AMD Zen
Images are sometimes not shown due to bandwidth/network limitations. Refreshing the page usually helps.
You are currently reading a thread in /g/ - Technology
You are currently reading a thread in /g/ - Technology
https://www.youtube.com/watch?v=kpEK8rOjC6A
Does the black man speak the thruth?
Turn over the polaroid and read "Don't believe AMD's lies" again
>>53258738
You rekon its still worth waiting to see if it at least makes intel price their kabby lake cpus more competativly?
>>53258752
If you want a competitive CPU get a 4690K.
I've never heard of Intel dropping their prices for AMD, not even Nvidia does that.
>>53258764
>Not even Nvidia does that
What is the 970, 980 and specifically 980Ti for £100, Ken?
>>53258730
AMD reckon they can achieve at least (if not more) than 40% increase on previous releases.
He's not lying about the numbers; he's completely correct if we assume that AMD was being honest about their improvements.
IF AMD can turn it around and put out a product that even rivals Haswell, they have a damn good chance.
Honestly, I think they've done it, due to the fact they got Jim Keller back to do it for them.
Still, the only way to know for sure is to wait for the actual release.
But for your original point, yes, the black man in the video was being completely honest with the figures provided to us.
>>53258764
Il bide my time and get an i7 at least
>>53258730
IPC means jack shit without clockspeed. I can have 1 ipc and still beat the last intel cpu if I have 1THz processor.
We need to know the clock speed before doing any calculation.
>>53258730
Call me illiterate, my ee-class was a year ago and I forgot most of it, but wouldn't one cycle allow for one instruction per core?
The main speedups (at least from what I remember, and what our simplified CPU had) come from pipelining, jump prediction, efficient low-level cache usage, microcode efficiency (and afterwards of course clock speed too).
>>53258812
>AMD reckon they can achieve at least (if not more) than 40% increase on previous releases.
I actually have a pretty sneaking suspicion that this is through VISC-like behaviour, as in; it takes single threads and does magic to them to run them on multiple threads- technically improving IPC.
However, this type of performance improvement only works in benchmarks, and AMD's claim is nebulous, we have no idea what was ran to get that improvement.
Likely highly optimised benchmarks.
Like I said earlier, turn over the polaroid you took of AMD. On the back it says "don't believe their lies"
No offense to AMD though, every single polaroid I have of tech companies says that on the back.
>>53258840
They're running on known silicon so the clockspeeds will be in the 3-4GHz range.
So IPC actually matters a fuckload since we know the technology the chips are going to be on.
>>53258853
If they were still on Piledriver based architecture, I'd agree, but they're not.
Zen is a brand new architecture hand designed with a team led by Jim Keller.
Jim Keller, aka, the most respected, sought after CPU engineer in the business alive today.
>>53258853
stop taking polaroids of companies, anon. you're creeping people out.
>>53258864
That's the problem anon, to do a VISC-like trick they need an entirely new architecture.
>>53258885
Nah, they're all about getting as much multi-threaded performance as possible also. I don't see them sacrificing multi-threaded performance just for single thread.
>>53258730
every speculation is based on muh 40% and we have no idea if that's true.
wouldn't even be mad if it wasn't since marketing always does this kind of shit and that little part of me that was susceptible by hype died long ago.
>>53258812
>What is the 970, 980 and specifically 980Ti for £100, Ken?
nvidia didn't drop the prices of those GPUs at all, they're still the same as their launch MSRP was in 2014. the 970/980 actually forced the price of the 290 and 290x (down from $430 and $550 respectively), the fury x was also rumored to be $850 before nvidia launched the 980ti and destroyed AMD.
>>53258895
VISC-like tricks don't sacrifice multi-thread performance. They split up single threads into multiple ones, but the effect on things that are already multi-threaded is almost zero. I recommend listening to the tech report's podcast on VISC. Awkward host aside, they have someone on who's an actual industry expert talking about it.
>>53258910
Oh? I figured it'd potentially damage multi-thread performance if you were to be multi-tasking, which I guess is still very possible but then again, if you multi-task to that level, you're going to expect some degradation.
Either way, even if they do use tricks like that to improve the 'single' core performance; if it works universally and doesn't need specially written code, would it really be an issue?
As long as AMD can bring themselves back to a respectable level, I don't personally care how they go about doing it.
>>53258932
It might even improve multi-tasking in lots of cases, ie; a program that's programmed to use 4 cores is put on a processor that has 16 and the VISC-like system distributes what it can and speeds up that. For infinite CPU multithreading there's no difference at all.
>if it works universally and doesn't need specially written code
That's the problem. It usually works very badly on general purpose and code not specifically written to take advantage of that system.
