|
MaxxBot posted:They're not lazy, it's that desktop CPUs are less of a revenue generator than server and laptops CPUs so it makes little economic sense to focus on them, combined with the fact that competition from AMD has been basically nonexistent for the past several years until now. From that position it makes sense to milk desktop users for minimal effort because what else are they gonna do? Who knows, maybe Ryzen+ or Ryzen 2.0 will finally force Intel's hand.
|
# ? Mar 19, 2017 23:49 |
|
|
# ? May 21, 2024 15:57 |
|
JnnyThndrs posted:I wouldn't go that far, but $100 for hyperthreading? That's just a big 'gently caress you' to consumers. Same thing for the prices of the upper crust chips on the HEDT platform, $1100 and $1700 is just ridiculous bullshit because they can. AMD used to take people to the cleaner's too, back when they could get away with it. The high-end Athlon 64s and X2s were over $1000 back in the early 2000s, which works out closer to $1400 today. Paul MaudDib fucked around with this message at 01:32 on Mar 20, 2017 |
# ? Mar 19, 2017 23:51 |
|
MaxxBot posted:They're not lazy, it's that desktop CPUs are less of a revenue generator than server and laptops CPUs so it makes little economic sense to focus on them, combined with the fact that competition from AMD has been basically nonexistent for the past several years until now. From that position it makes sense to milk desktop users for minimal effort because what else are they gonna do? You're not addressing the actual statement though. It's unlikely any amount of special research effort would have resulted in Intel being able to put out consumer price range CPUs with 12 full cores and 3x the IPC on each of them, even though a naive projection based on the progression from the Pentium 4 to the early Core i series chips might have indicated that would happen.
|
# ? Mar 20, 2017 00:02 |
|
MaxxBot posted:They're not lazy, it's that desktop CPUs are less of a revenue generator than server and laptops CPUs so it makes little economic sense to focus on them, combined with the fact that competition from AMD has been basically nonexistent for the past several years until now. From that position it makes sense to milk desktop users for minimal effort because what else are they gonna do? The client computing group still makes a lot more revenue for Intel than the data center group FYI. (Profits on the other hand, )
|
# ? Mar 20, 2017 01:12 |
|
redeyes posted:Intel moved the one thing that can break to the motherboard. pins Yeah but why would AMD do that, too, for server parts but make some presumably lower production number consumer with special snowflake pins?
|
# ? Mar 20, 2017 01:22 |
|
fishmech posted:You're not addressing the actual statement though. It's unlikely any amount of special research effort would have resulted in Intel being able to put out consumer price range CPUs with 12 full cores and 3x the IPC on each of them, even though a naive projection based on the progression from the Pentium 4 to the early Core i series chips might have indicated that would happen. Noone is actually asking for 3x IPC and 12 full cores though, and it was just a facetious comment? The more realistic response, I thought, is that Intel has simply focused on reducing power while maintaining/slightly increasing IPC each generation. I think what a lot of people actually wish, is that Intel followed the console market approach, where slight improvements continue throughout the console's lifespan but prices continue to go down. So maybe at times when IPC has only increased by 5% or such, Intel might have offered that generation at a lower price than the previous. But Intel has never and will never do that, so whatever. It is silly thought that there's still such a price discrepancy between i5s and i7s for hyper-threading, but I blame consumers for buying i7s and not forcing Intel to reduce the price delta between the two each generation.
|
# ? Mar 20, 2017 01:35 |
|
I thought that Intel's Hyper-threading added substantially more complexity to their CPUs and therefore cost more to produce and that's why i7 chips cost substantially more than i5 chips for the consumer.
|
# ? Mar 20, 2017 01:59 |
|
SourKraut posted:Noone is actually asking for 3x IPC and 12 full cores though, and it was just a facetious comment? I mean, I wouldn't say no... SwissArmyDruid fucked around with this message at 02:22 on Mar 20, 2017 |
# ? Mar 20, 2017 02:19 |
|
spasticColon posted:I thought that Intel's Hyper-threading added substantially more complexity to their CPUs and therefore cost more to produce and that's why i7 chips cost substantially more than i5 chips for the consumer. Nope, same chips.
|
# ? Mar 20, 2017 02:28 |
|
spasticColon posted:I thought that Intel's Hyper-threading added substantially more complexity to their CPUs and therefore cost more to produce and that's why i7 chips cost substantially more than i5 chips for the consumer. i5s are the same chip. Intel only uses two different dies for their desktop lineup, there's the 4-core that gets used for i3s/i5s/i7s (also for Xeon E3), then the 12-core die that gets used for HEDT i7s and E5s/E7s. All their other product distinctions are done with binning and e-fuses. Die harvesting is definitely a thing, but only Intel would be able to tell you their yield rates. In general though hyperthreading doesn't take much extra space (hence why you would bother with it versus just adding more cores). I imagine there are very few i7s that are fully functional except for the second thread, certainly not enough to fill the huge demand for i5s.
