|
Khorne posted:I am sitting here on a 5200u (2c) and facebook is instant even with a bunch of messenger tabs, a video playing in chrome, visual studio code, sublime text, a video call, a few ssh sessions, and some other crap open.
|
#
?
Jul 28, 2017 05:28
|
|
|
|
| # ? Apr 27, 2024 19:17 |
|
|
Isn't Ryzen generally agreed to be roughly Haswell IPC? That's only 3-4 years old, and generational improvement on Intel's side has slipped to only 5% or so. Then you have the thermals shitshow going on right now making those improvements moot if it starts throttling on you. Ryzen cores are 'good enough', which combined with cheaper cores makes it the default best value for any task not bound by IPC performance. Even for gaming there's an argument to be made that games are going to become more heavily threaded in the future and it might be worth sacrificing some of those top end frames to avoid bottoming out during context switches. It's not winning the absolute performance crown, but for just about any other task it's your bang for buck.
|
|
#
?
Jul 28, 2017 05:33
|
|
|
Khorne posted:Nah, I have an i7 3770k and it has around 10% higher single and quad core performance than an overclocked ryzen. I paid around $330 for it 5 years ago.
|
|
#
?
Jul 28, 2017 05:35
|
|
|
Khorne posted:20%-40% of your 2600k is probably only using 2 cores. It does the same to my 3770k. ...And it's definitely not the only thing I do at once.
|
|
#
?
Jul 28, 2017 05:37
|
|
|
SourKraut posted:Source? I'm curious. If you mean on benchmarks, check user submitted benchmarks or any aggregate site. The numbers are roughly, 3770k just flat out beats 1700 stock/oc, it's even with 1700x stock and has a ~6% edge oc, it gets beat by 1800x stock, has ~2% edge oc, and obviously at that slim of a margin it gets smashed on anything that leverages the extra cores. Right, and quad core the 1700x and 1800x both outperform the 3770k slightly if I remember right. Kinda moot given that no one should buy a 3770k in 2017 because even used they demand too high of a price, but it's not like performance got worse on later gen intel cpus. PerrineClostermann posted:...And it's definitely not the only thing I do at once. I don't think web browser justifies more cores in any way, and even things that benchmark new cpus with web browsers will agree as far as I've seen. 8+ cores is definitely a big reason I am looking forward to upgrading in a year or two. At the rate things are going it looks like I'm waiting for my computer to die a horrible death or for 7nm to hit, though, which will probably be longer than a year or two. Khorne fucked around with this message at 05:52 on Jul 28, 2017 |
|
#
?
Jul 28, 2017 05:41
|
|
|
Khorne posted:If you mean on benchmarks, check user submitted benchmarks or any aggregate site. The numbers are roughly, 3770k just flat out beats 1700 stock/oc, it's even with 1700x stock and has a ~6% edge oc, it gets beat by 1800x stock, has ~2% edge oc, and obviously at that slim of a margin it gets smashed on anything that leverages the extra cores. Right, and quad core the 1700x and 1800x both outperform the 3770k slightly if I remember right.
|
|
#
?
Jul 28, 2017 07:33
|
|
|
Being hurt about Skylake-X being bad when the 5820k and its boards are still on sale and ripe for the picking (not for very long though)
|
|
#
?
Jul 28, 2017 07:40
|
|
|
Khorne posted:Is -j6 faster than -j9/-j12 with the 5820K and that particular build? Yeah, 1.5-2C is the usual recommended number of jobs. Depending on the build they might be running out of memory first though. In the modern era of SSDs it's less important to have threads to cover blocking IOs and 1C usually isn't hitting you too much though, and the more RAM your threads are eating the less the OS has to play with for caching. The only regret I have about putting 32 GB on my 5820K (a whole $126, for DDR4-3000 CAS15) is that I didn't buy an extra 32 GB and just max out my processor back before it doubled in price. RAM for the RAMthrone. Win10 is extremely aggressive about caching poo poo in RAM if it has free space to play, which is a massive overall boost to the performance of various tasks. The first time I read/transfer a file it goes into RAM, then everything happens at lightspeed. It's also way better about writing large chunks to HDDs by using large RAM buffers if it has headroom - I've literally seen it use several gigabytes to buffer a large file transfer, the transfer "completed" and network traffic stopped but my drive traffic was maxed for quite a while longer. (is there a way to make Linux cache/buffer similarly aggressively? forget swap, I want to put more stuff in RAM) Both video encoding and rendering will definitely scale out almost trivially to more cores too, especially if you turn the encoding quality way up (veryslow). You could probably use Threadripper for those pretty efficiently. Paul MaudDib fucked around with this message at 08:02 on Jul 28, 2017 |
|
#
?
