|
ConanTheLibrarian posted:I, too, look forward to buying far more cores than I need. I'm glad I didn't get a 8700k when they came out. The jump from 4 to 6 cores looked great at the time, but kind of anaemic now. Spin up VMs, stream while playing games, compile vast quantities of code. Fun fact: I am in the process of adding OpenJDK to a project I contribute too called BuildRoot. I can compile Java in 5 minutes with a 2970wx. It takes almost 20 on my old 6800K.
|
#
?
Jan 25, 2019 19:24
|
|
|
|
| # ? May 21, 2024 02:19 |
|
|
dead comedy forums posted:One thing that I am curious about is with AMD pushing ALL THE CORES for the consumer market, how software development will work from here? I am a student and programming stuff to use multiple cores is waaaaaay beyond my level, but I am fascinated by the implications of software becoming efficient enough (one day) to properly use many cores. Some tasks simply do not scale, however just google multithreading in language and click on a stack overflow link and you are halfway to being a professional.
|
|
#
?
Jan 25, 2019 19:47
|
|
|
dead comedy forums posted:One thing that I am curious about is with AMD pushing ALL THE CORES for the consumer market, how software development will work from here? I am a student and programming stuff to use multiple cores is waaaaaay beyond my level, but I am fascinated by the implications of software becoming efficient enough (one day) to properly use many cores. AMD has sold eight cores since 2010. Any spec that isn't matched by Intel's mainstream line is just going to be niche. Zen2 (and specifically Epyc) will allow AMD to actually enjoy a profit margin above and beyond their usual priority of keeping the lights on, but they're never going to command influence on what people want from consumer CPU. To think otherwise is to ignore just how friggin' enormous Intel's sales really are. Craptacular! fucked around with this message at 19:51 on Jan 25, 2019 |
|
#
?
Jan 25, 2019 19:48
|
|
|
dead comedy forums posted:One thing that I am curious about is with AMD pushing ALL THE CORES for the consumer market, how software development will work from here? I am a student and programming stuff to use multiple cores is waaaaaay beyond my level, but I am fascinated by the implications of software becoming efficient enough (one day) to properly use many cores. You can do more and heavier tasks realtime. The major point for having a 8+ core CPU now is that you can play Fortnite or whatever on cores 1-4, and do a high efficiency real time encode of your flailing and failures for upload to twitch.
|
|
#
?
Jan 25, 2019 19:59
|
|
|
Or you could just have a basic 4-core CPU for streaming because games will slowly require more power while encoding will likely remain the same. (Don’t see many hours of 4K streams taking off with these capped residential connections.)
|
|
#
?
Jan 25, 2019 22:05
|
|
|
These are gonna be much more expensive than people are thinking it they perform in the neighborhood of AMD's cinebench comparison. They are also gonna use much more power than Zen 1 (but still probably less than CL-R). AMD's cinebench comparison machine was at 133 watts (vs 179 watts for the 9900k)
|
|
#
?
Jan 25, 2019 23:13
|
|
|
.
sincx fucked around with this message at 05:50 on Mar 23, 2021 |
|
#
?
Jan 25, 2019 23:27
|
|
|
sincx posted:I'll pay up to $450 for a 12 core that can boost to 4.8+ GHz and $600 for a 16 core that can do the same. I don't think AMD will charge more than that. Going by AMD's cinebench numbers, I would bet you are gonna be very disappointed.
|
|
#
?
Jan 25, 2019 23:35
|
|
|
Risky Bisquick posted:Some tasks simply do not scale, however just google multithreading in language and click on a stack overflow link and you are halfway to being a professional. Read two SO pages and now your understanding of multithreading is equal to the average professional! 
|
|
#
?
Jan 26, 2019 01:42
|
|
|
isndl posted:Read two SO pages and now your understanding of multithreading is equal to the average professional! A creeping sense of dread and a looming pit of existential despair as all your hard work refuses to do what you thought it would do, what you told it to do, what you pleaded with it to do, and what the conclave of duckies were unable to fix. Sounds about right for the average professional's multi-threading experience. That and the word gently caress in a variety of fonts, colors, and languages.
|
|
#
?
