AMD’s New High Performance Processor Cores Coming Sometime in 2015 – Giving Up on Modular Architectu

[Martini1991]

All they need for many people is not performance parity, but if they can close to what the Phenom II X6 could do to Intels offerings at the time (And especially on a price/performance route) they become fully viable.

AMD currently only have the price/performance in their favour, but the disparity between what an i5 4670K can do and what an FX83 can do is massively different due to the IPC gap, as compared to what the Thuban could do to the i5's at the time (The Thuban could truly beat the i5 in tasks ; Encoding, to a much bigger degree than the FX83 slightly out muscles the i5 4670k, and when it was behind, it wasn't *too* behind)

 

[eddyr]

Maybe not, my post was mostly speculation I admit. However, Intel are a very wealthy company and I'd like to think that they are already working extensively on the CPUs "of tomorrow".

Mine also. I expect a lot of early proof of concept and more late stage experimental prototype dev going on in parallel. Just anything planned for the 3-4 years time frame is still nowhere near finished but has relatively defined design goals and won't be too acceleratable.

Not the same but USB3 I feel is a sort of an example of their approach, they will simply not be rushed, even when the industry are temporarily ahead in one aspect.

 

[Phixsator]

If AMD can just match the mainstream i7s performance when it releases this enthusiast grade cpu(if they do) while giving people access to all the features that comes with an Intel X platform (like x79 and x99), that could be full support for all sort of Virtualization, tripple or quad channel DDR4 ram, Thunderbolt for workstation use(doubt it unless it comes from a third party vendor) and im sure the list continues, at a lower price than the Intel X platform then i see them having a winner. But ofcourse i would'nt mind if it was even faster, but one has to be reasonable

 

[Boomstick777]

AMD CEO: Beyond the ‘Unhealthy Duopoly’ of PC Chips

WSJ: Let’s talk a little about server, where your share of the market–

Read: –Has gotten much smaller.

WSJ: Has gotten really small. Are you giving up on x86 server chips?

Read: No. So let’s be perfectly clear on that. I think that Bulldozer (AMD’s recent processor design) wasn’t the game-changer that people had thought it would be.

We went out and got world-class designers and world-class leaders like Jim Keller or Raja Koduri that came over from Apple.

They’re working on next generation work in that space, in the server space, in the cores space. And that’s going to be both ARM- and x86-based.

Them are fighting words, looking like the new architecture AMD put out to replace the iterations of Bulldozer is going to be very good. With traditional IPC performance back in the game. Be great if things get competitive again.

 

[ubersonic]

About damn time

 

[Martini1991]

Potentials!

 

[drunkenmaster]

It's not like Jim Keller makes it a certainty that the next chip will be epic but Keller is one of the top couple of guys in the entire world in his field. With his leadership and AMD's tiny budget he helped make K8 a chip that led the market for quite a while. Intel had their own good people and came back and Keller left(I can't remember the reasons why and google is a bit sparse on older details, I think he just left for a more senior role as so many do at such companies. Lead engineer leaves to found a new company or take a VP/ceo type role).

You could using fingers on one hand the guys you'd bring in to make a brilliant new architecture and Keller could well be the first person on that list.

It's still unclear where AMD are going. it sounds a little(but not certainly at all) like Keller is pushing for their highest performance core to be based around the ARM ISA. He helped come up with x86-64 but while good it's still fundamentally based on a very old x86 ISA and he seems to believe the ARM ISA is newer and better.

The difficulty there in the future is(for us as mostly gamers) if their ARM chip say beat an i7 and anything AMD make in x86, how will games support it and what OS would it run.

Would MS ramp up windows to support both ISA's, they've tried an ARM windows, it kind of sucks but in 2 years could they have it working like normal windows, supporting DX. Or would AMD want to get ARM based support via android in lower end devices and pick a dedicated for support Linux distro, maybe steamOS for gaming. Get Mantle support to linux in the next two years, but if most AAA titles get linux/steam/openGL or Mantle support for the next two years then switching to a steamos/linux + arm chip set up may be a not bad transition.

This is likely primarily why they are making an x86 core at the same time, because lots of business buyers won't switch OS that quickly and I think the switch in the future for gaming to potentially ARM could be difficult.

Next gen consoles could be an interesting push for the desktop industry though. If the new consoles go ARM + GCN(or whatever AMD's next architecture is) then the natural desktop compatible port would be AMD ARM + gcn chips + whatever OS supports those chips the best.

