|
Twerk from Home posted:You know, the flip side of this is that I can keep riding the 2500K that I got for $200 in 2011 for the forseeable future. My 2600k is still absurdly powerful. What a world.
|
#
?
Apr 16, 2015 16:06
|
|
|
|
| # ? May 28, 2024 15:11 |
|
|
Twerk from Home posted:You know, the flip side of this is that I can keep riding the 2500K that I got for $200 in 2011 for the forseeable future. So true... my Phenom II X4 840 has since been transferred to a different build and relegated to Plex server duties only.
|
|
#
?
Apr 16, 2015 17:21
|
|
|
We live in an interesting world concerning performance. The 2011's A8 Llano still works surprisingly well for a CPU that was underclassed by Intel chips from 2008. Back in 1999, your 166MHz Pentium MMX you bought two years before was a paperweight when compared to the 450MHz K6-2 your dad just brought home. I don't even want to upgrade. After just dropping $799 on a Yoga Pro 2 for my mobile needs, I'll wait for Cannonlake or Zen for the desktop.
|
|
#
?
Apr 16, 2015 17:53
|
|
|
SwissArmyDruid posted:AMD owns Seamicro. I think that's the use case. You don't build a server chip that you can't use in that microserver company that you own. Well they don't have to worry about that problem anymore. http://anandtech.com/show/9170/amd-exits-dense-microserver-business-ends-seamicro-brand
|
|
#
?
Apr 17, 2015 00:34
|
|
|
Ragingsheep posted:Well they don't have to worry about that problem anymore. Motherfucker, seriously?! Edit: Sigh. ...Well. On the other hand, AMD isn't selling SeaMicro, they're retaining the IP portfolio, and I can only assume they're going to take what they know about dense computing, merge it how they play nice with ARM, and... gently caress it, I don't know anymore. Maybe the interconnect fabric scales up to traditional servers, who the gently caress knows anymore. SwissArmyDruid fucked around with this message at 01:36 on Apr 17, 2015 |
|
#
?
Apr 17, 2015 01:32
|
|
|
SwissArmyDruid posted:Motherfucker, seriously?!
|
|
#
?
Apr 17, 2015 01:34
|
|
|
Well they did, until people realized that hardware-accelerated virtualization actually WAS their chance to have their cake and eat it too.
|
|
#
?
Apr 17, 2015 02:01
|
|
|
Seamicro had one major customer and the hardware was... problematic.
|
|
#
?
Apr 17, 2015 02:10
|
|
|
Stock under $3. It's like those nerds who said AMD was a bargain when it started its freefall to $5 and below didn't know what they were talking about
|
|
#
?
Apr 17, 2015 05:28
|
|
|
And then there's this bit of news, with Samsung using some AMD chips: http://fudzilla.com/news/processors/37549-amd-bald-eagle-flies-for-samsung-digital-media Not that it's going to boost AMD's stock price any.
|
|
#
?
Apr 17, 2015 06:13
|
|
|
WhyteRyce posted:Stock under $3. It's like those nerds who said AMD was a bargain when it started its freefall to $5 and below didn't know what they were talking about Hey come on now, HP-Compaq recovered 
|
|
#
?
Apr 17, 2015 07:43
|
|
|
Pretty bad Q1 results. http://www.anandtech.com/show/9172/amd-posts-q1-2015-results-180-million-net-loss I was looking forward to the R9 390, but I'm seriously questioning whether AMD will be able to stick around and provide driver support in a few years.
|
|
#
?
Apr 17, 2015 08:01
|
|
|
Hopefully their mobile platform is actually good!
|
|
#
?
Apr 17, 2015 08:25
|
|
|
Carrizo seems like it'd be a OK to decent low end APU. Not gonna set any sales records though. At this point even if Zen is a home run AMD has had the consequences of so many bad decisions baked into their financial cake that they'd probably still lose money or barely break even for a quarter or 2. If Zen does badly then yea I could see them folding in a few years time.
|
|
#
?
Apr 17, 2015 09:39
|
|
|
PC LOAD LETTER posted:Carrizo seems like it'd be a OK to decent low end APU. Not gonna set any sales records though. Zen is basically their make or break product. Now that samsung/glofo have "14nm" (sort of) intel's foundry lead is shrinking which means that AMD has a slim chance.
|
|
#
?
Apr 17, 2015 12:14
|
|
|
This is depressing.
|
|
#
?
Apr 17, 2015 15:19
|
|
|
Angry Fish posted:This is depressing.
|
|
#
?
Apr 17, 2015 17:00
|
|
|
Is Carizzo cut down? I'm wondering that, even if in it's most refined stage, the bulldozer uarch was always incapable of matching Sandy.
|
|
#
?
Apr 17, 2015 17:16
|
|
|
Maybe we can start a WinChip thread....
|
|
#
?