Jul 28, 2017 07:45
|
|
|
That seems to be what its saying, unfortunately. My 4790k has a ~10% ipc advantage at any rate, and a marginal quad core one
|
|
|
#
?
Jul 28, 2017 13:18
|
|
|
Where do i get these used 10 core xeons for $70? People on Ebay want a lot more for their E5-2680 V2's.
|
|
#
?
Jul 28, 2017 13:26
|
|
|
Paul MaudDib posted:Yeah, 1.5-2C is the usual recommended number of jobs. Depending on the build they might be running out of memory first though. In the modern era of SSDs it's less important to have threads to cover blocking IOs and 1C usually isn't hitting you too much though, and the more RAM your threads are eating the less the OS has to play with for caching.
|
|
#
?
Jul 28, 2017 13:39
|
|
|
NihilismNow posted:Where do i get these used 10 core xeons for $70? You don't. He's thinking of the days of $70 E5-2670s. Something like that will happen again soon with v2 E5 chips, but until then its slim pickings. Also LGA 2011 boards aren't plentiful or cheap.
|
|
#
?
Jul 28, 2017 14:04
|
|
|
Khorne posted:It's a pretty solid platform and the first time AMD has been anywhere near competitive in ages. This is, frankly, the big reason people are excited about Ryzen; not that it's a slam-dunk gently caress you leap-frog ahead of Intel and now there's no reason to buy an i3/5/7/9 product, but that it's actually competitive, as there are legitimate markets for it where it is actually a Good Product At A Good Price. Sure, it might not win the IPC/single-thread crown, but it is fast enough in those metrics that your normal user will never know the difference, and it's cheaper, which is a win for Joe Average who is still on some <25Mbps lovely DLS or cable internet anyhow. For Bob the Renderer, cheaper cores is more-better regardless, so it's a win for him, too. As for used parts, you can still get a E5-2670 for <$100 on eBay (e; the v2's are $200+), which is a good deal if you're on a budget and need that sort of performance profile. But it's silly to say that the existence of used $100 E5-2670's invalidates the potential success of ThreadRipper anymore than it invalidates the potential success of the E5-2670v2, v3, v4, or any E5-4xxx chip; there will always be cheaper used parts that give you a much better bang for your buck than a new product, but support, warranties, "what exactly am I buying," corporate purchases, etc., all make that fact basically irrelevant from a macro level. DrDork fucked around with this message at 14:49 on Jul 28, 2017 |
|
#
?
Jul 28, 2017 14:45
|
|
|
So we've gotten to the bottom of this: You're a sucker if you buy CPUs less than 3 years old. Why get a Xeon v4 or Gold/Platinum when you can get a v1 or v2? Why buy an X299 chip when you can buy 2014's finest 5820K? Why buy memory in 2017 when you could time travel to when it cost half as much 18 months ago?
|
|
#
?
Jul 28, 2017 15:07
|
|
|
Khorne posted:Why are people so hyped about ryzen/threadripper? It seems competitive on paper, but if you look at benchmarks it has 7%-10% worse single core performance than a 6 year old intel architecture. Which is mostly what matters on the desktop. It's nowhere near competitive with recent intel stuff unless your life is encoding videos, and at that point intel has a better solution there too: buy used 10 core xeons for $70 each and put them in a dual socket motherboard. The used 10-core Xeon is gonna have really low clocks, like 2-2.5GHz compared to 4GHz on Threadripper.
|
|
#
?