Big move in the industry committing to making a very high IPC super fast ARM chip, no one else is moving that way(or has suggested moving that way) any time soon.

The other big issue is, if AMD switched desktop gaming from Windows to ARM based steamOS/linux, then they would effectively switch from having Intel as the only x86 competitor, to potentially having anyone making arm chips willing to make a higher power version, or relatively simple through a few more cores and more memory bandwidth and make a good gaming chip.

Keller may believe ARM ISA has a much better future than X86, but they'd have only one competitor if they kept gaming on x86 for the foreseeable future, if they move down the ARM path sooner then they could get into a fight with a heck of a lot more companies. On the other hand, they could be the first and best to make really high performance ARM cores and dominant that area before anyone else gets started in it.

 

[2013]

I have it on good authority. They aren't done with Enthusiast level CPU's yet

Thats some good news, the way amd was pushing apus I thought they had forgotten about CPUs

 

[2013]

And there was much rejoicing Right now, good single thread support still matters for PCs, and it's nice that AMD have realised it (finally).

Just hope this won't become Bulldozer v2. Full of hype and fanfare and excitement about the new powerful 8 core CPU architecture that would tear up the benchmarks! I remember waiting ages for the first reviews, on the edge of my seat with anticipation... then immediately buying Intel when the NDA expired and the facts came out.

Lol, I did the same.

 

[2013]

+1, Having said that AMD can do better than Vishera even today, dropping the modular design and another year or so development they can do even better still.

Personally i very much doubt they can pull off another Athlon XP with Intel also not standing still, but they don't have to beat Intel, the performance just needs to be good, much better than it is today, and then compete on price.

I agree with both of you here, intel wont wait around, but i dont think they will push the boat out till after AMD's next gent cpu, even then there will likely be a big difference in price ?

with amd providing the apus for the consoles, they will have a bigger budget this time round and a better idea.

 

[2013]

Intel X99 is the upgrade that makes the most sense if you're building this year. On a more than 4 core chipset, DDR4 support, plus future compatibility with Broadwell -E. Even when the new AMD stuff is out in 2016 X99 could very likely still be competitive. Prob get at least 3-4 years at the highest end with X99. Def my next upgrade. Then move back to AMD in the future when things are better on the CPU front.

I think im gong to give DDR4 a year or so before i use it, it mabe ready by the time AMDs next cpu is ready.

 

[2013]

Interesting, but AMD appears to have lost a few years of R&D on a platform they are now discontinuing for mainstream performance CPUs. That's going to hurt.

Im hoping the will use some of the R&D from the APU.

 

[2013]

I'm not sure, I suspect that the case with Intel is that they have a very structured plan all the time for their next few years ahead (they seem like that kind of company). So, with that assumption, it wouldn't be unreasonable to think that they already have prototypes of much more powerful CPUs waiting behind the scenes to be released in X years time while they continue to make sales on their current processors within that X year time gap. Therefore, I'd imagine that if AMD released something which threatened Intel's performance in the CPU space (which I doubt anyway) then Intel would immediately act to bring about an earlier release of the more powerful processors they have prepared.

You maybe right here.

 

[opethdisciple]

If they are dropping the modular design, does that mean they are also not going to be able to release 8 'core' cpu's in the future?

 

[~Dockie]

If they are dropping the modular design, does that mean they are also not going to be able to release 8 'core' cpu's in the future?

I wouldn't say that, they had hex-cores out on 45nm so it's still possible.

 

[2013]

Could it be possible they may reduce the cores like intel.

 

[Yamahahahahaha]

If they are dropping the modular design, does that mean they are also not going to be able to release 8 'core' cpu's in the future?

Short answer, no, they can still release multicore chips easily.

Long answer, AMD's tradeoffs when implementing 'CMT' directly impacted on their single-threaded performance. So rather than getting extra threads of equal computing performance (Which Intel do with HT, the bottlenecks lie elsewhere), the CMT modules lowered performance-per-thread but increased the number of threads. This might be a good tradeoff if your workload specifically benefits from massively-parallelised processing.

Three things happened (or rather, one did, two didn't). Firstly, massively-parallelised processing hasn't taken off in general. The challenges facing multithreading haven't been conquered, and the industry is still full of programmers who haven't been exposed to designing software for multithreading. The rise of GPU compute has also eaten some of their lunch. Secondly, single-threaded performance became more and more important, extending Intel's lead in the space. AMD's desktop chips cannot compete. Thirdly, AMD's software-designed CPU's approach (Parts of Intel chips are still designed 'by hand' to maximise transistor density and speed) took way too long to mature and meant their desktop architecture fell way behind Intel's. The benefits of the approach did not come together in this generation.