Apr 17, 2015 17:22
|
|
|
Angry Fish posted:Maybe we can start a WinChip thread.... How about a Rockchip one? Google seems invested in using those chips in lower-end Chrome devices.
|
|
#
?
Apr 17, 2015 17:23
|
|
|
teagone posted:How about a Rockchip one? Google seems invested in using those chips in lower-end Chrome devices. ARM cores are getting more and more efficient. Maybe there will come a time when ARM designs beat x86-64's in performance-per-watt?
|
|
#
?
Apr 17, 2015 17:41
|
|
|
Don't ARM beat x86 on performance per watt already on the lower range of things? I thought that was the point of ARM...
|
|
#
?
Apr 17, 2015 17:46
|
|
|
Angry Fish posted:ARM cores are getting more and more efficient. Maybe there will come a time when ARM designs beat x86-64's in performance-per-watt?
|
|
#
?
Apr 17, 2015 17:49
|
|
|
Truga posted:Don't ARM beat x86 on performance per watt already on the lower range of things? I thought that was the point of ARM... Not really, that's been a myth since like, forever? Maybe back in the P4 days that was true when no one making X86 chips gave a flying gently caress about power. http://anandtech.com/show/8357/exploring-the-low-end-and-micro-server-platforms/17 The ARM based server they're benching with there isn't on the newest process, but that's what is actually commercially available. The article notes that the newer version should be 50% more efficient than the one they reviewed, but even that's not enough to get close to the Xeon E3 -L series.
|
|
#
?
Apr 17, 2015 18:17
|
|
|
Yeah, I think that's part of the reason nobody uses micro servers. If you only have a small amount of electrical power then ARM is better than Intel, because they consume less electrical power. If on the other hand you have "unlimited" electrical power then you get more computing power for the same number of watts with Intel.
|
|
#
?
Apr 17, 2015 19:52
|
|
|
Wheany posted:Yeah, I think that's part of the reason nobody uses micro servers. What you're saying is that the ARM chips are less efficient at turning watts into flops than Intel chips? In what world is the former what anyone would want?
|
|
#
?
Apr 17, 2015 21:40
|
|
|
Boiled Water posted:What you're saying is that the ARM chips are less efficient at turning watts into flops than Intel chips? In what world is the former what anyone would want? The smallest ARM chips are much lower power usage than Intel chips. They are still worse performance per watt, but good luck finding 2W Intel parts. Edit: ARM chips are also way better at using less power when idling.
|
|
#
?
Apr 17, 2015 21:46
|
|
|
Wheany posted:Yeah, I think that's part of the reason nobody uses micro servers. I agree with, but why do cell phones all use ARM designs instead of a modified older x86 design? Like what would happen if you took a Pentium era design and scaled it down to modern process size?
|
|
#
?
Apr 17, 2015 21:48
|
|
|
Twerk from Home posted:The smallest ARM chips are much lower power usage than Intel chips. They are still worse performance per watt, but good luck finding 2W Intel parts. I should have specified: Who would want them in a server farm/park?
|
|
#
?
Apr 17, 2015 22:00
|
|
|
Crotch Fruit posted:I agree with, but why do cell phones all use ARM designs instead of a modified older x86 design? Like what would happen if you took a Pentium era design and scaled it down to modern process size? That's most because of licensing. ARM freely licenses out their designs and architecture. If you want to build an ARM core of your own, great, sign here, pay a fee, here's the documentation, have fun. Intel, on the other hand, wants to control everything. You'll use their processors designed to their specifications, and assuming no contra-revenue tricks, you'll be paying for it, too.
|
|
#
?
Apr 17, 2015 22:04
|
|
|
Crotch Fruit posted:I agree with, but why do cell phones all use ARM designs instead of a modified older x86 design? Like what would happen if you took a Pentium era design and scaled it down to modern process size? Intel do make tiny X86 chips for phones, they're just not very popular. Off the top of my head the only "real" manufacturer who uses them is Asus.
|
|
#
?
Apr 17, 2015 22:06
|
|
|
Twerk from Home posted:Edit: ARM chips are also way better at using less power when idling. In what sense?
|
|
#
?
Apr 17, 2015 22:12
|
|
|
Intel didn't have anything anyone wanted when smartphones were first taking off and now things are pretty entrenched. Intel doesn't have an integrated modem like Qualcomm and Apple and Samsung are pretty happy with their own stuff.
|
|
#
?