You're right in that I don't care how they improve performance either, but their benchmarks could be some tricky shit that just improves array sorting in a highly controlled environment but no benefit to complex and unpredictable things like browsers, games and general purpose. Since that's how that technology works- to spread the threads out it needs to predict parts of the single thread, the more predictable the better performance improvement. So benchmark cheating is super useful.
It could even mean that they'd have like a selection of 10 benchmarks- an 80% improvement on 5 of the synthetic ass ones and a 0% improvement on the real-world general purpose ones and then they say it's a 40% IPC improvement overall.
Sneaky fucks.
>>53258900
>fury x was also rumored to be $850
That was just bullshit obviously. Like GPU/graphic cards stories on worse tech sites were never made up before.
(I write about this for one site, so I have pretty good idea of the SNR in this field. GPU news are full of speculations and made up stuff, while CPU news tend to be more solid. It's because GPUs are of interest to fukcing little gaymer kiddos I reckon.)
If they manage to increase the IPC while still keeping the relatively high clocks and number of cores they'd be better than haswell. DESU I'd be happy even if it matched haswell, I guess we'll have to wait and see.
>>53258730
This is the person in the OPs video, he is a moron.
>>53258730
A real black man would never support AMD! A real black man HATES THE WHITE DEVIL and thus would only buy from a company that discriminates against whites.
That nigga is obviously not a nigga!
KILL WHITEY! BUY INTEL!!!
>>53258730
>Does the man speak the truth?
No. Zen is said to increase performance by 40% on the same die size (32nm), he isn't taking the IPC gains from the die shrink into account. Zen will almost certainly have higher IPC than Skylake. Moreover I expect them to push the clock rate fairly high (4Ghz+), and the core count will be higher than Intel's current gen. We're talking up to 32 cores here, and these aren't half cores like Bulldozer, they're the equivalent of Intel cores. Plus Zen will have actual hyperthreading, meaning at the upper end of the scale, we could see something like this:
Zen IPC: 180%
Turbo boost: 4.2Ghz
32 cores
64 threads
Realistically they may be keeping their 32 cores for server cpu's. I do expect no less than 12 cores at the high end desktop level though.
>>53259703
I am optimistic about Zen. It looks very promising.
But you're overhyping this something fierce and your expectations are unrealistic. You're also making several assumptions that make you look like a complete and utter idiot.
My prediction will remain as: It will be roughly the same as Skylake.
>>53259828
Literally nothing of what he posted came from AMD. AMD claim a 40% IPC increase over excavator. That idiot seems to think die shrinks have a direct effect on IPC. He's also pulled a 180% figure out of his ass, somehow.
>>53259831
Best case is it'll be better than sandy bridge.
If you hope for more, you're both retarded and in for a shock when it comes out and it's ass.
Yes based keller, no AMD will find a way to fuck it up.
Bump. I'm enjoying this thread.
AMD need to get their shit together. I want to upgrade my FX 6300, but I don't want to buy Intel.
>>53258812
>>53258864
yeah they had Jim Keller back for 3 years but he left right when his contract was up. not sure that's such a good sign for Zen.
engineers are usually not the type to drop the mic and leave after making something new.
>All this hype for ZEN
>>53258730
Practical CPU performance is never the same as theorized
>>53260894
Keller is pretty much freelance. Him leaving means he was done. What would you have him do, sit around twiddling his thumbs waiting for the rest of the work for a CPU to be done?
Keller leaving doesn't mean Zen will be bad or anything like it. It means that AMD doesn't have a processor after Zen that Keller can work on.
Zen is AMD's last processor. AMD has nothing else in the pipeline but iterations of Zen
amd you better not fuck up this time
>>53258730
There is only one thing that anyone needs to know about Zen, and it came straight from Lisa Su just a couple weeks ago, but everyone on /g/ missed it since they're too stupid to read any industry news.
AMD's upcoming Zen based Opteron line of processors will address 80% of the Market. In other words she was saying that their chips will be competitive in 80% of all common enterprise workloads, and they would lose out in the other 20%.
AMD isn't going to touch intel's FPU performance so thats a given.
>>53259703
Excavator has 40%~ higher IPC than Excavator
Excavator is only found in 28nm Carrizo/Bristol Ridge
Process node has literally nothing to do with IPC
Stop talking out of your ass, jesus christ.
>>53261634
Zen* has 40%~ higher IPC than Excavator
>>53261634
Oh, and Summit Ridge is only an 8 core part.
There is no 12 or 16 core consumer chip coming, and AM4 only supports dual channel memory
>tfw my $299 shitposting machine has fm2+ zen ready socket.