Jul 28, 2017 15:24
|
|
|
NihilismNow posted:Where do i get these used 10 core xeons for $70? Ryzen is favorable to getting an old xeon for desktop use. The TDP difference alone if you plan on using it heavily should pay for itself, it likely has better gaming or single threaded performance, and not having to "hack" the bios for nvme support is a big plus. If it's not going to be under constant heavy load, if you don't want nvme, if single thread performance isn't a sticking point, and if you value ECC the water gets much murkier. Technically, ryzen/threadripper support ecc, but do/will affordable motherboards exist with ecc support? Twerk from Home posted:So we've gotten to the bottom of this: You're a sucker if you buy CPUs less than 3 years old. Why get a Xeon v4 or Gold/Platinum when you can get a v1 or v2? Why buy an X299 chip when you can buy 2014's finest 5820K? Why buy memory in 2017 when you could time travel to when it cost half as much 18 months ago? I do lots of cpu intensive stuff on the hpc clusters I have access to, but if the task is 4 or less cores it runs notably faster on my desktop even vs fairly decent broadwell xeons. I still don't usually run them on my desktop because sending it to something that does nothing else and checking back tomorrow is real convenient, and in all honesty depending on what I was doing at home I'd build a system with a large number of cores that's not my desktop and do the same drat thing. I realize this isn't an option for certain serious hobby/work stuff. Khorne fucked around with this message at 15:38 on Jul 28, 2017 |
|
#
?
Jul 28, 2017 15:29
|
|
|
Are there many common tasks that require serious computing power but cannot effectively take advantage of a large number of threads? Obviously gaming is one of them but are there a lot of others? That would be the only sort of task where Intel would beat Ryzen/Threadripper by a substantial margin.
|
|
#
?
Jul 28, 2017 15:30
|
|
|
NihilismNow posted:Where do i get these used 10 core xeons for $70? You buy ES (engineering sample) chips. Though they are typically more expensive than $70. Ive built a couple of audio workstations for a guy who did dual 12 core ES chips on socket 2011 (non v3)
|
|
#
?
Jul 28, 2017 15:38
|
|
|
MaxxBot posted:Are there many common tasks that require serious computing power but cannot effectively take advantage of a large number of threads? Obviously gaming is one of them but are there a lot of others? That would be the only sort of task where Intel would beat Ryzen/Threadripper by a substantial margin. Synthetic Benches!
|
|
#
?
Jul 28, 2017 15:52
|
|
|
Rest of the Skylake-X specs are out:
|
|
#
?
Jul 28, 2017 16:46
|
|
|
i'm dumb
|
|
#
?
Jul 28, 2017 16:58
|
|
|
Cygni posted:Rest of the Skylake-X specs are out: I wonder how much these cost to produce vs threadripper?
|
|
#
?
Jul 28, 2017 16:59
|
|
|
Khorne posted:The problem is if you have a competitive desktop chip from the past 5 years there's very little incentive to upgrade, and a lot of extra cores is a potential incentive, just like nvme. However, when the ipc is worse than your existing chip from 2012 it really makes it a hard sell unless your workflow is hungry for more cores and you can't offload it. The point, again, being that Ryzen/ThreadRipper is actually a viable option now, vs hilarious piles of poo poo like Bulldozer. MaxxBot posted:Are there many common tasks that require serious computing power but cannot effectively take advantage of a large number of threads? Obviously gaming is one of them but are there a lot of others? It's pretty limited these days. Games, as you noted, are usually the biggest point on that issue. Javascript and lovely webpages can also get into that territory, but any modern desktop chip is already so fast that you're talking fractions of a second difference in rendering speed in most cases; it's a far bigger issue when you look at the ULV chips in lovely laptops and chromebooks where they can legitimately still struggle.
|
|
#
?
Jul 28, 2017 16:59
|
|
|
Khorne posted:I don't think web browser justifies more cores in any way, and even things that benchmark new cpus with web browsers will agree as far as I've seen. in my experience it *does* depend upon what type of sites you are browsing. terrible awful messes like facebook and twitter (even youtube if you have it playing a video in the background) will totally suck up multiple cores. in my use case i'm sitting in queue for 5 ish minutes in a game and have a few active tabs going, my quad core is pegged on one core at 100 and the rest at 40-80 percent utilization. granted i've got an old potato but theres no way that would be a satisfactory experience on a dual core :P
|
|
#
?