|
# ? Mar 20, 2017 02:29 |
|
SourKraut posted:
Er, but that's exactly what they've done and people complained about : console revisions pretty much only reduce power used to get the same performance. You normally don't get any improvements in speed. The PS4 Pro and Xbox Scorpio buck the trend by being actually improved speed/performance devices, rather than just being process shrinks and higher chip integration the way the revisions to the PS3/360 or PS2 or PS1 were.
|
# ? Mar 20, 2017 02:42 |
|
Paul MaudDib posted:i5s are the same chip. Intel only uses two different dies for their desktop lineup, there's the 4-core that gets used for i3s/i5s/i7s (also for Xeon E3), then the 12-core die that gets used for HEDT i7s and E5s/E7s. All their other product distinctions are done with binning and e-fuses. I could have sworn that the Celerons / Pentiums / i3s were the dual-core dies that they use for mobile -U CPUs, but really leaky. i3s are actually a 2 core disabled i5?
|
# ? Mar 20, 2017 03:03 |
|
Paul MaudDib posted:I imagine there are very few i7s that are fully functional except for the second thread, certainly not enough to fill the huge demand for i5s. This is also where my complaint with Intel is mostly, around Haswell era the stack should have been Celeron (2C/2T), Pentium (2C/4T), i3 (4C/4T), i5 (4C/8T), i7 (6C/12T+). This wouldn't have been an impossible demand on Intel's margins either, and had they done that loving Ryzen would have been the wettest drat fart.
|
# ? Mar 20, 2017 03:14 |
|
Twerk from Home posted:I could have sworn that the Celerons / Pentiums / i3s were the dual-core dies that they use for mobile -U CPUs, but really leaky. i3s are actually a 2 core disabled i5? I don't know about leaky -U or whatever else their production chain might be made from, but at least the physical evidence speaks on this: http://www.xtremesystems.org/fugger/a7.jpg some OCer's delidded 7700 & i3 7350, the i3 is a smaller die. obviously not the same chip as a 4 core.
|
# ? Mar 20, 2017 04:10 |
7350: 7700:
|
|
# ? Mar 20, 2017 06:48 |
|
https://twitter.com/CPCHardware/status/843099198799187968 Translation: There will be a 16C/32T part with an HEDT focus on chipset X399, within 4-6 months. Clocks 2.4 - 2.8 GHz. 2-die MCMs. Quad-channel DDR4. Socket LGA SP3r2. 150W TDP.
|
# ? Mar 20, 2017 17:52 |
|
SwissArmyDruid posted:Translation: There will be a 16C/32T part with an HEDT focus on chipset X399, within 4-6 months. Clocks 2.4 - 2.8 GHz. 2-die MCMs. Quad-channel DDR4. Socket LGA SP3r2. 150W TDP. Are there multiple sources for this other than this OrangeKhrush fellow? I was under the impression R7 was AMD taking on HEDT and none of this sounds like anything AMD has telegraphed.
|
# ? Mar 20, 2017 18:18 |
|
I can't make much sense of it either. The release date seems too late to be a beta test for Naples. Corecount is nice but the clockspeed makes it somewhat niche. Most enthusiasts who can use all those cores would probably prefer a cheap dual e5-2670 build. 16C/32T @ 2.6-3.3 Ghz, 224W combined TDP, 40MB cache, octa-channel DDR3 and $90 per CPU. Perhaps they're going to release 8C/16T CPUs for that platform? How about a R7 1800X with quadchannel and a couple of the most glaring launch issues fixed?
|
# ? Mar 20, 2017 18:37 |
|
I would like to note that the CanardPC tweet says 2-die MCMs, not 2-socket boards. In other words, knowing what we know about AMD: interposers.
|
# ? Mar 20, 2017 18:55 |
|
I'm still lolling that this means OrangeKrush likely has a real leaker in the industry, and that AMD is going to loving roll fixed silicon into bad silicon and lol at early adopters. Seriously, if you're buying Ryzen or looking at Ryzen, please wait until like, August-September. Plenty of boards, likely higher density of good silicon versus bad, fewer and less annoying BIOS updates/issues. Also get some small entry CPU because Pinnacle/Zen2 is apparently Q1 2018, LMAO.
|
# ? Mar 20, 2017 19:22 |
|
You mean "end of Q1 2018", with a non-zero chance of "slip to Q2".
|
# ? Mar 20, 2017 19:26 |
|
The best part of all this will be reviewers getting fixed R3 silicon to test and having it perform better at similar clocks compared to the older R7/R5 they might have and concluding with more cores = bad.