>>53261690
There are no FM2+ Zen parts. AM4 only.
>>53261554
He could've worked on finishing K12 instead of abandoning them, like a rat leaving a sinking ship.
>>53261717
AMD isn't made of money
>>53261717
Keller was a department head at AMD, he was in charge of the teams doing the grunt work, he wasn't directly working on the designs himself. The man isn't a low level technician.
Zen and K12's high level designs were long finished before he left the company. Work on the Zen+ core was already underway by March of 2015 when AMD gave their last FAD presentation.
>>53261741
Pretty sure they can afford to pay one guy if they really wanted to...
>>53261705
Fuck. I was just going by a chart I saw on wikipedia.
>>53261771
When that "one guy" is worth more than several research teams then maybe they don't. Also they might not need him anymore, did you think about that? That he just did his job?
>>53261802
keller tends to jump around companies a lot, either he is hard to work with or he doesn't like to stay in one place for long.
>>53261802
I think it'd be worth it just to stop Intel from scooping him up.
>>53258730
seems mostly impartial and conservative in its speculation
even at equal or slightly lesser IPC zen is projected to have higher core count
with dx12 and vulkan the gamble might work for them this time around
>>53261828
Or he gets paid several million dollars per short term contract and is an expert on making himself independently wealthy
I was looking at some comparisons and AMD wipes the floor with Intel when it comes to audio and video encoding. It also has better onboard graphics. It only performs slightly less well in games and some more esoteric apps. For cost at initial checkout it seems AMD has the edge. But in terms of heat, power and gaming Intel/Nvidia are the better option (mainly because of Goyworks and having more buying choices when it comes to Gsync monitors). If cost and use for intensive multimedia applications is what you are after AMD is a better choice.
I really see no big gains with Zen and Intel will have it's next CPU out soon after anyhow.
>>53261901
Zen won't make Vulkan or DX12 much faster than a deddy GPU will which is what most gamers are interested in anyhow. If Zen falls in the middle of Broadwell and Skylake the only advantage I see is on price. Basically 'OK' for poorfags.
>>53261947
keep in mind that's between broadwell and skylake with 8 real cores. expect Intel to finally roll out 8 core consumer chips at a decent price to keep AMD from getting too competitive.
>>53261947
So, an 8 core "Broadlake" CPU for the price of a 4 core Skylake?
I'm sold. Being a poorfag sure has its advantages.
Also,
>middle of Broadwell and Skylake
As if there is any performance difference between those two.
>>53261989
>>53261947
Summit Ridge isn't going to be cheap, and intel will never price their i7 Extreme line for the mainstream market.
Both product lines are high end desktop, not intended for the mainstream market. They will never voluntarily decimate their own profit margins here.
>>53261947
the new APIs signify a change of direction in the industry towards multi-threaded workloads in game engines, basically what we've been expecting for half a decade. it took three gens for intel to beat piledriver at it (not that they were trying particularly hard).
you can see vulkan showcases on youtube that display CPU usage by core. there is a fuckhuge difference. it directly benefits from moar coars rather than fast single thread.
>>53261999
what >>53261989 says is probably true
intel has quasi-infinite R&D. they're holding off on tech to lay down the smack on AMD should they stick their neck out again.
that, or they've been so busy trying to break into the mobile market they don't have anything new for the x86 desktop.
>>53262011
I'm willing to bet good money the high end Zen CPUs will be priced around the i7 K-line chips, not X.
>>53262043
AMD should know there's zero point to it
the kind of retards that build X99 SLI Titan setups are fidelized to the marrow
>>53262043
The high end Summit Ridge will likely be around $500 with the high end of Raven Ridge around $200-$250.
AMD isn't going to be the cheaper alternative, came straight from Lisa Su.
>>53262031
Remembering that they've spent billions on developing chips for mobile and trying to push companies to use these chips by ether giving them out for free or flat out paying companies to use them, I wouldn't be all that sure that they have something to genuinely lay the smackdown on AMD just waiting to be used if they challenge them again.
My personal feeling is that it's just going to end up like it was in the 90's where AMD was cheaper, but Intel had the fastest chips if you could afford them.
>>53259531
your the retard here
>>53262092
>My personal feeling is that it's just going to end up like it was in the 90's where AMD was cheaper, but Intel had the fastest chips if you could afford them.
this isn't feasible in the late 2010's
even the average ledditor can make an informed decision now. if the price/performance ratio isn't there, it's dead. both intel and AMD will have models at every price point (except maybe the ultra-high-end in AMD's case)
there's some truth to what you said if you consider the 6300 and how it competed with similarly priced i3's for so long. that chip is probably the main reason AMD survived piledriver.