Jul 28, 2017 17:01
|
|
|
DrDork posted:It's pretty limited these days. Games, as you noted, are usually the biggest point on that issue. Javascript and lovely webpages can also get into that territory, but any modern desktop chip is already so fast that you're talking fractions of a second difference in rendering speed in most cases; it's a far bigger issue when you look at the ULV chips in lovely laptops and chromebooks where they can legitimately still struggle.
|
|
#
?
Jul 28, 2017 17:10
|
|
|
That a bunch of mobile chips still struggle with Facebook does, indeed, say a lot. Now whether it says more about the chip or about Facebook's code and insistence on auto-play EVERYTHING is a different question.
|
|
#
?
Jul 28, 2017 17:25
|
|
|
At least they don't autoplay the loving audio which lots of lovely websites love doing. Whoever decided it was a great idea to have autoplaying videos with audio that follow you around as you scroll should be tortured to death.
|
|
#
?
Jul 28, 2017 17:35
|
|
|
Twerk from Home posted:So we've gotten to the bottom of this: You're a sucker if you buy CPUs less than 3 years old. Why get a Xeon v4 or Gold/Platinum when you can get a v1 or v2? Why buy an X299 chip when you can buy 2014's finest 5820K? Why buy memory in 2017 when you could time travel to when it cost half as much 18 months ago? Does it fill you with rage to know that I've been plotting out a X99 build with a Xeon for a new home NAS?  (X99 lets me get 40 spare lanes into a tiny-rear end U-NAS NSC-810a case that looks like a Synology unit, but I leave myself tons of expansion room vs the prebuilt or Ryzen/LGA1151. I'd be open to using Threadripper in theory, but TDP is a problem and high clocks will beat tons of cores here. Also, I need a very particular configuration to make this work (8+ SATA lanes onboard) and few boards match, and even fewer coolers fit into the absurd 52mm clearance.) Nah, but, that's what happens when IPC gains stop happening. Apart from power consumption, a computer from 5 years ago is pretty much just as fast as a modern PC for a lot of gaming/workstation tasks. A 6-core Sandy Bridge-E is still a wicked perfomer, and the 5820K really has yet to be bested. Including by Ryzen - Ryzen is a win on value and I wouldn't recommend it for new builds anymore, but X99 really almost a strict superset of Ryzen's performance and capabilities, and it's still a pretty great value offering considering you could have been running a 5820K for years now. Including, yes, when RAM was half the price it is now. X299 sure isn't any faster for most gaming stuff, barring some massive patches or BIOS fixes or something. And new high-clocked Xeons are absurdly expensive, at least for anything faster than 4C, so it's really worth looking at used stuff, engineering samples, or Threadripper once that launches. Paul MaudDib fucked around with this message at 17:59 on Jul 28, 2017 |
|
#
?
Jul 28, 2017 17:55
|
|
|
Anime Schoolgirl posted:the fact that core m3/m5 chips from the same generation are being clowned in browser performance by braswell is pretty loving sad trombone material what do you mean when you use this term, braswell?
|
|
#
?
Jul 28, 2017 22:23
|
|
|
ArgumentatumE.C.T. posted:what do you mean when you use this term, braswell? Braswell is the codename for netbook class Atom CPUs from 2015. Celeron and Pentium branded.
|
|
#
?
Jul 28, 2017 22:26
|
|
|
MaxxBot posted:Are there many common tasks that require serious computing power but cannot effectively take advantage of a large number of threads? Obviously gaming is one of them but are there a lot of others? That would be the only sort of task where Intel would beat Ryzen/Threadripper by a substantial margin.
|
|
#
?
Jul 28, 2017 22:54
|
|
|
Twerk from Home posted:Braswell is the codename for netbook class Atom CPUs from 2015. Celeron and Pentium branded. the official name for an intel product is Bra Swell? and people make fun of AMD for the name EPYC?
|
|
#
?
