AMD prepares Zen for CES 2017 launch

Pieter3dnow

Supreme [H]ardness
Joined
Jul 29, 2009
Messages
6,784
According to Digitimes :

The high-end X370 chipsets will be launched at CES 2017, while mainstream B350 and low-end A320 will be unveiled later. X370-based motherboards will begin shipments in December.

AMD's Zen-based CPUs and its corresponding high-end X370 chipsets will make their arrivals at CES 2017 in January. To achieve a smooth transition, AMD has begun adjusting product prices in order to accelerate inventory digestion. However, since Zen-based products are unlikely to start mass shipments until after February, they are unlikely to start contributing profits until the second quarter 2017, according to sources from the upstream supply chain.

Seems we are getting there finally and if Zen goes on sale in February it should finally be over ;)
 
I don't trust anything AMD claims or leaks anymore. Still waiting for a decent video card from them to upgrade mine.
 
I wish it was sooner. The USB3 ports on my X79 Deluxe just took a dump and I need an upgrade. I guess I can limp along until Feburary with a USB2 hub if nothing else dies. I really wish I'd just stuck with my FX 8320 rig, I gave it to my girlfriend and everything still works great. I thought I'd see a performance difference with the Intel 4930k but it's mostly been a waste of money.

I wont be buying another ASUS motherboard, that's for sure. This X79 Deluxe was $350 and now both the wifi and USB3 have died... The Gigabyte 990FX UD3 in the AMD machine was ~$120 when I got it. Overclocked to 4.8 GHz all these years, no problem. :rolleyes:
 
Just buy a PCI express usb 3.0 card , would function as a backup as soon as you buy a new computer ;)
 
Well, I might just upgrade both my computers at that time. However, even then, I am not certain of that since I am have no issues at present with my existing FX 8 core computers. Looking forward to whatever AMD releases, regardless.
 
Well, I might just upgrade both my computers at that time. However, even then, I am not certain of that since I am have no issues at present with my existing FX 8 core computers. Looking forward to whatever AMD releases, regardless.
I will most likely wait for their top end and see if it is necessary. Not seeing any reason now for an upgrade of any sort.
 
great at tax time I get to decide If I am going to bother upgrading from my FX-8320@ 5ghz, or just keep on waiting for skylake-e.

in the meantime I'm upgrading my home brew nas, with sfp link to my desktop, and 5-4tb drives. Yay for 200+mb/s transfer speeds via network storage.
 
I wish it was sooner. The USB3 ports on my X79 Deluxe just took a dump and I need an upgrade. I guess I can limp along until Feburary with a USB2 hub if nothing else dies. I really wish I'd just stuck with my FX 8320 rig, I gave it to my girlfriend and everything still works great. I thought I'd see a performance difference with the Intel 4930k but it's mostly been a waste of money.

I wont be buying another ASUS motherboard, that's for sure. This X79 Deluxe was $350 and now both the wifi and USB3 have died... The Gigabyte 990FX UD3 in the AMD machine was ~$120 when I got it. Overclocked to 4.8 GHz all these years, no problem. :rolleyes:

My RIVE the bluetooth module died a fw years ago. I will probably never buy an ASUS board again either. I am going to get Zen if it is as fast or faster than my 3930K but if not then I am not worried about it.
 
Sweet Stuff. I wonder if they'll have anything worth upgrading to.
 
Zen launch lines up nicely with when I will have the funds together for my next upgrade, I want to see the mid end Opteron stuff however as I would like more then 8 cores and ECC memory support. If AMD has good info on what their Opterons will look like at the point I will wait for that otherwise, I am looking at a e5-2620 v4 dual socket machine as my next box.
 
Zen launch lines up nicely with when I will have the funds together for my next upgrade, I want to see the mid end Opteron stuff however as I would like more then 8 cores and ECC memory support. If AMD has good info on what their Opterons will look like at the point I will wait for that otherwise, I am looking at a e5-2620 v4 dual socket machine as my next box.
I was thinking the same thing, in the past you could get some great Opteron's and motherboards way cheaper than Intel.
 
  • Like
Reactions: N4CR
like this
My RIVE the bluetooth module died a fw years ago. I will probably never buy an ASUS board again either. I am going to get Zen if it is as fast or faster than my 3930K but if not then I am not worried about it.

If you're basing your future with Asus based on a bluetooth module is a bit much. Quite possibly the best manufacturer available and you're going to terminate your relationship with them over a module that probably wasn't even built by them. Did you call and they their CSRs told you to go eff yourself? Seems like an extreme response.
 
If you're basing your future with Asus based on a bluetooth module is a bit much. Quite possibly the best manufacturer available and you're going to terminate your relationship with them over a module that probably wasn't even built by them. Did you call and they their CSRs told you to go eff yourself? Seems like an extreme response.

Nah I was just being sarcastic and having fun, but at X AM in the morning I failed to articulate myself properly haha..

