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

[ShrimpBrime]

I miss AMD. More FX cores would be fun. 10 count maybe 12.... yea And I don't mean APU either.

[vario]

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?

Youre of course in large part right ,however i dont think its that simple.Intel has its tentacles in MUCH more stuff than AMD, so it costs more thats first, second it was similar before and AMD had done it either way(i mean win or come close) and there are a lot of examples of smaller companies winning in the long run and becoming big themselves.I think that the core of the problem began with Hector Ruiz and then with the board after him i mean after the A64 on 90nm there wasnt all that much that was done.Also, low R&D doesnt really explain why were standing still performance wise for so long.I mean lets look at it realistically , intel is standing still (almost) from the first bloomfields(in the mainstream) ,and yet perf difference is getting bigger and bigger.
Also, Cat cores, pretty darn good fighters given the miniscule budget .Its doable, but management has to be stable , concrete and has a good grasp on the technology, and i dont think thats the case.

[tbone8ty]

even just an 8 core FX part with updated/tweaked L3 cache and steamroller/excavator core would be nice. heck make one for fm2+ with a small igpu just for generic video. i dunno they need something until 2016

[vario]

even just an 8 core FX part with updated/tweaked L3 cache and steamroller/excavator core would be nice. heck make one for fm2+ with a small igpu just for generic video. i dunno they need something until 2016

Taking into account that they have a shrink available for some time, they could even do a 10-12 core no prob.But this deafening silence, makes me think that they just abandoned mainstream market as a company policy from now.I mean no roadmap shows anything more than puny apus.

[The Stilt]

Things that you would like to have are quite different to what can actually technically be done.

- Steamroller supports only up to 3 compute (6 cores) units.
- TDP for a six core Steamroller with same clocks as 7850K would be 125W MINIMUM.
- The die size would be massive, around 290 sq .mm (expensive & low yields).
- Incompatible with most of the current FM2+ motherboards (power infra), unofficial compability could be had on high-end boards (ASUS Pro, ROG & Gigabyte UP4 boards).

Only ARM (Seattle), Orochi, Kabini and Mullins based dies support up to 8 cores by the design.
Two Orochi dies can be installed on the same package (G34 / G2012) allowing the core count to increase to 16.
Komodo supported up to 10 cores per node by the design.

All of the current APU designs support only up to 4 (Trinity / Richland) or 6 (Kaveri) cores.
Multiple dies per package is not possible as there are no internal links (HT) available for the cross-node communication as there is on Orochi or Komodo.
Neither is multiple packages (MCM) for the same reason.

Excavator will probably increase the supported number of cores...
However without a much better manufacturing process I can't see them really turning into flesh.
A 8 core Excavator made on the current 28nm BULK would pretty much mean clocks similar to Mullins (up to 2.2GHz base).
Not because of the process is not capable for higher but simply because of the TDP.

Excavator is and has to be (unfortunately) compatible with the current FM2+ motherboards and therefore the infrastructure itself is limited to 95W max.
In fact AMD has already confirmed that Excavator will be up to 65W, which effectively means it has only up to 4 cores and they are operating at even lower clocks than the Steamroller cores in Kaveri does.

Unless of course AMD has a functional (= low risk) <20nm high-performance manufacturing process in their disposal to be used on Carrizo.
And that's not going to happen in the given schedule

[vario]

Things that you would like to have are quite different to what can actually technically be done.

- Steamroller supports only up to 3 compute (6 cores) units.
- TDP for a six core Steamroller with same clocks as 7850K would be 125W MINIMUM.
- The die size would be massive, around 290 sq .mm (expensive & low yields).
- Incompatible with most of the current FM2+ motherboards (power infra), unofficial compability could be had on high-end boards (ASUS Pro, ROG & Gigabyte UP4 boards).

Only ARM (Seattle), Orochi, Kabini and Mullins based dies support up to 8 cores by the design.
Two Orochi dies can be installed on the same package (G34 / G2012) allowing the core count to increase to 16.
Komodo supported up to 10 cores per node by the design.

All of the current APU designs support only up to 4 (Trinity / Richland) or 6 (Kaveri) cores.
Multiple dies per package is not possible as there are no internal links (HT) available for the cross-node communication as there is on Orochi or Komodo.
Neither is multiple packages (MCM) for the same reason.