Jul 28, 2017 23:55
|
|
|
ArgumentatumE.C.T. posted:the official name for an intel product is Bra Swell? It was the only way the Intel engineers were ever going to touch a bra 
|
|
#
?
Jul 29, 2017 00:03
|
|
|
Pablo Bluth posted:Adobe Lightroom. Stuff like export should be embarrassingly parallel but Abobe can't program for poo poo. You can start multiple export jobs to parallelize out, just launch (say) 4 exports with 1/4 of your photos each. It'll happily saturate my 5820K that way. But yeah, you shouldn't have to.
|
|
#
?
Jul 29, 2017 00:27
|
|
|
Parallel jobs is the major thing that got me to switch from Handbrake to Staxrip. Not only is the UI easier and the components more updated because they're not integrated into the core like with Handbrake, but parallel jobs are as simple as setting up another rip and clicking the button. It might be a little particular to my workflow because my machine with a bluray drive isn't the same machine I encode on, I rip on my low powered NAS, stripping DRM there, then the DRM free 1:1 rip gets encoded on my 5820K machine. I think the only thing Staxrip might be lacking on is support for transcoding straight from DRMed DVDs and blurays, I've never checked out that functionality and there's probably a good reason people like Handbrake. For blurays the bit rate's high enough to use all 6 cores, but for DVDs it tends to only use somewhere between 3 and 4 for a job so being able to run them in parallel basically doubles my encoding speed. If X265 isn't going to use all my cores natively at least I figured out a workaround to do it. I should have my entire DVD library that I ripped years ago to Divx reripped to 265 by the end of the year.
|
|
#
?
Jul 29, 2017 12:24
|
|
|
I have three laptops I use regularly, a Surface Pro 3 with an i5, an HP 2570p with a dual core Ivybridge i7 and an Asus 'gaming' monster with a quad core Haswell i7 For browsing only the Surface Pro struggles which I'm guessing is more down to cooling (or lack thereof) than anything else, there is no noticeable performance difference between the other two.
|
|
#
?
Jul 29, 2017 12:30
|
|
|
craig588 posted:Parallel jobs is the major thing that got me to switch from Handbrake to Staxrip. Not only is the UI easier and the components more updated because they're not integrated into the core like with Handbrake, but parallel jobs are as simple as setting up another rip and clicking the button. It might be a little particular to my workflow because my machine with a bluray drive isn't the same machine I encode on, I rip on my low powered NAS, stripping DRM there, then the DRM free 1:1 rip gets encoded on my 5820K machine. I think the only thing Staxrip might be lacking on is support for transcoding straight from DRMed DVDs and blurays, I've never checked out that functionality and there's probably a good reason people like Handbrake. For blurays the bit rate's high enough to use all 6 cores, but for DVDs it tends to only use somewhere between 3 and 4 for a job so being able to run them in parallel basically doubles my encoding speed. If X265 isn't going to use all my cores natively at least I figured out a workaround to do it. I should have my entire DVD library that I ripped years ago to Divx reripped to 265 by the end of the year. StaxRip is so good, thank you. HalloKitty fucked around with this message at 16:41 on Jul 29, 2017 |
|
#
?
Jul 29, 2017 13:37
|
|
|
dissss posted:I have three laptops I use regularly, a Surface Pro 3 with an i5, an HP 2570p with a dual core Ivybridge i7 and an Asus 'gaming' monster with a quad core Haswell i7 My work thinkpad struggles with browsing (unusable on battery) because we have 170 processes running by default
|
|
#
?
Jul 29, 2017 15:55
|
|
|
fivehead posted:My work thinkpad struggles with browsing (unusable on battery) because someone thought it was ok to issue laptops with 4GB RAM, a 5400RPM HDD, and then require full disk encryption and McAfee software, and we have 120 processes running by default, most of which don't do anything useful. Fixed for me.
|
|
#
?
Jul 29, 2017 18:15
|
|
|
|
| # ? Apr 27, 2024 19:17 |
|
|
Some more details and names. 6 cores coming to i5:  Weird that the 8700 is nearly identical in clocks to the 8700k, but only a 65w TDP.
|
|
#
?
Jul 29, 2017 23:13
|
|