like they're just 4C/4T SKUs. The leaked Raven Ridge was a single CCX (4C/8T) with a 704SP graphics unit on-die. Just like Bristol Ridge, this will get cut down/clocked down to hit the mobile power power targets, but will all be the same die. When it actually shows up is another question. If its competitive in mobile, all of the dies will go there for a while cause it has tastier profit margins and a much bigger market.
|
|
#
¿
Jul 14, 2017 20:51
|
|
|
DRAM/Flash is also completely price fixed (still), so theres that.
|
|
#
¿
Jul 15, 2017 20:29
|
|
|
A theoretical 8 core TR would be 2/2+2/2. You could also do a comical 4 core with 1/1+1/1, which would join the Epyc 7251 8 core (1/1+1/1+1/1+1/1) in "this is for a very specific use case" territory, ha. Lots of cache/memory bandwidth, lots of PCIe, not a lot of raw proc power.
|
|
#
¿
Jul 16, 2017 05:11
|
|
|
I dunno, Zen2 could blow everyones labia off, but this is Keller-less AMD on an internal Glofo node were talking about. It's also not going to be on shelves until late 2018/2019, which means its theoretically gonna be facing 10nm Cannon Lake/Ice Lake from intel, a process which took them 3 years longer than they expected to develop. Hard to say what any of this stuff means yet.
|
|
#
¿
Jul 19, 2017 21:55
|
|
|
Seamonster posted:So is zen+ the die shrink? Or is that zen2? Zen2 is what AMD is now calling what used to be Zen+ on their slides. The 14nm+ version of Zen doesn't have a specific code name that I've seen, and I don't think they are expecting too much of a performance boost.
|
|
#
¿
Jul 19, 2017 23:46
|
|
|
I'm mostly sure thats just a Noctua NH-14S with a wider contact plate...? I've got a Bequiet Dark Rock 3 that is def a much bigger heatsink than that.
|
|
#
¿
Jul 20, 2017 22:08
|
|
|
Anarchist Mae posted:AMD Ryzen 5 1600; 96% of the performance at 85% of the watts for 50% off the price of an Intel Core i7-7800X. I like that the best looking CPU in that comparison is the one not in the title. 7800x is gonna be essentially redundant next month with Coffee Lake, too. Woof. SKL-X is a mess.
|
|
#
¿
Jul 21, 2017 13:40
|
|
|
FaustianQ posted:Intel is promising a lot with Coffee Lake but I'm just assuming it's going to be Skylake with 50% more cores and the accompanying thermal and power issues, and I feel it's success depends on if it's compatible with 1151. They've said a "30% performance increase" over Kaby, but it has 33% more cores so yeah, it's pretty much Kaby Lake with 2 extra cores at 95W. If it is close to 7700K clocks though, thats pretty drat formidable. The leaks had the 8700k at 3.7ghz base.
|
|
#
¿
Jul 21, 2017 19:09
|
|
|
FaustianQ posted:Fair enough, we're still in the kind of mid-late 2001 phase but AMD seems poised to deliver Tbred again while Intel futzes about with at least two distinct uarchs, one of which massively underperforms. Also AMD was kind enough to pencil in everything ahead for us this time. I wouldn't say SLX massively underperforms, I would say its not the victory lap that some people were expecting. It's slightly faster than Broadwell-E at much cheaper prices per core, and with a more modern platform. Those expecting a zero-compromise, mega overclocking many core Skylake experience might be disappointed, but that was never really in the cards. I thought we all learned not to listen to computer part manufacturer marketing, but I guess not. Ryzen competes favorable in a price/performance view with some of the lineup, and thats a good thing! But I would be careful putting too much faith in AMD. This is AMD were talking about, afterall. Take it for what it is right now.
|
|
#
¿
Jul 24, 2017 05:26
|
|
|
Shaocaholica posted:Anyone making or will make a proper dual socket Ryzen workstation? Regular Ryzen won't ever be dual socket, only Epyc supports dual socket. 16 core Ryzen workstation parts are due very soon (Threadripper).
