AMD Reveals High-Performance Zen CPU, GPUs With HBM, And FinFET

[elbert]

I wonder if AMD will use the HBM memory in some way to reduce latency? Would be a shame if they cant utilize all the throughput for there CPU's. Maybe a 2~4GB HBM APU. Make the APU's GPU part finally be used as a high end math co-processor.

 

[gamebrigada]

Does NOBODY else see SMT on that? Everyone is fighting over the 40% figure but NOBODY notices that Hyperthreading is FINALLY COMING TO AMD.

 

[jimmysmitty]

Does NOBODY else see SMT on that? Everyone is fighting over the 40% figure but NOBODY notices that Hyperthreading is FINALLY COMING TO AMD.

Because it was inevitable. While CMT is a more advanced version of SMT, it doesn't do a good enough job. It takes up more space than SMT, quite a bit, and due to that its performance return is just not worth it.

I had a feeling that AMD was going to eventually drop the module design and move back to a normal CPU design and throw in SMT as SMT is an easy way to get about a 20% boost with very little in terms of extra transistors or power use.

 

[ITechFreek]

Well, intel already has a 15 core 2.8 Ghz CPU, (Xeon E7-8890 v2) They probably planned ahead of time for this.

 

[firefoxx04]

Fool me once, shame on you AMD. Fool me twice, shame on me. Seriously, a 40% increase will not be enough to combat 2016 intel.. and even if it is, when has a large claim like this actually panned out to be real when coming from AMD?

I love AMD, I really do. I use tons of their chips.. but I just dont buy these massive performance increases that never happen.

 

[anthony8989]

If they could pull out an APU with the efficiency / processing power of the 8370E and the graphics power of say a 7790/7850 I'd buy into that platform for under $200 USD for the APU. Sadly I don't think AmD could afford to charge so low for such a product. Kaveri was a little over priced for its market segment in my opinion. Great chip - just a bit too expensive.

In addition, I can't wait until FinFET is fully adopted and old news , because for some reason the word FinFET annoys me. I'm sick of seeing it.... I'm weird. FinFET.

 

[atminside]

Will this provide a noticeable improvement over a 955X4? I know it's old but the current FX lineup doesn't really provide a performance gain worth me getting a new mobo and CPU.

 

[kinggremlin]

From the results I have seen, at the same clock speed on SINGLE threaded benchmarks, haswell is ~60% better than the 8xxx series. If this 40% holds up, AMD will seriously close that gap, and create some SERIOUS competition with multithreaded tasks which is currently their strong point. I can't wait to see how their lineup unfolds.

Barring something highly unexpected, skylake will be out this year, so when these Zen chips are finally released at some point in 2016, they won't be competing with Haswell. They could end up only closing the gap by 50% or so against skylake, which is nothing for the enthusiast to get too excited about. AMD would again have to win marketshare based on cost. Anyone focused solely on performance, will still have only Intel as an option.

 

[Jaroslav Jandek]

40% IPC improvement is huge, although Intel CPUs will still have the advantage, since they have approx. 60% (prob. 65% in 2016) higher IPC - especially in synth benchmarks (e.g. Cinebench, 75%). Still, it will make AMD vastly more competitive, especially if they can keep their lower prices.

it should consume more than 50 percent less power

Great phrasing . Did you mean "it should reduce its power consumption by at least 50%"?

 

[Creme]

This is very exciting news, but why keep the FX name? It's only associated with power hungry inefficient processors, might as well introduce a new nomenclature.

AM4? I thought they were going to be only FM3.

 

[daglesj]

My prediction -

Zen will come up 20% short.

Carrizo will be largely ignored by the OEMs like all the previous incarnations.

Retail really doesn't need or want AMD anymore.