Apr 17, 2015 22:14
|
|
|
Twerk from Home posted:The smallest ARM chips are much lower power usage than Intel chips. They are still worse performance per watt, but good luck finding 2W Intel parts. Intel server chips go from 6-20 watts for the Atom C2xxx series (The C2750 atom tested in the article I linked is a 22nm 8 core 20 watt part), and 13-84w for the latest Xeon E3 v3 line. (Most notably, 13, 25, and 45w for low power Xeons. If you hop up to the E5 you can get an 8 core chip @ 55w.) So yea, phone SOCs may come in at about 2 watts, but what's even the point of a 2w chip in a server environment? Low power use is nice but you do have to be able to do useful work at a low enough latency. (Also Intel's own phone SOCs come in right at about the same power, 2 watts. Broadwell-Y is 4.5 watts, and beats the snot out of equivalent ARM tablet SOCs.) And uh, idle power use for any modern chip is not even relevant, because they all use power/core gating to completely power down cores that aren't in use. Idle power difference between the chips is going to be milliwatts either way. That's a long way from "Way better." The whole "OMG ARM is going to kill intel!!!!" thing has been floating around for years, but somehow it never really materializes. The major advantage ARM has is just more varied companies using it and working on building their own implementations of it- More of a chance for one of them to innovate, etc, vs Intel's monolithic view. On the other side, Samsung is the only one that actually has similar vertical integration to Intel, in terms of controlling the design and manufacture of their chips end to end. And they're still a ways behind on process technology, since even their "14nm" doesn't actually shrink component sizes down to the same level Intel's 14nm process does. And the not-samsung foundries are even further behind. Gwaihir fucked around with this message at 23:26 on Apr 17, 2015 |
|
#
?
Apr 17, 2015 23:20
|
|
|
The A8X from Apple and the Samsung designs do a really good job competing with the Core M chipsets from Intel, but they're using more cores. Core Vs. Core, ARM still has fewer MIPS, but you're comparing apples to oranges what with ARM lacking integrated memory controllers and utilizing poorer bus utilization and less cache on the dies. Single threaded performance lags significantly, but so does AMD so. 
|
|
#
?
Apr 17, 2015 23:36
|
|
|
Yea, those chips do pretty well (Especially the GPU portions, against non-Iris versions of Intel's integrated graphics), but the latest ones are also pretty power hungry. It would be interesting to see them in a format like a module HP's Moonshot chassis, where there's no artificial power and sharp temperature limits like there are in Tablets meant to be held in hand while in use. That's turning out to be a major limitation to Broadwell-Y, as well, since there's only so much you can do when you drop to a totally fanless design. Still, though, look at the WebXPRT (CPU load) and Tablet 3dmark(GPU) tests from http://anandtech.com/show/9136/the-2015-macbook-review/9 and http://anandtech.com/show/9136/the-2015-macbook-review/10. Remove the thermal limits/put the chips in a position where they have active cooling and it would probably be much closer. At the ultra low power end like these devices operate in, performance is basically gated by "How hot can you afford to let your chassis get"/"How well does it passively dissipate heat?" Hence the all aluminum Asus Zenbook out-performing other units with worse dissipation or lower temperature limits, while it has a lower rated CPU in it.
|
|
#
?
Apr 17, 2015 23:52
|
|
|
Angry Fish posted:Maybe we can start a WinChip thread....
|
|
#
?
Apr 18, 2015 00:01
|
|
|
JawnV6 posted:In what sense? AFAIK there's no x86 equivalent to something like the 4+1 core ARM chips. Being able to sleep your powerful cores but still respond if needed is really great. The only thing that comes to mind is that tech Intel is working on that lets the CPU go to a full sleep but still respond to network connections. Damned if I can remember what they called that though. I think it was an evolution of the C6/C7 processor state. They weren't looking at putting it on their super low power stuff either, just server stuff from what I remember. Paul MaudDib fucked around with this message at 05:33 on Apr 18, 2015 |
|
#
?
Apr 18, 2015 05:31
|
|
|
Paul MaudDib posted:AFAIK there's no x86 equivalent to something like the 4+1 core ARM chips. Being able to sleep your powerful cores but still respond if needed is really great. Powergating all your cores except one and running the remaining one at low speed. It's really simple and has been available in dual core form since the mid 2000s and more cores since slightly later.
|
|
#
?
Apr 18, 2015 05:34
|
|
|
|
| # ? May 28, 2024 15:11 |
|
|
Nintendo Kid posted:Powergating all your cores except one and running the remaining one at low speed. It's really simple and has been available in dual core form since the mid 2000s and more cores since slightly later. The idea on 4+1 is that you build it using slower, lower-power processes than the big cores. Power usage doesn't follow a perfect linear scale. Taken to an extreme example, you'll never get a 4790k to consume 0.1W like a microcontroller given a similar computational load. At the end of the day you can never gate off and clock down a performance core far enough to match the power consumption of a core specifically designed to consume nothing. The tradeoff is that your software has to be smart enough to take advantage of it. If your kernel treats the battery-saver core like a normal core you're going to have issues. Paul MaudDib fucked around with this message at 05:52 on Apr 18, 2015 |
|
#
?
Apr 18, 2015 05:43
|
|