What can we expect from AMD? Take a look at the CPUs in their 'Fusion' chips. They're pretty competitive with Intel's lower-end offerings already. By stripping out CMTs and replacing it with Hyperthreading (or something very close) would mean AMD are right back in the mid-range trading blows with Intel's i3 and i5 chips, this is where the sales volume is which will help AMD grow revenues. The lack of 'premium products' remains, but with Xeon chips eating into the Operaton market, AMD need to fight hard to reclaim the space.

 

[humbug]

Short answer, no, they can still release multicore chips easily.

Long answer, AMD's tradeoffs when implementing 'CMT' directly impacted on their single-threaded performance. So rather than getting extra threads of equal computing performance (Which Intel do with HT, the bottlenecks lie elsewhere), the CMT modules lowered performance-per-thread but increased the number of threads. This might be a good tradeoff if your workload specifically benefits from massively-parallelised processing.

Three things happened (or rather, one did, two didn't). Firstly, massively-parallelised processing hasn't taken off in general. The challenges facing multithreading haven't been conquered, and the industry is still full of programmers who haven't been exposed to designing software for multithreading. The rise of GPU compute has also eaten some of their lunch. Secondly, single-threaded performance became more and more important, extending Intel's lead in the space. AMD's desktop chips cannot compete. Thirdly, AMD's software-designed CPU's approach (Parts of Intel chips are still designed 'by hand' to maximise transistor density and speed) took way too long to mature and meant their desktop architecture fell way behind Intel's. The benefits of the approach did not come together in this generation.

What can we expect from AMD? Take a look at the CPUs in their 'Fusion' chips. They're pretty competitive with Intel's lower-end offerings already. By stripping out CMTs and replacing it with Hyperthreading (or something very close) would mean AMD are right back in the mid-range trading blows with Intel's i3 and i5 chips, this is where the sales volume is which will help AMD grow revenues. The lack of 'premium products' remains, but with Xeon chips eating into the Operaton market, AMD need to fight hard to reclaim the space.

Just one thing, AMD aren't fighting Intel directly in the Server space, not anymore, they can't....

What they doing instead is reinventing parts of it and getting the technology jump on Intel.

AMD are the parent Company of SeaMicro, well, they are one and the same now, they have been pioneering high density low power servers for some years now using Jaguar and ARM SoC's, they are very good at massive parallel work, they are gaining traction and significant contracts in things like Cloud Computing.

 

[Yamahahahahaha]

Just one thing, AMD aren't fighting Intel directly in the Server space, not anymore, they can't....

Intel's performance is pretty rampant at the moment and AMD aren't competitive in the new install space. It would take something pretty special to take the Xeon market away from Intel, I seriously doubt they can pull that off.

What they doing instead is reinventing parts of it and getting the technology jump on Intel.

Yep! It's the only way they can compete. Intel's process advantage is pretty alien to every other company so even then, they have to keep the wraps on it until it's ready.

AMD are the parent Company of SeaMicro, well, they are one and the same now, they have been pioneering high density low power servers for some years now using Jaguar and ARM SoC's, they are very good at massive parallel work, they are gaining traction and significant contracts in things like Cloud Computing.

The workloads for that kind of kit is pretty specific, mostly low-utuilisation webserving. Intel's resurgent Atom offerings are competitive that space. Most cloud servers are chunky Xeon servers with tons of RAM behind them.

 

[humbug]

Yes, it is pretty specific at the moment.

As you said, Multi-threading, even 64Bit Computing to an extent never materialised, it never will, Bulldozer was conceived under the impression that we would be very multi-threaded at this point. AMD got that very wrong, partly because AMD did nothing to push their own technology to the fore.

GPGPU computing both locally and remotely i dare say is the future, having learnt a hard lesson they are pushing that, HSA is starting to appear in productivity software, entertainment software and even web browsers.

GPGPU computing is something Intel can do, but, to an open standard (which is crucial) AMD do it extremely well, they are years ahead of Intel, for all of Intel's Money they are still not going to find it easy to catch up let alone push AMD of their perch.

Its not like Intel are fighting AMD on their own x86 turf, this is Intel fighting AMD on their Heterogeneous Architectures, a daunting task even for Intel.