Asus is just fine in my book!
 
great at tax time I get to decide If I am going to bother upgrading from my FX-8320@ 5ghz, or just keep on waiting for skylake-e.

in the meantime I'm upgrading my home brew nas, with sfp link to my desktop, and 5-4tb drives. Yay for 200+mb/s transfer speeds via network storage.

Seriously, I assume you mean 200 MegaBytes, right? Because 200 mbits is not good at all. ;)
 
My friend and I were discussing Zen APUs, as these will probably be the larger market. is there any news/leaks/theories on what to expect from the AM4 Zen APUs? Are we going to be stuck with yet more 4core 8CU chips... or will AMD start taking it's integrated graphics seriously again?
 
Raven Ridge will be 4c/8t with 12CU igpu(for 768steaming core gpu). And will incorporate some sort of on package HBM2 in some fashion as of yet unrevealed. I would not be shocked to see higher CU count versions as time goes on however. or maybe even more cpu cores. HBM is low latency(as compared to DDR4) and would make a great l4 cache!, so maybe it will work like iris pro, tho might just be gpu memory. AMD being surprisingly tight lipped on this one.
 
Raven Ridge will be 4c/8t with 12CU igpu(for 768steaming core gpu). And will incorporate some sort of on package HBM2 in some fashion as of yet unrevealed. I would not be shocked to see higher CU count versions as time goes on however. or maybe even more cpu cores. HBM is low latency(as compared to DDR4) and would make a great l4 cache!, so maybe it will work like iris pro, tho might just be gpu memory. AMD being surprisingly tight lipped on this one.

It wont have HBM2. 12 CU wouldn't do much either besides die size waste. And if we look on the Polaris 11 failure, it would be very power hungry.
 
My friend and I were discussing Zen APUs, as these will probably be the larger market. is there any news/leaks/theories on what to expect from the AM4 Zen APUs? Are we going to be stuck with yet more 4core 8CU chips... or will AMD start taking it's integrated graphics seriously again?

It really depends on what OEM wants , in the future you could see HBM taking place when it comes to desktop or laptop for "small" machines. This can happen if there is some movement in the market.
 
My friend and I were discussing Zen APUs, as these will probably be the larger market. is there any news/leaks/theories on what to expect from the AM4 Zen APUs? Are we going to be stuck with yet more 4core 8CU chips... or will AMD start taking it's integrated graphics seriously again?

Its also a conflict of interest. IGP is something you more or less have to give away for free. It adds TDP to an already high amount. And you lose sales of dGPUs.
 
It wont have HBM2. 12 CU wouldn't do much either besides die size waste. And if we look on the Polaris 11 failure, it would be very power hungry.

Any sources supporting those claims?
 
Any sources supporting those claims?

Even Raja have stated that HBM isn't nowhere near ready for anything but high end. The cost structure also kills something like an APU completely.

The question is rather, what at all points to a consumer APU with HBM2 from AMD? Where are the AMD statements on it besides wishful thinking?

You are simply using the reversed proofing method. I could just as well say prove that Intel wont use Nvidia IGP and HMC memory. And if you cant it must be true? ;)

AMD couldn't sell Polaris with HBM for 250$ for the same reason.
 
Even Raja have stated that HBM isn't nowhere near ready for anything but high end. The cost structure also kills something like an APU completely.

The question is rather, what at all points to a consumer APU with HBM2 from AMD? Where are the AMD statements on it besides wishful thinking?

You are simply using the reversed proofing method. I could just as well say prove that Intel wont use Nvidia IGP and HMC memory. And if you cant it must be true? ;)

AMD couldn't sell Polaris with HBM for 250$ for the same reason.

I suspected the guy you responded to was merely reading his Christmas wish list.

Meaning both of you are speculating. That's all wanted to know, really.
 
I suspected the guy you responded to was merely reading his Christmas wish list.
Meaning both of you are speculating. That's all wanted to know, really.

What is speculating on those hardware parts that make sense? Look at the mobile space, less chips bigger profits as soon as that happens for HBM then you have everything in one chip , Been happening these past several years so why should it stop now ?

But if tomorrow some OEM comes to AMD and says we want APU with HBM2 then you think they are going to say no?
 
What is speculating on those hardware parts that make sense? Look at the mobile space, less chips bigger profits as soon as that happens for HBM then you have everything in one chip , Been happening these past several years so why should it stop now ?

But if tomorrow some OEM comes to AMD and says we want APU with HBM2 then you think they are going to say no?
Well, to start with, AMD won't sell at loss, and as such will just say no unless OEM is willing to pay a hefty price.
 
An HBM apu would probably have to be a high end product which might be a hard sell since 'integrated graphics' still has a bad name. I think it would make more sense as a server product first as you could probably fit two into a 1U space and sell it as an 8 core + gpu 1U server. Just speculation on my part of course.
 
HBM isn't that expensive, its the TSV that adds cost, and cpu package isn't that far off in cost to a full on TSV anyway, APU likely uses a single HBM2 stack, that will still provide more then double the bandwidth of a quad channel ddr4 memory bus, and with lower latencies.
 