|
# ? Mar 20, 2017 19:33 |
|
Remember when the GPU thread's subtitle was "Buy a 970"? The Something Awful Forums > Discussion > Serious Hardware / Software Crap > AMD CPU and Platform Discussion: Buy a 7700K
|
# ? Mar 20, 2017 20:00 |
|
For gently caress's sake, it's P L A T F R O M. Respect the loving marketing. PLATFROM.
|
# ? Mar 20, 2017 20:10 |
|
This is AMD so I'm calling it now, consumer Zen 2 will have lower IPC but better overall performance than present Ryzen.
|
# ? Mar 20, 2017 20:19 |
|
SwissArmyDruid posted:For gently caress's sake, it's P L A T F R O M. Respect the loving marketing.
|
# ? Mar 20, 2017 20:32 |
|
Kazinsal posted:Remember when the GPU thread's subtitle was "Buy a 970"? If you're making a gaming PC then yeah, if you're doing other stuff it's not that simple.
|
# ? Mar 20, 2017 20:45 |
|
Kazinsal posted:Remember when the GPU thread's subtitle was "Buy a 970"? In strictly gaming scenario's that's correct. If you want to do anything else, the 1700 is the better choice IMHO. eames posted:This is AMD so I'm calling it now, consumer Zen 2 will have lower IPC but better overall performance than present Ryzen. Er, how?
|
# ? Mar 20, 2017 20:55 |
|
SwissArmyDruid posted:For gently caress's sake, it's P L A T F R O M. Respect the loving marketing. Can we get a minor thread title change? GRINDCORE MEGGIDO fucked around with this message at 21:07 on Mar 20, 2017 |
# ? Mar 20, 2017 20:59 |
|
https://twitter.com/BitsAndChipsEng/status/843864982320267265 What's all this then?
|
# ? Mar 20, 2017 21:50 |
|
Truga posted:https://twitter.com/BitsAndChipsEng/status/843864982320267265 Comments sound like Windows 10 Game Mode, which means the 7700k would likely also get a boost.
|
# ? Mar 20, 2017 21:53 |
|
pixaal posted:Comments sound like Windows 10 Game Mode, which means the 7700k would likely also get a boost. 35% is absolutely unprecedented for Game Mode
|
# ? Mar 20, 2017 21:56 |
|
Ok here's a before and after: This would certainly explain things.
|
# ? Mar 20, 2017 21:58 |
|
So the NT kernel team figured out how to schedule things on Ryzen properly. Neat. Wonder if this is going to completely invalidate existing game benchmarks.
|
# ? Mar 20, 2017 21:59 |
|
If it's as easy as that then AMDs decision to launch in that state is even more baffling to me
|
# ? Mar 20, 2017 22:03 |
Kazinsal posted:So the NT kernel team figured out how to schedule things on Ryzen properly. Neat. This is really odd, even AMD said that there were no problems with the scheduler, so WTF is going on here? God, they should have sent a few systems with Ryzen chips over to MS a month before launch.
|
|
# ? Mar 20, 2017 22:07 |
|
Kazinsal posted:So the NT kernel team figured out how to schedule things on Ryzen properly. Neat. That is kind of a big deal. Any sources?
|
# ? Mar 20, 2017 22:21 |
|
I think people were interpreting "no problems with the scheduler" to mean "Windows 10 performance is as expected" and that wasn't the correct interpretation. IIRC the current theory is that at launch time Windows 10 was attempting to save power through techniques such as core parking; this update must be turning off some of those power optimizations and allowing the processor to do more managing its own power/performance.
|
# ? Mar 20, 2017 22:21 |
|
Isn't that before and after the same as disabling HT?
|
# ? Mar 20, 2017 22:25 |
|
|
# ? May 21, 2024 15:57 |
|
redeyes posted:That is kind of a big deal. Any sources? Sorry, that came out a bit too authoritative. I meant based on the screenshots of before and after a recent Windows 10 update that Truga posted. The before has an 8-thread benchmark eating up core time all over the place. The after has eight threads being used, and eight sitting idle. The after is about 20% faster. This implies that they've fixed the scheduler to not bounce stuff around needlessly in a way that'll cause major performance issues. AVeryLargeRadish posted:God, they should have sent a few systems with Ryzen chips over to MS a month before launch. Mmmyep. Back in Windows 95's development, they got samples of pretty much everything old, new, and upcoming for testing and writing helper code to support troublesome devices. And hey, it worked pretty well on launch! As long as you didn't have any horrible nasty DOS device drivers, which if you had a CD-ROM drive, you probably did. Fun fact: one of the driving forces in Windows 95's development was getting the VDM to run DOOM in a window Kazinsal fucked around with this message at 22:31 on Mar 20, 2017 |
# ? Mar 20, 2017 22:28 |