Jan 26, 2019 01:49
|
|
|
Cygni posted:These are gonna be much more expensive than people are thinking it they perform in the neighborhood of AMD's cinebench comparison. They are also gonna use much more power than Zen 1 (but still probably less than CL-R). AMD's cinebench comparison machine was at 133 watts (vs 179 watts for the 9900k) 133 watts puts it in the 65watt TDP category. For an 8 core that gets 2000+ Cinebench, which is what a full-turbo 9900k gets at 95w top, if they price it at $300 then thats a steal for anyone whose been paying attention over the last few years
|
|
#
?
Jan 26, 2019 01:53
|
|
|
https://www.anandtech.com/show/13405/intel-10nm-cannon-lake-and-core-i3-8121u-deep-dive-review/7 tl;dr version: Intel's 10nm while interesting from a process design PoV, in its current iteration is a hot piece of garbage.
|
|
#
?
Jan 26, 2019 02:57
|
|
|
dead comedy forums posted:One thing that I am curious about is with AMD pushing ALL THE CORES for the consumer market, how software development will work from here? I am a student and programming stuff to use multiple cores is waaaaaay beyond my level, but I am fascinated by the implications of software becoming efficient enough (one day) to properly use many cores. The other real problem with parallel programming comes from complex applications with multiple developers and inconsistent practices. Things will do things you don't expect and cause rare race conditions and stuff. If you are coding it, provided you are aware of the gotchas, parallel stuff isn't bad. There are already some solutions to this that are commonly used. Pseudo-functional or encapsulated stuff, schedulers, and message passing instead of directly mutating state all go a long way toward easing problems in complex applications. The good news is chrome will use 16 cores and 64gb of RAM no problem. Just open 16 tabs in our javascript-riddled world. Jokes aside, that's where the practical benefit of more cores comes from. Individual applications can be more aggressive about dominating a single core, and things that can work well parallel don't have to be shy. Nothing really changes with more cores available. The industry has already been shifting toward parallel when reasonable for some time. Khorne fucked around with this message at 03:31 on Jan 26, 2019 |
|
#
?
Jan 26, 2019 03:02
|
|
|
Lube banjo posted:if they price it at $300 then thats a steal for anyone whose been paying attention over the last few years Which is why I dont think its gonna be anywhere close to $300. I'm expecting $450, if not straight price parity with the 9900k at ~$500. Guess it also depends on what Intel does, cause it sounds like Zen2 isnt gonna show until mid 2019.
|
|
#
?
Jan 26, 2019 03:23
|
|
|
If AMD priced their CPUs so that they had no value advantage over Intel no one would buy them, that's like their main selling point. Yeah 7nm is more expensive but no way is AMD going to come out with a new CPU lineup 2.5x more expensive than their previous one, all of the excitement over Zen 2 desktop would disappear in an instant.
MaxxBot fucked around with this message at 04:19 on Jan 26, 2019 |
|
#
?
Jan 26, 2019 03:53
|
|
|
MaxxBot posted:If AMD priced their CPUs so that they had no value advantage over Intel no one would buy them, that's like their main selling point. Yeah 7nm is more expensive but no way is AMD going to come out with a new CPU lineup 2.5x more expensive than their previous one, all of the excitement over Zen 2 desktop would disappear in an instant. They have to steal market and mindshare now while they're demonstrably better than Intel and while people are sick of Intel's bullshit *lake refreshes. Leaving some cash on the table to get that is a good move going forward most likely.
|
|
#
?
Jan 26, 2019 04:24
|
|
|
It's really not, especially not in current year. If you have the best product, you price that motherfucker sky high and sell off the halo effect.
|
|
#
?
Jan 26, 2019 04:36
|
|
|
Yeah, prices are set by the top. If Intel is selling [power] for $200, you do not sell [power] for $140. AMD has been cheaper because there are compromises. If there's no compromises in performance, there's precious little compromise on price.
|
|
#
?