>>53262155
I'm fairly sure Intel can keep carving out a dominant marketshare despite AMD being capable of giving better price-performance. That's what they did around 2009 when it was K10 (Phenom II) vs Nehalem and I can't see why this couldn't be repeated.
>>53258752
Intel dropping their prices would be admitting that AMD has similar products for a cheaper price, which would be detrimental to their sales because a good chunk of their market share is due to the fact that Intel is viewed as superior in most of the people's mind, at every level, when this is not globally true.
>>53258730
>Begging for likes
Typical nigger, always demanding handouts.
>>53259797
Nice list of examples and reasoning you've got there, asshole.
>>53261634
Explain then the incremental IPC improvements Intel has been enjoying due to die shrinks. Or are you going to claim a 1.2Ghz Pentium 3 at 130nm is going to have equal IPC to modern Intel processors at 14nm?
AMV literally said that Zen would have 40% higher IPC than their previous gen ON THE SAME PROCESS NODE. This 40% increase in performance does NOT account for the additional performance increase Zen will see due to the die shrink.
Yes I'm being otherwise hopeful, maybe overly-so, but I do think that most of you are underestimating Zen, particularly in light of the man behind the scenes. So what if it's only 8 cores. If they price match it with Intel's 8 year old quad core line (Nehalem, and note Skylake is STILL fucking quad core, it's pathetic) then we've got a good upgrade.
8 core Zen vs 4 core skylake, at roughly equal price and IPC. Zen wins.
>>53262306
>I'm such a tech illiterate retard that I have literally no idea what any of these words mean
IPC = instructions per clockcycle
The process node has literally no impact on this whatsoever, it is a function of the architecture design itself. If you don't understand something this absurdly simple then you don't belong on this board.
>>53262321
I do know that you moron. Let me spell it out to you:
Higher IPC = improved performance.
Separately, smaller process node translates to improved performance as well.
>>53262306
>HURRRRRRRRRRRRRFFFFF
A smaller process lets you use less voltage per clock, you can hit higher clocks in a given power envelope. It does not impact IPC in any way, shape, or form.
Architectural changes and process shrinks are totally independent from one another.
You're such a tech illiterate retard that you're conflating entirely different things as if they're the same.
All that AMD stated is that Zen would have 40% higher IPC than Excavator. Everything else you're pulling out of your ass. Tech. Illiterate. Retard.
>>53262217
yeah well that's just capitalism
although even AMD is big to the point where it can take a lot before disappearing even now.
>>53262237
not really, you would be surprised just how fidelized people can be
some of the people I work with... holy shit...
>I'll literally buy X no matter what happens because I heard through hearsay that Y sucks 15 years ago
in the server market it's even worse. people would consider POWER8 chips over opterons.
>>53262258
literally every youtube channel does this
go right back to /pol/ with the rest of the sub-100-iq dregs
>>53262306
haswell's got 8 core CPUs though, maybe we'll see a new flotilla of those at lower prices
Do people still give the benefit of doubt for AMD considering that they haven't delivered in years? Watching /g/ claiming that Zen is the next best thing since sliced bread is very amusing to me
>>53262375
you sound mad as fuck. can you behave like an adult?
>>53262473
>gets his shit pushed in
>lel u mad
Right.
>>53262473
>tech illiterate retard shitposting with his back to the wall after proving how clueless he is
What a surprise
>>53262470
we're being optimistic about zen because the prospect of it failing is too horrible to imagine
unless you literally enjoy the current trudge that is intel's lineup
>>53262488
not the same guy, stop being a feisty cunt
>>53262519
>we're being optimistic about zen because the prospect of it failing is too horrible to imagine
this, I want to get off of Mr. Intel's wild ride
>>53262519
M8 if they manage to pull a Haswell equivalent core per core, use the number of cores they use today (6-8) for consumer grade CPUs and implement Simultaneous Multi-Threading (basically hyper-threading) they'll push Intels shit in.
>>53262375
Then why did AMD specifically mention the phrase "40% higher IPC at the SAME PROCESS NODE". Read into what they're saying. They're claiming 40% improved performance if Zen were on 32nm. Why would they specify this? Doesn't it seem a little odd to you?
>>53262790
>tech illiterate retard just keeps making an ass of himself
https://www.youtube.com/watch?v=sgqE2bbmvWw
>>53262838
Heres the only other official mention from AMD
http://www.overclock3d.net/articles/cpu_mainboard/amd_s_zen_will_have_a_greater_than_40_ipc_improvement/1
Greater than 40% uplift comes from comparing Zen based Opterons against their now outdated Vishera based Opterons, instead of any Excavator based chip.
