next big core AMD late 2015/or Q1-Q2 2016 - info, speculation

[FlanK3r]

So, AMD dont give up enthusiast CPUs. Only next two years we will be with Visheras or with some APUs (Carizzo) at FM2+.

There is news:
http://news.mydrivers.com/1/305/305092.htm
Bernd Lienhard: Last year we introduced a Piledriver (hammers) architecture, and achieved good market performance, there are about 30% growth, Kaveri also used Steamroller updates (excavator) architecture. FX Series of high-performance market positioning in the future will, within two years you will definitely see an update.

[vario]

So, AMD dont give up enthusiast CPUs. Only next two years we will be with Visheras or with some APUs (Carizzo) at FM2+.

There is news:
http://news.mydrivers.com/1/305/305092.htm
Bernd Lienhard: Last year we introduced a Piledriver (hammers) architecture, and achieved good market performance, there are about 30% growth, Kaveri also used Steamroller updates (excavator) architecture. FX Series of high-performance market positioning in the future will, within two years you will definitely see an update.

"within 2 years" puts it in half of 2016, so how youre getting late 2015 ?Realisticaklly speaking , its coming end of Q1 2016 at the earliest.

[AbortRetryFail?]

"within 2 years" puts it in half of 2016, so how youre getting late 2015 ?Realisticaklly speaking , its coming end of Q1 2016 at the earliest.

If you really want to split hairs, the FX branding will return "within 2 years" ... with the FX-7600P.

That said, I wouldn't mind a new stepping of Vishera at 32nm if AMD can make their 'typical' node improvements/enhancements. Carizzo should be sweet, too.

AMD can't seem to 'shrink' worth a crap (see their first attempts at 65-, 45- and 32nm) but they tend toward quite decent improvements on their second and third shots on a node.

[vario]

If you really want to split hairs, the FX branding will return "within 2 years" ... with the FX-7600P.

That said, I wouldn't mind a new stepping of Vishera at 32nm if AMD can make their 'typical' node improvements/enhancements. Carizzo should be sweet, too.

AMD can't seem to 'shrink' worth a crap (see their first attempts at 65-, 45- and 32nm) but they tend toward quite decent improvements on their second and third shots on a node.

Dunno, 90nm and 45nm seemed to be good from the get go, then after a while even better while 65nm was crap until the end as for 32nm its hard to say because they were doing a) Bulldozer which has its own problems and budget oriented cpus which never scaled much (look at athlon II vs phenom II, athlons with smaller dies should be getting more headroom yet they were crap in phenom II comparison).28nm seems to be bad tho, or maybe its that pesky BD architecture.Anyhow Flanker is saying about the new arch and he is being overly optimistic with the time frame.
I would buy SR FX or even a good revision update to the line too .But AMD stated something like a half year ago that they dont want any more money from AM3+ owners and AMD enthusiast market in general.AMD let itself to have a big F gaping hole in the lineup for 4 years.Thats just bad in every way.

[FlanK3r]

Hard to say mention as SR FX...

1) Vishera FX is good CPU for almost 2 years old CPU...OC ability up to 5 GHz
2)in theory FX Steamroller 28nm could be similar in TDP, but with lower clocks potential - with some luck up to 4.5 GHz. With some improvements is Steamroller B very awesome against Piledriver, but in other cases are only few procent or the same as Piledriver or a bit lower (of course, clock to clock is SR better tahn Piledriver)
So 4.5 GHz FX SR could be average at similar performance as 5 GHz Vishera FX. In wprime 1024 40% better, but in winrar 20% worse etc...

28nm SMP is not good now for big high clocks cores. Maybe Excavator core in future or soemthing new.

[vario]

Hard to say mention as SR FX...

1) Vishera FX is good CPU for almost 2 years old CPU...OC ability up to 5 GHz
2)in theory FX Steamroller 28nm could be similar in TDP, but with lower clocks potential - with some luck up to 4.5 GHz. With some improvements is Steamroller B very awesome against Piledriver, but in other cases are only few procent or the same as Piledriver or a bit lower (of course, clock to clock is SR better tahn Piledriver)
So 4.5 GHz FX SR could be average at similar performance as 5 GHz Vishera FX. In wprime 1024 40% better, but in winrar 20% worse etc...