Jan 26, 2019 04:43
|
|
|
K8.0 posted:It's really not, especially not in current year. If you have the best product, you price that motherfucker sky high and sell off the halo effect. I mean if you have a product like the 2080 Ti which really does have no equal then sure but that's not really comparable to merely matching a competing product (9900k) which will have been out for 8 months at that point.
|
|
#
?
Jan 26, 2019 04:48
|
|
|
MaxxBot posted:I mean if you have a product like the 2080 Ti which really does have no equal then sure but that's not really comparable to merely matching a competing product (9900k) which will have been out for 8 months at that point. Most consumers aren't as desperate to upgrade CPUs compared to GPUs anyway.
|
|
#
?
Jan 26, 2019 04:56
|
|
|
I don't know about that, the chiplet strategy should help towards increasing sales volume to gain market share. They can still have a halo product and have pricing that encourages lots of sales. Production costs might be higher because there's now two to three dies in a package, however they've already been packaging Threadripper and Epyc processors like this for over a year, I'd expect them to have done what is necessary to streamline that process by now. The 7nm chiplet should be shared amongst all of their upcoming products, with the possible exception of the next generation of APUs. Using these across Epyc, Threadripper, Ryzen and next gen consoles should result in each chiplet being stupidly cheap. The 14nm IO die for Ryzen is smaller than the current Ryzen die and yields will be stupidly high, and the the additional cost is likely negligible due to their special deal with GlobalFoundries that forces them to pay even if they don't use the capacity. I would honestly expect AMD to price aggressively, and to keep pushing their advantages. Previously they had a core count advantage for retail customers, and I don't see them trading that out for only a per core performance advantage while Intel is unable to respond, let alone after that advantage has been lost.
|
|
#
?
Jan 26, 2019 05:14
|
|
|
WSA ends next year, is unlikely to be a factor going forwards.
|
|
#
?
Jan 26, 2019 05:29
|
|
|
Measly Twerp posted:I would honestly expect AMD to price aggressively, and to keep pushing their advantages. Previously they had a core count advantage for retail customers, and I don't see them trading that out for only a per core performance advantage while Intel is unable to respond, let alone after that advantage has been lost. 100% of the power for 70% of the cost, all the way up through the halo products, which are 140% the power for 100% of the cost. Best of both worlds.
|
|
#
?
Jan 26, 2019 05:35
|
|
|
SwissArmyDruid posted:WSA ends next year, is unlikely to be a factor going forwards. It's still a factor right now however, and they're going to use that to their advantage as much as they are able. Methylethylaldehyde posted:100% of the power for 70% of the cost, all the way up through the halo products, which are 140% the power for 100% of the cost. Best of both worlds. I'm still anticipating some gaming deficit, but yeah, exactly.
|
|
#
?
Jan 26, 2019 05:41
|
|
|
Anybody got a clock for clock comparo between 2700x and 9900k?
|
|
#
?
Jan 26, 2019 06:33
|
|
|
Seamonster posted:Anybody got a clock for clock comparo between 2700x and 9900k? Going by Cinebench, Intel IPC advantage is only 6.5%.
|
|
#
?
Jan 26, 2019 07:14
|
|
|
Palladium posted:https://www.anandtech.com/show/13405/intel-10nm-cannon-lake-and-core-i3-8121u-deep-dive-review/7 The CPU is literally from a low budget Chinese laptop made solely for their schools, I feel that this isn't exactly a fair analysis for now They are struggling with the process but this CPU is clearly not meant to be compared alongside desktop CPUs
|
|
#
?
Jan 26, 2019 10:28
|
|
|
Yup, using almost 50% more power than 14nm KBL at a mere 2.2GHz really inspires confidence for 10nm desktop
|
|
#
?
Jan 26, 2019 10:50
|
|
|
Zedsdeadbaby posted:The CPU is literally from a low budget Chinese laptop made solely for their schools, I feel that this isn't exactly a fair analysis for now They did a detailed comparison of it against the i3-8130U which is a laptop CPU with the same 15W TDP and pretty much the same core just in 14nm instead of the i3-8121U's 10nm. There are some benches of it against various other CPU's later, some of which were desktop CPU's, in that review but it was pretty clear what the really relevant comparison here was. Yeah its pretty clearly a low end laptop but it was also pretty clear that Intel's 10nm is doing pretty lovely right now and putting it in a better laptop won't change this. They even mentioned that when using extra fans to cool the CPU and give it some help there was no improvement either in performance or power use so you can't reasonably blame the chassis its in here at all.
|
|
#
?