 

[lekzero]

40%? Or 400% like in graphics? xD

 

[wtfxxxgp]

This is very good news. I don't doubt that Intel has been working on something to shock us all with in the meantime too though. Intel has bumped up performance very slightly between new chips simply because they can - without a serious competitor, they are allowed to milk every little iteration of what they could actually produce - I believe Intel will be waiting patiently for the benchmark results and then shock everyone again by providing something that puts AMD right back in non-contention land for the next few years again (on the high end only though). This is very good news to a consumer: we'll benefit out of this no matter what - the rich will get to spend lots of money on the latest and greatest and probably get to upgrade quite often, but I think the real winners here are the lower-to-middle income earners - we'll get many benefits in terms of pricing and speed on modules we can afford. I say "Go AMD! SHOW UP!"

 

[GreaseMonkey_62]

Sweet. Finally looking forward to something good from AMD.

 

[Haravikk]

AMD's technology in the APU space is very exciting, but their hardware disadvantage on CPUs is making it really hard for developers to take advantage of HSA and other great features. If AMD can actually pull out a significantly better CPU architecture, then we might actually see these things surfacing at last; they wouldn't even need to overtake Intel, just get close enough to gain popularity, as HSA and other features could then give them the performance edge.

 

[ubercake]

Fool me once, shame on you AMD. Fool me twice, shame on me. Seriously, a 40% increase will not be enough to combat 2016 intel.. and even if it is, when has a large claim like this actually panned out to be real when coming from AMD?

I love AMD, I really do. I use tons of their chips.. but I just dont buy these massive performance increases that never happen.

I agree. It really is kind of hard to believe their hype any more. I want it to be real, but based on AMD's past lack of delivery on their lip service, I have a lot of doubts.

 

[Steveymoo]

Let's all hope they pull something competitive out of the bag. Although I'm not going to hold my breath.

 

[mosu]

I wonder how stacked memory will work in high power configurations from the heat dissipation point of view.

 

[leo2kp]

Please be true. Please be true. I miss the glory days of competition!

 

[LookItsRain]

I hope AMD charts and word holds up to thier release product, we need competition in the market.

 

[Jake Hall]

x86 is sooooo two-thousand and late...

 

[alidan]

This is very good news. I don't doubt that Intel has been working on something to shock us all with in the meantime too though. Intel has bumped up performance very slightly between new chips simply because they can - without a serious competitor, they are allowed to milk every little iteration of what they could actually produce - I believe Intel will be waiting patiently for the benchmark results and then shock everyone again by providing something that puts AMD right back in non-contention land for the next few years again (on the high end only though). This is very good news to a consumer: we'll benefit out of this no matter what - the rich will get to spend lots of money on the latest and greatest and probably get to upgrade quite often, but I think the real winners here are the lower-to-middle income earners - we'll get many benefits in terms of pricing and speed on modules we can afford. I say "Go AMD! SHOW UP!"

this is likely amd's last shot. so yea... i hope they come within 5% of intel and undercut them in cost, and if intel comes up with something new, amd can match or surpass it.

the last thing we need is a monopoly so i REALLY hole intel has no answer.

 

[f-14]

something is misleading with this chart, excavator on the chart is supposed to double bulldozer, and then there's this 40% increase over excavator.
amd flagship is 60% behind intels flagship, to double from bulldozer that would be a 30% increase and make amd -30% of an intel, but the excavator core is only 200% of the bulldozer core 100% that should only be a 20% increase vs intel and also a 60% gain for Zen which would put Zen on par with intel or better depending on how you view the chart.

there's nothing developmental being cited as the reason for the increase either, just like the bulldozer press releases at the same time of it's announcement and year prior press releases.
seems like excavator is an intel tock and bulldozer to Zen is just a die shrink performance gain with some minor new chip stacking tech added in.

 

[xaephod]

When I see it I'll believe it.

 

[photonboy]

"Thus, its performance per watt should be 3x that of GDDR5, and on top of that, it should consume more than 50 percent less power."

Um... What???

So the new memory is 300% more efficient but consumes 50% less power? Wouldn't it then be at least "200%" more efficient?

Also, where did the "thus" part come from? There's no obvious thought path that leads to the new memory being 3X more efficient just because it's stacked. That may be true but you can't say "is 3D thus 3X...".