|
If you need a good chip right now just buy an intel product and get it over with.
|
#
?
Oct 31, 2016 19:46
|
|
|
|
| # ? May 15, 2024 12:03 |
|
|
Just limp for a few months with some cheapo CPU, evaluate after CES which is supposedly the launch date with mass availability by February.
|
|
#
?
Oct 31, 2016 20:38
|
|
|
Was WCCFtech the site that gets everything right because they publish literally every rumor, and thus have at least one correct? http://wccftech.com/new-amd-zen-blender-benchmarks/
|
|
#
?
Nov 3, 2016 06:33
|
|
|
That would be lovely if it is true. AMD getting good again would be awesome, my Phenom 840 was amazing and I miss it.
|
|
#
?
Nov 3, 2016 07:05
|
|
|
I saw that article, but dismissed it as not-news because the Xeon E5 2680 v2 they're comparing it to is 22nm Ivy Bridge EP. If that's the best they can do on 14nm, then things are dire indeed.
|
|
#
?
Nov 3, 2016 11:14
|
|
|
SwissArmyDruid posted:I saw that article, but dismissed it as not-news because the Xeon E5 2680 v2 they're comparing it to is 22nm Ivy Bridge EP. As in the test between an Ivy Bridge Xeon and a Zen processor came out equal? If so then I guess goodbye amd CPU division.
|
|
#
?
Nov 3, 2016 11:16
|
|
|
No, as in the unnamed chip and the Intel chip that wccft found to have an identical score. A lot of the sheen is going to come off the Jim Keller name, that's for sure, though. SwissArmyDruid fucked around with this message at 11:23 on Nov 3, 2016 |
|
#
?
Nov 3, 2016 11:20
|
|
|
Well, according to that benchmark it beats out some of the low clocked 12 core Haswell Xeons, so it might not be completely useless i guess 
|
|
#
?
Nov 3, 2016 12:29
|
|
I CANNOT EJACULATE WITHOUT SEEING NATIVE AMERICANS BRUTALISED!
|
blowfish posted:Well, according to that benchmark it beats out some of the low clocked 12 core Haswell Xeons, so it might not be completely useless i guess Really depends on the amount of cores I guess. 
|
|
#
?
Nov 3, 2016 12:34
|
|
|
SwissArmyDruid posted:If that's the best they can do on 14nm, then things are dire indeed. Intel's focus on performance per watt for the last few generations is why many are still running Sandybridge after all. There just hasn't been any reason to pay Intel prices for the level of performance improvement they've been offering for a long time with their newer chips. So if AMD prices their stuff right then they're fine. If they try to get top dollar then yeah they're screwed.
|
|
#
?
Nov 3, 2016 12:44
|
|
|
Isn't Samsung 14nmFF fairly similar to Intels 22nmFF? If so then Ivy-Haswell single core performance make sense I suppose, and apparently Broadwell level multithreaded.
|
|
#
?
Nov 3, 2016 20:14
|
|
|
Given that we're basically going to see Intel stuck on the same architecture for two years, too...
|
|
#
?
Nov 3, 2016 20:28
|
|
|
PerrineClostermann posted:Given that we're basically going to see Intel stuck on the same architecture for two years, too... Not in server-land, the Purley platform with Skylake-EP and EX should be a big loving deal. AMD could be arriving with a good Sandy / Ivy / Haswell / Broadwell server competitor finally when Intel changes platforms for the first time since 2012.
|
|
#
?
Nov 3, 2016 20:33
|
|
|
Twerk from Home posted:Not in server-land, the Purley platform with Skylake-EP and EX should be a big loving deal. AMD could be arriving with a good Sandy / Ivy / Haswell / Broadwell server competitor finally when Intel changes platforms for the first time since 2012. And that's the crucial bit, because if server farms are going to have to move to a new platform anyways, why *not* see what the competition can bring, instead of "Oh man, we'd totally have gone with you guys... except that we JUST bought the new Intel poo poo."
|
|
#
?
Nov 3, 2016 20:49
|
|
|
We appear to have some new leaks, saw these linked on Dresdenboy's twitter so not just random crap here.quote:Apologies if there's another thread about AMD's upcoming CPU architecture and of course desktop CPUs, if there is, delete or perhaps merge this info. Its a 4 page thread so I just c/p'd the most relevant or interesting parts here. The BIOS being on package/die and requiring 30 min. to clear is one of the oddest tidbits to come out so far about Zen. Maybe just a thing for their engineering samples? I dunno.
|
|
#
?