Jan 26, 2019 13:22
|
|
|
Zedsdeadbaby posted:The CPU is literally from a low budget Chinese laptop made solely for their schools, I feel that this isn't exactly a fair analysis for now It's also being used in Intel's latest NUC models which are definitely meant for desktop use (albeit low-power office/media center type desktops).
|
|
#
?
Jan 26, 2019 15:14
|
|
|
Seamonster posted:Anybody got a clock for clock comparo between 2700x and 9900k? 2700X is already ahead of the 9900K in Cinebench, clock for clock. As you'd expect, AMD is putting their best foot forward in their demo - they picked the benchmark with the highest IPC advantage and then tuned the clocks so that they'd barely beat Intel, at the lowest clocks they could guarantee a win, so that they could show the maximum possible power advantage. Per Lisa Su, "clocks are not final yet", AKA the test wasn't necessarily a particular SKU that will exist in the retail lineup, and it was probably on a minimal engineering-sample board with no extraneous peripheral controllers (vs a flagship Z390 board). Stuff will improve over the engineering samples, granted, but AMD was putting their best foot forward. That's what you do in a super-high-profile public demo. In Cinebench, it's probable that Zen2 will dump on Coffee Lake by >10%, maybe 15%. But Zen1 literally already won against coffee lake. Cinebench is a completely lovely benchmark that has no vector optimizations, no cross-thread communication, and no impact from memory latency, it's not representative of any other task, including other renderers like Blender. Real world tests? Well, Zen2 seems to have ~15% higher memory latency than Zen+ in that UserBenchmark leak. Admittedly that's with crappy RAM, but it doesn't have the L4 on the IO die that some people were hoping for either. Frankly the chips of interest here are going to be the 8C variants since those are single-die, and the Threadripper 32C. The 16C is just going to be a weird Baby Threadripper on an AM4 socket. Yeah, some people will like it for pseudo-HEDT builds, but it won't be a good value for gaming, and Threadripper will destroy it in workstation/HEDT/performance type tasks. Overall I think Zen2 will keep up with the 9900K, when overclocked to the brink, or be close enough it doesn't matter (0-5% behind). The relevant questions here are whether performance tanks when you spread across multiple chiplets like Threadripper does, and how easy it is to cool that 7nm chip at those peak overclocks with a ~doubling of heat density vs 14nm (in the active areas of the cores). Power consumption doesn't go down if you are pushing the clocks right to the firewall, and transistor density has doubled on 7nm, relative to 14nm. 8Cs will probably be $300-350, maybe $400 for a top bin, but the 12C/16C variants are undoubtedly going to sell for $400-500. 7nm is expensive, and AMD has never hesitated to price "high" in absolute terms (see: FX-57, $1031 in 2005 dollars), but particularly when they think they have a better value proposition than Intel. AMD sold the top tier of Ryzen chips at $530, remember. 16C at $500 MSRP would sell like hotcakes, a high-binned 16C at $600 MSRP would still sell great. Remember, even if core-to-dollar stays the same, you'd still be getting a big bump in single-thread performance/performance-per-core. And companies measure value msrp-vs-msrp, not against the clearance pricing you're seeing right now. Value-priced Zen+ will absolutely be competitive vs MSRP Zen2, that's just how things work, people just whine about it when NVIDIA does it. Clearance-priced Ryzen 1000 series was competitive vs the 2000s, clearance-priced Fury was competitive vs the Polaris, clearance-priced 900 was competitive vs the 1000, clearance-priced 700 was competitive vs the 900s, ad infinitum. The whole point of clearance pricing is getting you to buy the old stock, if the new stock was better value-for-dollar than nobody would buy the old stock and companies would be stuck with it and have to write it off. Value of a new lineup is always measured msrp-vs-msrp. And if they can get the 32C Threadrippers anywhere near $1000 I'm totally down. Between the core count increase and full-rate AVX2 they could easily be three times as fast for x265 encoding or parallel x264 encoding. Paul MaudDib fucked around with this message at 16:20 on Jan 26, 2019 |
|
#
?