Excavator will probably increase the supported number of cores...
However without a much better manufacturing process I can't see them really turning into flesh.
A 8 core Excavator made on the current 28nm BULK would pretty much mean clocks similar to Mullins (up to 2.2GHz base).
Not because of the process is not capable for higher but simply because of the TDP.

Excavator is and has to be (unfortunately) compatible with the current FM2+ motherboards and therefore the infrastructure itself is limited to 95W max.
In fact AMD has already confirmed that Excavator will be up to 65W, which effectively means it has only up to 4 cores and they are operating at even lower clocks than the Steamroller cores in Kaveri does.

Unless of course AMD has a functional (= low risk) <20nm high-performance manufacturing process in their disposal to be used on Carrizo.
And that's not going to happen in the given schedule

I understand you are talking about Fm2+ infrastructure, that doesnt explain lack of SR/EX cores for AM3+ with no GPU ,about die size ,well PH II x6 BD and Vishera were all larger than that, i guess it all boils down to FM2+ being build for low end chips (which is understandable) , however for larger chips AM3+ would be fine, if i remember correctly you even said that there were working komodo prototypes.As for TDP, well AMD introduced 200W chips on 32nm so...
Also isnt kaveri build on some other 28nm flavour than high end chips would be because of its GPU and low power target ?
Dont get me wrong, i fully understand that no mainstream/high end chip will come.I however think that its because of AMD policy, not that it isnt doable.

http://www.xbitlabs.com/news/cpu/dis...echnology.html

[The Stilt]

There isn't much point if any in releasing a brand new design for AM3+ socket.

The platform is completely outdated in terms of the power infra (SVI 1.0) and the CPU package itself.
Worse yet the chipsets are completely outdated: Two module structure (NB + SB vs. single FCH), No USB3.0, no PCI-E 3.0 and even the amount of available lanes is lacking.

Since the chipset operations are now at ASMedia I find the chance for a new chipset to be non-existent.
I'd say 990FX + SB9x0 & A88X FCH are the last external chipsets made by AMD.

[FlanK3r]

I knwo Asmedia doing some external chips, but they have some experience with chipsets too? Or AMD propose layout and Asmedia will produce it? I heard something about LGA in future desktop of AMD .

[The Stilt]

I knwo Asmedia doing some external chips, but they have some experience with chipsets too? Or AMD propose layout and Asmedia will produce it? I heard something about LGA in future desktop of AMD .

In future the chipset (FCH) will be integrated to the C/A-PU.
Excavator is the first non-Cat core to have a FCH inside it.

For desktop Carrizo this changes nothing as the external FCH will be used.
The mobile and tablet parts which are based on new packages the BOM will be cheaper because the external FCH is no longer needed.

[vario]

In future the chipset (FCH) will be integrated to the C/A-PU.
Excavator is the first non-Cat core to have a FCH inside it.

For desktop Carrizo this changes nothing as the external FCH will be used.
The mobile and tablet parts which are based on new packages the BOM will be cheaper because the external FCH is no longer needed.

So, we dont know nothing about the new cpus being launchd in 2016/2017, however as you say, package and infrastructure for them would tell us a lot about designed target for them...
but i guess even you doesnt have such information ;-)

[The Stilt]

but i guess even you doesnt have such information ;-)

You're spot on that.
Until the official release everything is pure speculation.

Let's hope for the best and fear the worst

[P.A.D>]

These were my thoughts exactly when I started reading this thread. Also ,more cores? For who? Not the end user market ,doesn't make sense.

"There isn't much point if any in releasing a brand new design for AM3+ socket.

The platform is completely outdated in terms of the power infra (SVI 1.0) and the CPU package itself.
Worse yet the chipsets are completely outdated: Two module structure (NB + SB vs. single FCH), No USB3.0, no PCI-E 3.0 and even the amount of available lanes is lacking.

Since the chipset operations are now at ASMedia I find the chance for a new chipset to be non-existent.
I'd say 990FX + SB9x0 & A88X FCH are the last external chipsets made by AMD. "

[FlanK3r]

AM3+ is definetely death for future. And AMD know it..

There is older news, but about LGA...I dont be surprise, if new highend AMD will be LGA

http://wccftech.com/amd-16nm-opteron...#ixzz3A47tgNYc

[tbone8ty]

Zen

[tbone8ty]

http://wccftech.com/amd-disclose-det...n-transistors/

Carrizooo