|
Until fairly recently, characterizing large ICs (such as a CPU, for example...) has been a lengthy and expensive process - so companies preferred to just assume each CPU is going to have nothing but the slowest possible transistor, and market its top frequency accordingly. (Okay, they did do some binning, but the test had a pretty wide margin of error - so a lot of performance was still left on the table). Now they have better tools and can quickly and cheaply characterize a large IC, and then help themselves to all of the headroom they were previously forced to leave on the table due to economic concerns. This is what we're seeing now... well, for designs coming out of the usual foundries. Intel of course has always preferred to build their own tools, so it's possible they've gotten complacent on the characterization side of things. They'll probably catch up one day, though. RIP overclocking.
|
#
?
May 20, 2018 18:12
|
|
|
|
| # ? Apr 19, 2024 04:21 |
|
|
eames posted:
The PS4 and Xbox both use Cat cores not Construction cores. If you wanted a rough approximation of IPC you'd have to look at Kabini. https://www.anandtech.com/show/7933/the-desktop-kabini-review-part-1-athlon-5350-am1
|
|
#
?
May 20, 2018 19:40
|
|
|
Khorne posted:I strongly disagree considering modern processors cap out at about where you want to run them. You need to over volt to the point of degradation to push an 8700k further than 5.0-5.2 which is where you want it to begin with. AMD's current processors aren't even really worth OCing even though the previous generation were. People bitch about Buildzoid crapping on B-boards etc and hating on low end stuff, but other people point out his channel is called “Actually Hardcore Overclocking”. He’s pointed out in the past that he doesn’t run his daily machine as high as he can get it, but he does take part in competitions and LN2 projects and he buys his components with that stuff in mind. Most people don’t have those needs. Saying 5.0-5.2 is “where you want it to begin with” suggests such numbers are common mainstream. Adoredtv’s research into SiliconLottery etc suggests that isn’t the case. He’s incredibly eager to draw conclusions that make Intel look bad, but he’s done more research on the matter than I ever will. quote:AM4 still has ~2 generations of CPUs left on it and it's likely they'll be power hungry. Again this is a difference between what people say they want and what people actually do. Someone buying a CPU in 2018 for home office or gamer tasks and using it until it is no longer sufficient to do a job aren’t going to buy another CPU before 2020. As someone who is aghast at the used prices for his Z77 motherboard and 3770K, I get that there’s people with older equipment that was mid-range when new and who would rather buy the last, greatest part in their generation than change generations completely. But that happens many years down the road when these parts are EOL already. And besides, isn’t the new process supposed to make chips less power hungry? Doesn’t power savings and efficiency go up over time? Last time I looked that was the direction technology trended. Craptacular! fucked around with this message at 23:04 on May 20, 2018 |
|
#
?
May 20, 2018 23:02
|
|
|
It's 40% more performance for same power or 60% less power for same performance on GloFos 7nm compared to 14nm. TSMC seems to be in the same ballpark. So probably 80W for 4Ghz 8C/16T, 140W for 5Ghz.
|
|
#
?
May 20, 2018 23:34
|
|
|
Bottom end motherboards really do have underdimensioned VRM's, though, and they really do limit performance. Things like the Asus Prime Z370-P comes with a VRM that tops out at a sustained power draw of about 120-130 watts, and just enabling all-core turbo on an i7-8700K at stock clocks can easily draw 150W or even more in stress tests, unless the chip is a top bin overclocker to begin with. I guess enabling all-core turbo is technically overclocking, but most people wouldn't see it that way. To actually run an "average" clocking 8700K with all cores at max stock turbo for an extended period of time at all, you need a board that's pretty solidly into the mid-range of the market, and if you don't have airflow over the VRM's and plan to do extensive compute work like video encoding or Blender CPU rendering or whatever, you might want even more than that if you plan on keeping the system for five years or more like many people do these days. Buildzoid is also very clear on when he thinks a VRM is way overkill for air or even liquid cooling, by the way. You really don't need to buy top end boards for a 24/7 home overclock and I really don't think he says that either.
|
|
#
?
May 20, 2018 23:45
|
|
|
FaustianQ posted:It's 40% more performance for same power or 60% less power for same performance on GloFos 7nm compared to 14nm. TSMC seems to be in the same ballpark. So if your motherboard is capable of running your CPU right now, you should be able to support one that's 40% faster (theorizing they manufacture one that faster) in the later EOL years? That doesn't sound too bad.
|
|
#
?
May 21, 2018 01:05
|
|
|
eames posted:Sony's devs are working on LLVM compiler improvements for the current Zen architecture. This should be good news down the road for all gamers as it seems exceedingly likely that the PS5 will launch with some sort of Ryzen CPU/APU. Piledriver? Nope, Jaguar. A chip designed for the fashionable netbook or tablet of the day.
|
|
#
?