Jan 26, 2019 15:39
|
|
|
AMD absolutely should sell Zen 2 cheaper than the equivalent i7/9s. If they're the same price, people will just pick the one they consider a safer bet, and that's Intel. But I also agree that people who think they're getting more than 8 cores cheap are in for a disappointment.
|
|
#
?
Jan 26, 2019 15:58
|
|
|
Paul MaudDib posted:2700X is already ahead of the 9900K in Cinebench, clock for clock. As you'd expect, AMD is putting their best foot forward in their demo - they picked the benchmark with the highest IPC advantage and then tuned the clocks so that they'd barely beat Intel, at the lowest clocks they could guarantee a win, so that they could show the maximum possible power advantage. Per Lisa Su, "clocks are not final yet", AKA the test wasn't necessarily a particular SKU that will exist in the retail lineup, and it was probably on a minimal engineering-sample board with no extraneous peripheral controllers (vs a flagship Z390 board). Stuff will improve over the engineering samples, granted, but AMD was putting their best foot forward. That's what you do in a super-high-profile public demo. Wouldn't a test using AVX2 be a better showing since they doubled FP width on Zen 2?
|
|
#
?
Jan 26, 2019 18:37
|
|
|
Khorne posted:Jokes aside, that's where the practical benefit of more cores comes from. Individual applications can be more aggressive about dominating a single core. Made my decision to go dual processor way back during the Pentium II days a good one. Paul MaudDib posted:And if they can get the 32C Threadrippers anywhere near $1000 I'm totally down. Between the core count increase and full-rate AVX2 they could easily be three times as fast for x265 encoding or parallel x264 encoding. Combat Pretzel fucked around with this message at 18:45 on Jan 26, 2019 |
|
#
?
Jan 26, 2019 18:40
|
|
|
Combat Pretzel posted:Hell yea. I mean, I'm on a 2950X since August, but if there's a decent IPC and clock bump, and 32 cores, and faster AVX, I'll let the wallet bleed some more. After that, I should be set. I don't expect large jumps for the next few years. The industry people I follow on twitter say that Milan (Zen 3) is some crazy ambitious design but they're tight lipped on any actual details so far.
|
|
#
?
Jan 26, 2019 18:46
|
|
|
Risky Bisquick posted:Some tasks simply do not scale, however just google multithreading in language and click on a stack overflow link and you are halfway to being a professional. Please do not take this advice and then expect me to debug your code  Also check out Amdahl's Law.
|
|
#
?
Jan 26, 2019 19:27
|
|
|
Arzachel posted:Wouldn't a test using AVX2 be a better showing since they doubled FP width on Zen 2? Cinebench has been pretty AMD friendly, especially after Zen+ got a big ol boost for some reason. AT guessed that it was due to some SMT tweaks. I imagine thats why AMD was using it as their comparator, seems to fit the arch. Here is the Zen 1 vs Zen+ IPC graph. 
|
|
#
?
Jan 26, 2019 19:58
|
|
|
It's a shame that people are drones and will keep buying Intel because they think it's "the best" because it was the best for so long. Same reason people buy Honda and Toyota like crazy today - all because of the work their engineers did in the 90s, not the work they did in 2016. Brand goodwill goes far and there is a delay for the average joe to get with the times. AMD has to keep prices attractive, unfortunately for them.
|
|
#
?
Jan 27, 2019 22:01
|
|
|
Sorry about your Chevy Cavalier I guess
|
|
#
?
Jan 27, 2019 22:18
|
|
|
|
| # ? May 21, 2024 02:19 |
|
|
People keep buying Intel because their CPUs are still dramatically superior for gaming. It's not that complicated. If Zen2 changes that, AMD will sell like mad.
|
|
#
?
Jan 27, 2019 22:23
|
|