Nov 10, 2016 05:07
|
|
|
Apparently this leaker has some credibility which is interesting. Sounds like AMD is still struggling with GloFo (rebrand to "Albatross Industries") if they're getting 4.1Ghz stable and 5.2Ghz stable chips (likely lots of massive binning for 2017 up until close to 2018 or into 2018). How much of a performer would a Haswell-E chip be with a loss of 30% of it's performance? Also 299$ for the "6850K" 3.0Ghz 8C/16T chip running around apparently, that seems...lol, to get what he's claiming is essentially the 5960X. Like, that's Zen 4C/8T competing versus i3s, holy poo poo, I just can't see that happening, not with Haswell-E single threaded performance. That'd mean 6C/12T versus i5s (if those exist and I'm not sure why they wouldn't), and he says prices will go up to 699$ for likely the top tier handpicked Black Editions (Likely your 5Ghz on air ones). If 4C/8T are hanging around ~150$, I guess this forces Bristol Ridge to be all under 100$ to be competitive. He also makes it sound like board partners are excited because they can put in minimum effort for maximum return, like they can easily be made as cheaply as AM1 boards (I'm sure OEMs are wetting themselves over this) and the biggest difference will be VRM count and cooling. I'm already half way to ODing on this years election, I'm not buying it because it'd be too good of a reason to replace my current rig will a full AMD one and good things don't happen in the mirror universe. gently caress that performance and price expect 3000$ 24C/48T and 4000$ 32C/64T server CPUs to undercut Intel heavily and still make profit. Bogus, unreal, not a chance. Not one bit of salt from me.
|
|
#
?
Nov 10, 2016 06:48
|
|
|
I don't believe hexacores are a thing, IIRC the Zen blocking comes in 4C/8T blocks, no duals or half steps. Haswell-E performance has my eyebrow raised as well. Cubs won the World Series, Trump is president, weird poo poo is happening. Maybe AMD really have it.
|
|
#
?
Nov 10, 2016 11:03
|
|
|
NewFatMike posted:I don't believe hexacores are a thing, IIRC the Zen blocking comes in 4C/8T blocks, no duals or half steps. That seems odd to me because while CCXs would be 4C/8T each, it's still made of 4 individual cores and 2 being faulty but 6 not across the chip would seem to indicate making a 6 core chip rather than further disabling cores to a 4 core chip, sell for more and get more out of it. I dunno, maybe there is something to Zens design that would prevent this but that seems to be unusual and bad planning considering that's a whole lot of chips to start tossing. I mean I get with why bothering with a 2 core 4 thread chip because you've disabled so much already but a 6C/12T chip makes sense financially at least.
|
|
#
?
Nov 10, 2016 11:14
|
|
|
It's me, I'm the one that buys BE chips even though I have zero desire to overclock. (at the outset at least) I'm with Fat Mike on this, though. With Zen's design coming with core complexes in multiples of four, the only way I expect to see hexacore parts is if they have too many failed parts, and they want to make some money off otherwise defective silicon like the Phenom II X2 and X3s, (GloFo gonna GloFo) as opposed to designing a proper price point for them. Realistically? Assuming they designed the chip with this in mind, I expect AMD to saw it in half and sell a 4c/8t + a 2c/4t part, instead of having to build fallback cases for their ring interconnect to shuffle between four good cores and two bad ones way over there. (And that's assuming that the defective cores are the two that are farthest away from the known good core complex on an 8-core part. If the two defective cores were diagonally opposed to eachother? Whoo doggy, that's not a latency problem I want to be responsible for. No, better to saw the bad core complex off and sell it as a lower-spec dual-core.) (I don't even want to consider the case of two core complexes with three good cores each.) And I absolutely do NOT see them doing things like locking off cores on perfectly good chips to hit a lower-tier segment either, a la... gently caress, what was it. The 960T? SwissArmyDruid fucked around with this message at 11:26 on Nov 10, 2016 |
|
#
?
Nov 10, 2016 11:18
|
|
|
quote:- AM4/ZEN uses an SOC design, that means even CMOS/BIOS configuration is on package (not necessarily on silicon, I can't confirm this) so it is possible to clear the "BIOS" and still have old value applied 30 minutes later. How this will be addressed remains to be seen. Perhaps it won't be the same scenario for final silicon I for one look forward to botching my BIOS update and bricking my CPU.
|
|
#
?