May 21, 2018 07:18
|
|
|
5% uplift on Zen APUs with new drivers. notbad.png https://www.pcper.com/news/Graphics-Cards/AMD-Releases-Updated-Raven-Ridge-Desktop-APU-Graphics-Drivers
|
|
#
?
May 21, 2018 09:27
|
|
|
HalloKitty posted:Piledriver? Nope, Jaguar. A chip designed for the fashionable netbook or tablet of the day. Hahaha, you'd think so but clock for clock A4-5000 is less than 5% below a FX-4300 in single thread Cinebench 11.5.
|
|
#
?
May 21, 2018 09:31
|
|
|
AMD really should have just scaled up the Cat cores TBH, I mean that's without any L3 cache and halved L2, although the cat cores have 256kb to L1 to the 192kb L1 of Piledriver.
|
|
#
?
May 21, 2018 09:35
|
|
|
Arzachel posted:Hahaha, you'd think so but clock for clock A4-5000 is less than 5% below a FX-4300 in single thread Cinebench 11.5. It really only goes to show how much of a dog that whole Bulldozer architecture really was..
|
|
#
?
May 21, 2018 09:45
|
|
|
The most hilarious thing about bulldozer is that apparently it aged better than sandy bridge? Weren't there benchmarks recently showing that in brand new games, i7-2600k is just slightly ahead of bulldozers? I might be misremembering. It's only about 7 years too late in getting there, but I found it funny. 
|
|
#
?
May 21, 2018 10:45
|
|
|
I remember ordering a sandy the day bulldozer reviews launched. That was a sad day, I even had a nice sabertooth board waiting for it.
|
|
#
?
May 21, 2018 12:13
|
|
|
Truga posted:The most hilarious thing about bulldozer is that apparently it aged better than sandy bridge? Weren't there benchmarks recently showing that in brand new games, i7-2600k is just slightly ahead of bulldozers? I might be misremembering. It's a multithreading issue, an i7 still stomps the FX 8350 flat and it's why the i5-6600K and 7600K could find themselves struggling vs an R5 1600 in heavy game workloads (Crysis, Asscreed).
|
|
#
?
May 21, 2018 19:30
|
|
|
I still think the two cores per module idea had merit and could have been greatly improved on if AMD had the resources to see it through to maturity. Let's get a kickstarter going to fund a Cyrix reunion and provide them with the Piledriver IP and a fab partnership with IBM or Texas Instruments.
|
|
#
?
May 21, 2018 21:04
|
|
|
Truga posted:The most hilarious thing about bulldozer is that apparently it aged better than sandy bridge? Weren't there benchmarks recently showing that in brand new games, i7-2600k is just slightly ahead of bulldozers? I might be misremembering. I remember what you’re talking about but I thought it was vs the 2500k.
|
|
#
?
May 21, 2018 21:16
|
|
|
uh yeah. it was probably that, sorry
|
|
#
?
May 21, 2018 21:18
|
|
|
Bloody Antlers posted:I still think the two cores per module idea had merit and could have been greatly improved on if AMD had the resources to see it through to maturity. Let's get a kickstarter going to fund a Cyrix reunion and provide them with the Piledriver IP and a fab partnership with IBM or Texas Instruments. Reunion? They never broke up, they just took the tour to China Rastor posted:Centaur still Hauls, in China
|
|
#
?
May 21, 2018 21:41
|
|
|
Is the new Zeneration (get it?) due this year? I’m dreaming of a Zen+/Zen2 plus Volta/Ampere build but I’d rather not wait until 2019.
|
|
#
?
May 21, 2018 22:39
|
|
|
Zen2 is seemingly due 1H 2019, seems AMD will be focusing on Rome/EPYC2 first probably because initial yields on 7nm will still be pretty bad. God only knows when Volta/Ampere/Turing will happen. The two release opportunities seem June/July and September/October. But new Pascal poo poo has been coming in as mining has collapsed and prices have gone down so...
|
|
#
?
May 21, 2018 22:44
|
|
|
Boo. OK, thanks.
|
|
#
?
May 21, 2018 22:58
|
|
|
Oh, but Zen2 will be socket compatible with Zen, right?
|
|
#
?
May 21, 2018 23:06
|
|
|
Subjunctive posted:Oh, but Zen2 will be socket compatible with Zen, right? Yeah, should all be AM4 to 2020.
|
|
#
?
May 21, 2018 23:14
|
|
|
NewFatMike posted:Yeah, should all be AM4 to 2020. Great, thank you!
|
|
#
?
May 21, 2018 23:15
|
|
|
Subjunctive posted:Zeneration *summons the most condescending D&D tone possible* Yeah, this? Don't do this. Or else you're gonna be in AMDly of trouble with a ton of people ryzen against you. Central Processing Unit.
|
|
#
?