|
|
#
¿
Jul 24, 2017 07:06
|
|
|
Ihmemies posted:Probably the AIO takes a ton of room from the box. Good point.
|
|
#
¿
Jul 24, 2017 16:37
|
|
|
STH has a big article on memory bandwidth/latencies on Epyc. https://www.servethehome.com/amd-epyc-infinity-fabric-latency-ddr4-2400-v-2666-a-snapshot/  Pretty much as we expected. Latency spikes across NUMA zones are not ideal, but actually better than i was expecting honestly. Bandwidth shows pretty clearly that the game for Epyc (and Threadripper to a lesser extent) will be making sure that the OS and scheduler stash the right stuff in the right ram bank, because you really don't want to be reaching across IF: 
|
|
#
¿
Jul 24, 2017 18:50
|
|
|
Combat Pretzel posted:Oooooh wait, green/yellow and grey/blue are different sockets? In that case, woohoo! Yea, yellow is stuff in the local memory of that die, green is other dies in that socket, blue is the "partner pair" die in the other socket, grey are the other dies in the other socket. So TR/1 socket Epyc won't have those bigger hits (only yellow and green). And theoretically, a NUMA aware OS and a good scheduler will keep stuff local to avoid the blue/grey hits.
|
|
#
¿
Jul 24, 2017 22:31
|
|
|
Shaocaholica posted:So is the gaming community like already on this train? The best case (and most likely) is that Threadripper equals an 1800X in gaming for twice the price. The worst case is that NUMA doesnt work well at the OS level or games just don't play nice with it, and it ends up actually slower than an 1800X clock for clock. I don't think thats likely at all, but we will see soon. HEDT (or SHED, lol AMD) is not really a gaming platform so much, no matter how desperate AMD and Intel are to make gamers think they need $1000 CPUs to play overwatch. For content creator types though, TR could be a great value.
|
|
#
¿
Jul 25, 2017 03:01
|
|
|
PC LOAD LETTER posted:The 1600 is a good enough chip that why would you want to see 6C Coffeelake everywhere? Because we are consumers trying to get the best goods for as cheap as possible from huge billion dollar multinational corporations?
|
|
#
¿
Jul 27, 2017 05:17
|
|
|
i mean all the good motherboards let you turn the LEDs off and they cost pennies, thats why they are on everything. i dunno if its too worth getting worked up over, especially when unironically buying a cpu named _-~+=ThReAdRiPpEr=+~-_
|
|
#
¿
Jul 27, 2017 07:15
|
|
|
There isn't any iGP on the Zeppelin die and R3 is just that same die again. We wont see Zen with iGP until Raven Ridge, which seeing as how Bulldozer based Bristol Ridge just now showed up today, might be further away than we thought. Threadripper having 4 dies is so fuckin' weird... the chip he was delidding was clearly an engineering sample and non retail, so maybe the retail units will have 2 dies removed in the diagonal config? https://twitter.com/IanCutress/status/890593099952664576
|
|
#
¿
Jul 27, 2017 17:08
|
|
|
AVeryLargeRadish posted:der8auer said in the video that AMD has confirmed to him that retail TR will have four dies. i dont remember him saying that in the video? he said that AMD confirmed that 2 dies were disabled in his version, but he didnt say anything about retail that i remember. if Ian at AT thinks 2 dies is still a possibility, thats probably a better source than a youtuber honestly. either way, the whole thing is weird.
|
|
#
¿
Jul 27, 2017 17:43
|
|
|
|
| # ¿ May 2, 2024 14:03 |
|
|
Combat Pretzel posted:
Based on ServeTheHome's Epyc tests, having to hit IF for the memory calls would have a ~40% hit on latency and ~50% hit on bandwidth vs each cores integrated controller. X399 only supports 4 channel memory, so only 2 of the dies can have active memory controllers. The 2 extra cores on the engineering sample really are completely vestigial. Im putting my chips in the "they just reused Epycs as engineering samples and production models will only have 2 dies" pile, because its the only thing that makes sense to me. At least weve got somethin weird to talk about!
|
|
#
¿
Jul 27, 2017 18:57
|
|