Nov 10, 2016 11:30
|
|
|
Boiled Water posted:I for one look forward to botching my BIOS update and bricking my CPU.  Just because it's on phones...
|
|
#
?
Nov 10, 2016 11:37
|
|
|
NewFatMike posted:I don't believe hexacores are a thing, IIRC the Zen blocking comes in 4C/8T blocks, no duals or half steps. given glofo, they might be core harvesting
|
|
#
?
Nov 10, 2016 11:48
|
|
|
Anime Schoolgirl posted:JIM! At least the chipset is also on the chip, purpose of this being ... I don't know. I recall Intel trying this strategy (and/or VRMs on the cpu?) but never really settling on anything.
|
|
#
?
Nov 10, 2016 11:55
|
|
|
Having the chipset, especially when combined with the gpu, all on the cpu, is a huge huge thing for HPC. I know some HPC people, and they've been waiting for this for over a year, even to the point of delaying some upgrades, because they got to play with some samples. One guy works with CERN on a regular basis and he's saying it's the most requested thing out of there right now. Even if zen somehow craters on consumer market as bad as bulldozer (though I presume microsoft/sony are all over this poo poo for the next console ~pro~), it doesn't look like it'll matter that much in the long run for AMD. Their consumer share is already poo poo.
|
|
#
?
Nov 10, 2016 12:21
|
|
|
It may be just me but here is hoping that it's not directly on chip and there is at least a dual bios (one you can gently caress with and one locked out from all but the savvy).SwissArmyDruid posted:It's me, I'm the one that buys BE chips even though I have zero desire to overclock. (at the outset at least) Then theoretically another way to differentiate boards is core unlocking. Get that sweet 5th core and massively gently caress your latency, hell yea. EmpyreanFlux fucked around with this message at 12:23 on Nov 10, 2016 |
|
#
?
Nov 10, 2016 12:21
|
|
|
Truga posted:Having the chipset, especially when combined with the gpu, all on the cpu, is a huge huge thing for HPC. I know some HPC people, and they've been waiting for this for over a year, even to the point of delaying some upgrades, because they got to play with some samples. One guy works with CERN on a regular basis and he's saying it's the most requested thing out of there right now. How does chipset on CPU help in an HPC system?
|
|
#
?
Nov 10, 2016 12:33
|
|
|
Latencies, I'd imagine? Also definitely power efficiency.
|
|
#
?
Nov 10, 2016 12:35
|
|
|
AMD is re-using the SeaMicro interconnect fabric, right? If everything is entirely self-contained inside each package, that speaks to ease of scalability in HPC, doesn't it?
|
|
#
?
Nov 10, 2016 12:42
|
|
|
If the evolution of SeaMicro's interconnect is GMI, yes Zen uses that.
|
|
#
?
Nov 10, 2016 12:47
|
|
|
Undercutting Haswell-5960X by 3% (undercutting the 6900k by 10%) for the 8core would be a pretty good intro price for the premier 8 core SKU, and the board is likely a lot cheaper though you will only have double channel memory. I'm more interested in how the "runt" bins are going to work though. Zen is 4 core blocks, and I presume it needs all of them to actually function in some cases like AVX. Probably going to see Athlon/Sempron dual cores with the bells and whistles disabled like the Pentium/Celeron chips. Anime Schoolgirl fucked around with this message at 13:07 on Nov 10, 2016 |
|
#
?
Nov 10, 2016 13:05
|
|
|
FaustianQ posted:That'd mean 6C/12T versus i5s (if those exist and I'm not sure why they wouldn't) It's quite possible Intel has been holding back due to lack of competition and would also be able to release a consumer-socket 6C/12T in pretty short order.
|
|
#
?
Nov 10, 2016 23:36
|
|
|
Exciting for sure if true. Will be interesting seeing how this Bios on a CPU goes, will it also in a way remove the need for a CMOS battery or something? If they can bring the 5960X performance for 1/2 the price, that is going to do really well to bring that virtual kick to the nuts of Intel. The lack of Quad Channel memory will be a downer, but if the Gaming, Encoding and VM performance is still within throwing distance for 1/2 the price, that would still be pretty good IMO. The overclocking inside Windows though I find is going to be a iffy thing. I guess you get to see results instantly, but when you throw it out of range, I hope it has a good failsafe so you can boot back to stock and try again instead of having a sort of bootloop of death if you screw something up. Hmm
|
|
#
?