May 22, 2018 03:23
|
|
|
Bloody Antlers posted:I still think the two cores per module idea had merit and could have been greatly improved on if AMD had the resources to see it through to maturity. Let's get a kickstarter going to fund a Cyrix reunion and provide them with the Piledriver IP and a fab partnership with IBM or Texas Instruments. Khorne fucked around with this message at 05:46 on May 22, 2018 |
|
#
?
May 22, 2018 05:30
|
|
|
FaustianQ posted:Zen2 is seemingly due 1H 2019, seems AMD will be focusing on Rome/EPYC2 first probably because initial yields on 7nm will still be pretty bad. I dunno, AMD/GloFo was able to get pretty good 14nm yields pretty much right after moving to the node. Sure lightning probably won't strike twice but I don't think 7nm yields would be that dismal. Of course, unless they really do go hogge wilde and start pumping out 4.8-5.0 GHz parts for the same power and pricing as now. Then you won't be able to find them in stock until the socket is due to expire.
|
|
#
?
May 22, 2018 18:00
|
|
|
Seamonster posted:I dunno, AMD/GloFo was able to get pretty good 14nm yields pretty much right after moving to the node. Sure lightning probably won't strike twice but I don't think 7nm yields would be that dismal. They bought somebody else's fully functioning node for 14nm (after theirs failed), i dunno if thats gonna be comparable to making a brand new DUV node from scratch semi on their own. The research alliance apparently shares information, but their individual processes are separate. It's worth pointing out again though that Glofo is not making the first 7nm Zen2 parts that will ship, TSMC is. Glofo seems confident their process will show up in 2019, but AMD obviously thought better of putting all their eggs in that basket, to the level of spending who-even-knows how much money redesigning the entire chip for a separate process/fab. Tbh, I think thats a great move from AMD. Gives them huge flexibility if one/both/neither end up having issues.
|
|
#
?
May 22, 2018 18:11
|
|
|
The situation is that basically every other fab decided to share resources to get to "7nm" while Intel, alone, struggles with their 10nm, right?
|
|
#
?
May 22, 2018 18:17
|
|
|
Dual sourcing also mean AMD can meet demand easily. Considering Intel is going to be competing with 14nm offerings in 2019, that's...that might be a lot of demand.
|
|
#
?
May 22, 2018 18:17
|
|
|
PerrineClostermann posted:The situation is that basically every other fab decided to share resources to get to "7nm" while Intel, alone, struggles with their 10nm, right? TSMC isn't part of the alliance, so the two biggest fabs for the PC processing world are going solo, and the rest of the smaller fabs are sharing info.
|
|
#
?
May 22, 2018 18:25
|
|
|
I just caught the Adoree video on StoreMI and it looks pretty awesome. Are there any other good videos on it to see if it's actually all that?
|
|
#
?
May 22, 2018 18:57
|
|
|
Level1techs?
|
|
#
?
May 22, 2018 19:17
|
|
|
The computer of the day on woot is an HP "Gaming" Pavilion desktop. R7 1700, RX550, 16G of DDR4-2400, and a crappy 1TB HDD. But that's not a bad heap of componenets for $550. Edit: I'm more excited by what this represents, generally -- the beginning of endless cheap ryzen HP refurbs on woot. mdxi fucked around with this message at 22:54 on May 22, 2018 |
|
#
?
May 22, 2018 22:51
|
|
|
NewFatMike posted:I just caught the Adoree video on StoreMI and it looks pretty awesome. Are there any other good videos on it to see if it's actually all that? L1T.
|
|
#
?
May 22, 2018 23:45
|
|
|
https://www.youtube.com/watch?v=_3JHucl013Y Watching this to see if there is anything important in it. Doubling shipments from Q4 2017 to Q1 2018 sounds great depending on what numbers you're talking. EmpyreanFlux fucked around with this message at 06:39 on May 23, 2018 |
|
#
?
May 23, 2018 00:41
|
|
|
thanks for the broken video I guess
|
|
#
?
May 23, 2018 06:25
|
|
|
It was working when I posted it, but I hope I fixed my poo poo.
|
|
#
?
May 23, 2018 06:40
|
|
|
FaustianQ posted:Level1techs? It still seems good, which is pretty wild. I wonder if it's free on laptops with Ryzen mobile - StoreMI with a decent APU would make for one heck of a budget gaming machine. Lets you put more of that budget towards RAM.
|
|
#
?
May 23, 2018 13:10
|
|
|
|
| # ? Apr 19, 2024 04:21 |
|
|
Looks like Apple might be getting 7nm from TSMC for A12 this year. Does this mean 7nm is looking good for for Zen 2 as well?
|
|
#
?
May 23, 2018 14:45
|
|