An HBM apu would probably have to be a high end product which might be a hard sell since 'integrated graphics' still has a bad name. I think it would make more sense as a server product first as you could probably fit two into a 1U space and sell it as an 8 core + gpu 1U server. Just speculation on my part of course.

Take a look at consoles it is not "integrated graphics" and yet it is ;) And there not doing that badly and at this point in time something with HBM cost wise does not make much sense but in the foreseeable future things might change.
 
Don't get me wrong, I would love the thing. I just think servers is where it will show up first mostly due to the cost of HBM.
 
AMD is going to surprise some with the extent they use HBM, most of the FUD around the cost is nonsense pushed by Micron and Intel, whose HMC tech is looking more and more like a dead end. HBM isn't all that expensive, its the TSV that is spendy.

AMD will push hard to scale TSV's as much as possible much of their future server product line depends on dropping costs of the tech. They will be able to use TSV based MCM modules to bring out core counts at speeds intel would never dream of. Amd being able to use 8-16core dies in sets of 2 to 8 dies together on TSV MCM modules will give them major yields advantages over Intels monolithic 28-32+ core single die chips.

Their is also HBM3 and low cost HBM planned which will use simplier TSV designs by cutting the number of data lines in half from 1154 to 512(low cost HBM also drops ECC support), and clocking the lines faster.
 
AMD is going to surprise some with the extent they use HBM, most of the FUD around the cost is nonsense pushed by Micron and Intel, whose HMC tech is looking more and more like a dead end. HBM isn't all that expensive, its the TSV that is spendy.

AMD will push hard to scale TSV's as much as possible much of their future server product line depends on dropping costs of the tech. They will be able to use TSV based MCM modules to bring out core counts at speeds intel would never dream of. Amd being able to use 8-16core dies in sets of 2 to 8 dies together on TSV MCM modules will give them major yields advantages over Intels monolithic 28-32+ core single die chips.

Their is also HBM3 and low cost HBM planned which will use simplier TSV designs by cutting the number of data lines in half from 1154 to 512(low cost HBM also drops ECC support), and clocking the lines faster.
If HBM is not expensive, why do HBM manufacturers talk about cost cutting measures :)? And yes, whether that's cost of modules or TSV, it's basically a cost of HBM.

As for second paragraph, i'll just mention that AMD's 8 core dies also have uncore, that also gets doubled/quadrupled in MCM. Say hello to power consumption and GMI latency.
 
We are comparing a TSV to the cost of what they would have used for the CPU packaging in the first place, We are not looking at absolute cost of the TSV rather, we are looking at the difference. And no in this context the cost isn't that bad. We also know that TSV costs are relative to the number of traces on them, so a single stack HBM for an l4 cache/igpu memory, would actually cost much less then current 4stack TSV's used on radeon Fury cards.

We don't know enough about GMI yet, but AMD has always done well with their socket scaling, I don't see any reason to think otherwise at this point, They have more experience then Intel with NUMA, and have access to alot of good tech for putting their interconnect fabric together. As well the power cost will not be comparable to existing multi socket configurations, its well known that the significantly shorter traces in the TSV lower power consumption, whilst at the same time improving latency.
 
but AMD has always done well with their socket scaling

Looking at benches for 2S magny cours against benches for 2S 2603v3 convinces me of opposite.

As well the power cost will not be comparable to existing multi socket configurations, its well known that the significantly shorter traces in the TSV lower power consumption, whilst at the same time improving latency.
Except that uncore is not going to disappear from CPU die in MCM. And AMD are not thinking about replacing actual multi socket configs for now.
 
lol, Don't compare core to core performance, Socket scaling is about how much performance per thread is lost as the thread count increases, and granted this varies from bench to bench. AMD generally does pretty well on this score.
 
AMD also bought seamicro for their fabric tech which was hopefully good. We'll find out soon.
 
As far as the APUs go, I believe that AMD is pretty firm on their 4C limit, which is unfortunate IMO. It'd be nice to see them at least increase it to 6C. However, on the flip-side, the APUs are sort of gaining 1 to 2 cores over those of the Bulldozer-era, given these are full on 4 cores. This time it at least carries the bonus of 4 additional threads. :) The Zen APUs are slated for end of 2017, though, so anyone actually interested in them will have awhile to hold out, but at least you'll have the Excavator APUs that will be AM4 compatible to tide you over.

I've also heard speculation (on sites, not forum posts, but I didn't pay enough mind to where as I didn't think I'd need to cite anything lol) about the APUs around the 1H of 2018 that'll come equipped with HBM. As for how plausible or reliable that is, one can't be sure.
 
Seriously, I assume you mean 200 MegaBytes, right? Because 200 mbits is not good at all. ;)

200 megabits, is 200mbps, quite honestly that is quite fast for most older networks, they are usually based on 10/100 mbps. Max transfer speeds on a 200mbps network would be around 25mb/s

but yes i mean 200 megabytes per second, everyone I know abbreviates megabytes per second as either MBps, or MB/s
 
Back
Top