Note that AMD doesn't make any tech illiterate retarded statements conflating IPC and process nodes.
>>53258730
>AMD
>>53262596
They use 4 physical cores, just like intel.
It would mean they would be equal though.
>>53263722
ADD MORE CORES
>>53263722
They use two cores paired together under one scheduler sharing a 1+1 fpu core as well, the 8 core terminology is correct. Which is funny because hyperthreading is essentially intelligent scheduling of core resources.
>>53259703
You will be disapointed if you put up your hopes that high. It's better to keep your expectations low to be pleasantly surprised if they do end up delivering.
>>53262838
>>53262910
In no way did I confuse IPC improvements with process node improvements. Quit being a gigantic jackass.
>>53264115
If I put my hopes too low I might end up buying a Skylake, which I'm hoping to avoid. Even if Zen only matches Haswell, I'd rather go with Zen than Intel.
So what will /g/'s excuse be this time when Zen turns out to be mediocre as fuck like Fury did?
>>53262306
>Nice list of examples and reasoning you've got there, asshole.
I obviously don't need to give any examples. I pointed out your stupidity and /g/ took over. Seriously, where the fuck did you get the
>Zen IPC: 180%
Bullshit from? That's absolutely ridiculous. Your wild guesses at clock rates are also laughable as AMD have made no statements about it whatsoever.
Like I said, I'm a blind AMD fanboy but you're just an idiot. I'm halfway tempted to accuse you of being an Intel shill, overhyping expectations so you can point to it after it's released and say "See? They didn't live up to my unrealistic expectations!"
>>53258858
>muh IPC
clock speed is also influenced by architecture. You can sacrifice IPC for clock speed and vice versa.
There is also cores and Zen will have a moar cores advantage. Current rumors/leaks:
Zen Opteron: 32 cores
Broadwell-EP: 22 cores
Broadwell-EX: 24 cores
Skylake-EP: 26 cores
Skylake-EX: 28 cores
Intel will not reach 32 cores before they go to 10nm.
>>53259531
hes right though
>>53267014
Pretty sure the die shrink will bring an IPC increase as well. Think what you want, time will tell.
>>53267689
>Pretty sure the die shrink will bring an IPC increase as well
No, it won't. Making a transistor physically smaller does not change the interaction of several transistors or the chip as a whole. Could you please explain why you hold the idiotic belief that making a transistor smaller will impact IPC?
>>53259549
>Zen is looking promising
>its competitive against intels chips from 2 generations ago
yeah brace yourselves for disappointment.
it will be the typical AMD story
>great for certain applications that can make use of all the cores
>poor/mediocre performance when it comes to single threaded stuff
>>53267761
Well, when Intel went from 32nm to 22nm they saw small IPC increases. Ivy Bridge is a tick, and it's the tocks (Sandy, Haswel) that bring the core improvements.
>>53268100
>Ivy Bridge is a tick, and it's the tocks (Sandy, Haswel) that bring the core improvements.
THAT is your explanation? You need to go take a look at that again. They do minor tweaks with die shrinks. You didn't honestly think that Intel just dumped their old architecture on a new process, did you?
>>53267761
I thought it was due to being able to fit more transistors on the die?
example:
>2500k 32nm 1.1b transistors
>3570k 22nm 1.4b transistors
>>53268216
>I thought it was due to being able to fit more transistors on the die?
Yes it is, which means that they needed to make some pretty big changes. The performance increase is because of the added transistors and the tweaks, not the smaller process.
A smaller process does not inherently make a processor have better IPC.
>>53258730
>Does the black man speak the truth?
Ask him if he stole some shoes
Any answer other than "yes" you know he's a liar
current AMD cpus are clocked so much higher than Intel's, overclock so much better and have higher core counts because of their simpler architecture
generally, higher IPC => increased architecture complexity => lesser overclockability and core count
AMD claimed 40% improved IPC, they said nothing about the clock speed. It might(and probably will) be lower than current cpus, meaning less than 40% performance gain.
smaller ipc lets amd put more cores on a die and a better architecture will result in a better performance per core. I hope they make a 16 core zen fx for desktops, it would destroy any desktop i7 in multi threaded benchmarks. imagine a 8350 with 2x more cores and better ipc...
>>53268316
If you're white and you stole a black man's shoes, what does that make you?
>>53268316
>mouthposting
>>53268154
You're right. Intel actually went backwards with Ivy Bridge. That's what makes the IPC increase from the die shrink alone even more impressive.