28nm SMP is not good now for big high clocks cores. Maybe Excavator core in future or soemthing new.

So... youre making excuses for amd for NOT making SE AND EX FX chips ?
2) In THEORY we dont know sh1t as kaveri is an APU and we cant make any real world assumptions how cpu only chip would do, we cant make direct comparisons as every other APU was clocking WORSE than pure cpu counterpart.llano trinity richland were clocking worse than pure cpu chips, its probably process/design thing and the complexity of adding gpu.
SR IS faster IPC wise, thats a fact, lower node could add more cores/more speed or more L3 cache.
Also 5ghz visheras are a pipe dream, few chips are stable at this settings, and i havent seen one on air (no, a single run of cinebench isnt stable, and yes 8h run of video encoding/prime or occt is).
My point is, given the past, we cant read in too much to its cpu only potential.So 28nm is a "?" at the best.

[haylui]

less EN IT websites reporting about AMD APU 14 Beijing. I wonder why is not covered by western media...If you could get a lot of coverage from chinese websites....

[FlanK3r]

Lok at this, clock to clock is Steamroller B very good. Im sure, in some situation beat clock to clock FX "piledriver core with L3"

http://pctuning.tyden.cz/hardware/pr...ovani?start=10

[[XC] Lead Head]

Yeah, and just barely edging out Vishera is still garbage. Steamroller B is not good.

[vario]

Yeah, and just barely edging out Vishera is still garbage. Steamroller B is not good.

Thing is, Vishera has big ass L3 cache and kaveri has zilch.So thats not really a fair assessment.IF SR B on am3+ HAD L3 cache (maybe even more due to shrink) it would fair a lot better.
Thats just the thing with SR B in kaveri, its a low end cpu with a mid end GFX, its not meant really for desktop.Any comparisons on how high it can clock or architecture vs architecture arent really in place, becasue AM3+ was meant as a mainstream desktop.SR B MEANT for Am3+ would probably clock as high (or higher) and retain the IPC advantage at lower power and smaller footprint.But its a moot point, amd stays in the low end for the time being.

[FlanK3r]

SR B is Wprime and Superpi "monster". This CPU I beleive is quicker than Haswell in wprime clock to clock with 2C/4T and in Superpi 32 is very effective for AMD (with air you can get around 13min only with 4.7 GHz, before without L3 cache impossible for FX impossible too with same cooling at 5200 MHz)
Looking again at this graphs 4 GHz vs 4 GHz:

[NEOAethyr]

Yeah but 13mins though, that's half the speed in single threaded apps.
Oh wait a min, 16m, who the heck runs 16m lol?
I always figured it was either 1m for scores, and 32m for stability checks (even though it's not really worth the time anymore).

Anyways looking forward to 2017'ish 2018 whatever.
I probably won't be building a new rig for a long time anyways.

I've noticed some ht issues in linux on the intel chips.
Nothing major though, just a bit of stutter that can be gotten rid of if you taskset which cpu's you wanna run on for specific apps like wine.
I was doing some reading on it and they say that intel's share the fpu between each 2 core modules.
Though from what linux is telling me it's saying that all 12 cores have (access to) fpu's.

Have you guys ran any windows benches in linux yet?
I noticed aida64's read bandwith is a bit higher then in actual windows.
Linux has some scheduling issues or something though, but since this last night I've been noticing more and more just how of date windows seems :\.

If I remember right superpi was originally written in linux.
I'll have to look that up.

[Darakian]

I'm looking forward to their arm chips above their x86. I like the mention that it has a "bigger engine"

[sauria]

16 cores on the next FX?