Nov 10, 2016 23:37
|
|
|
EdEddnEddy posted:Exciting for sure if true. Will be interesting seeing how this Bios on a CPU goes, will it also in a way remove the need for a CMOS battery or something? This is the assumption apparently, that and it should have a dual BIOS to alleviate all problems. A little concerned though that they'd have potentially low security compared to Intel chips. FuturePastNow posted:It's quite possible Intel has been holding back due to lack of competition and would also be able to release a consumer-socket 6C/12T in pretty short order. I doubt it, Skylake-X has a planned 6C/6T chip and I don't think Intel is as price flexible as people assume.
|
|
#
?
Nov 11, 2016 00:11
|
|
|
FaustianQ posted:I doubt it, Skylake-X has a planned 6C/6T chip and I don't think Intel is as price flexible as people assume. Skylake-X isn't even on the mainstream cheap socket.
|
|
#
?
Nov 11, 2016 00:12
|
|
|
EdEddnEddy posted:The overclocking inside Windows though I find is going to be a iffy thing. I guess you get to see results instantly, but when you throw it out of range, I hope it has a good failsafe so you can boot back to stock and try again instead of having a sort of bootloop of death if you screw something up. Hmm My hope is that this would be done through an application which can be set to start on bootup but by default would need to be started manually. Once you've stress tested and are feeling good about your OC you could set it to auto, and if anything goes wrong you could use safe mode to recover.
|
|
#
?
Nov 11, 2016 00:31
|
|
|
FaustianQ posted:A little concerned though that they'd have potentially low security compared to Intel chips. Putting the BIOS on CPU/package for a consumer x86 chip is weird but I don't see a inherent security disadvantage from doing things that way at all. Of course I don't see much in the way of advantage for AMD there either. Maybe it makes it easier for them to do support/updates in the future through the OS?? I dunno, just guessing there. If they're going to have a secondary BIOS to control other stuff I anyways I don't see any cost savings for the mobo guys. The strangeness of it is what made me wonder if its just for the ES chips originally. edit: the software could probably detect if windows wasn't properly shut down and on reboot it'd just go to default clocks as a recovery mode and ask if you want to reapply custom settings. PC LOAD LETTER fucked around with this message at 04:10 on Nov 11, 2016 |
|
#
?
Nov 11, 2016 03:58
|
|
|
Watch as it applies the overclock when the service starts up in Windows and makes it impossible to overclock an AMD processor and then use Linux on it.
|
|
#
?
Nov 11, 2016 03:59
|
|
|
PC LOAD LETTER posted:How so? Isn't there plenty of stuff you can do to Intel chips through Windows software to screw things up if you want to be malicious too? I'm thinking that it just enables people to permanently brick the processor which is vastly more costly compared to boards. I should rephrase and say I think they'd be more vulnerable to attack, both from permanently shutting them down to and costing way more per attack compared to Intel; but I don't run a datacenter and can't evaluate whether this is a serious potential con.
|
|
#
?
Nov 11, 2016 04:09
|
|
|
FaustianQ posted:I'm thinking that it just enables people to permanently brick the processor OC software lets you change all kinds of things with Intel chips too in Windows if you're talking about stuff like hackers changing the volts to kill the CPU. Of course Overdrive/OC software would have to be installed for that to work. I don't think that will happen in a server environment. There may be some sort of vulnerability vs Intel chips but its just not apparent to me from the leak details. For all we know its only consumer chips that allow overclocking that may have any vulnerabilities here. Their sever chips could very well be locked up to prevent OC'ing. Though if they allow undervolting/clocking on the server chips, which I think they normally do, I guess a hacker could make everything run reeaaaalll slooooowww. But then that would apply to Intel chips too.
|
|
#
?
Nov 11, 2016 04:22
|
|
|
|
| # ? May 15, 2024 12:03 |
|
|
HALF A GODDAMN HOUR TO RESET YOUR BIOS?! gently caress, I thought you guys were kidding! The gently caress are they doing, are they including a UV flashlight with your processor so you can shine it on there like an FPGA?!
|
|
#
?
Nov 11, 2016 04:31
|
|