[Johan45]

I was under the impression that AMD next Enthusiast CPU was supposed to be a fully redesigned from the ground up offering and not based on the current BD/PD/SR/EX architecture at all. http://wccftech.com/amd-confirms-dev...#ixzz30wPt64KL

[robbicfc]

Good to hear. If FX series are in line for an update then perhaps the platform as a whole might get some love. PCIe V4 and more support for M.2 please

[FlanK3r]

up (update)

[FlanK3r]

Finally next information (not confirmed):
http://wccftech.com/amd-16nm-opteron...res-2016-2017/

The report starts by saying the FX and Opteron x86 processors are returning in 2016 or 2017 and are preparing to make quite an entrance when they do so. The processors in question will apparently be built on the 16nm or the 14nm fabrication process. The node will go to Samsung/GloFo/ TSMC depending on some variables. The architecture itself will be highly customizable and though they fail to mention so, this is probably the time when AMD introduces a non-Bulldozer architecture. This is also the same time AMD is supposed to shift from CMT to SMT. They finally go on to state that the processor will land on the LGA Socket (No, that does not mean AMD processors will be compatible on Intel Chipset Mobos). Patent No. US 20140043768 A1 also lends credence to the these claims if you ignore the fact that the report does make sense.
Also, the source specifically mentions that when they say upto 20 cores, they do not mean compute cores. The only thing that bugs me about this is that it fits and makes too much sense on nearly all aspects. This is why there might be a danger of it being a case of “to good to be true”. Now I must admit something, If all of this comes true, AMD will be making one helluva comeback in 2016. Intel has been running nonstop pretty much without any competition and some would argue that this has given them the ability to exploit the market. If you have a competitor on your heels, you can’t afford to take your merry time around a node. Its worth pointing out that this report also fits in with the schedule of TSMC. 20nm is supposed to arrive in 2015 so we should get high performance 16nm/14nm TSMC by 2017 at the latest.


Read more: http://wccftech.com/amd-16nm-opteron...#ixzz3A4a8Nk93

[haylui]

I am wondering if AMD keep following Intel footsteps, will it survive until now....when AMD has better FPU in k6-2 era, people crying for INT performance. During K8 AMD wins and K10 was a slip and need K10.5 badly.
Bulldozer was a monster in INT but....lol...people crying for FP performance now...

[FlanK3r]

I think, it must be balanced...Example K10.5 were very popular, it was sunshine after average years before. Bulldozer killed it, Vishera fixed reputation. But now is Vishera "too old" from launch date.
So the best way is creating some CPU with very good single thread performance what AMD may do and with classic multithread (this AMD can do). Question is, how good can be next new CPU core with lower budget than Intel?

[The Stilt]

Everyone needs to adjust their expectations for the new uArch, eventhou JK himself (along with several other talented persons) is working on it.
AMD's resources are extremely limited and it most definitely will have at least some sort of a effect on the outcome.

How limited?
So limited that AMDs entire R&D spending for the whole year would last less than 1.5 months for Intel.
And that includes everything, not just the CPUs (CPU, GPU, chipsets, drivers, APIs, software, custom solutions, etc).

So Intel is spending roughly 9 times more in R&D than AMD.
Still wonder why the results are more impressive?

[FlanK3r]

Yeah, I talked about it (maybe I got this words in bad conjunction with "?"). I believe, the smart team can surprise a bit . The GPU of AMD are awesome, the CPU are not bad. But we know, AMD surprised with K7/K8 and later with Phenoms II a bit and later with Thubans.

[ShrimpBrime]

The main problem AMD had while producing BD chips was the massive cut in transistors. Typically a shrink would produce room for more transistors. The fail can be seen while benching my A8-3850 VS FX-4300. With many benchmarks single and multi threaded, the A8 generally comes out with better numbers even while running a much lower clock speed. L3cache means nothing when you dont have enough compute transistors to do the reserved cache any justice.

Hopefully, AMD can or will change and move away from "module" or "cluster" cpu design.

[ilkkahy]

For amd, Samsung & GF:s 14nm manufacturing process cant come out soon enough. In the long run its propably not so bad that they went 28nm bulk but naturally it was kind of dissapointment to find out that Kaveri overclocks so badly compared to older 32nm parts.

[FlanK3r]

but these two are different architectures. You know, BD is for much higher clocks. And the issue is 1FP per 2 cores. But I understand what you mean. Interesting is watching the changes - example FX-4300 is better in games than PII x4. Tomshardware tested it few months ago.