New Zen microarchitecture details

[Atari2600]

I have to admit, only 4 SATA ports doesn't really float my boat...


Personally:
Not willing to give up a SATAe port to bring that up to 6, I already have an SM951 and want to leave the other free for the future.
If you need a couple of DVD drives*, that means you've only ports for 2 SSD/HD?

*they are still used and needed due to the inertia of others, so a lecture on the advantages of the cloud or usb sticks will not help.

I suppose a PCIe SATA card can solve it, but thats not ideal.

 

[Elixer]

I have to admit, only 4 SATA ports doesn't really float my boat...


Personally:
Not willing to give up a SATAe port to bring that up to 6, I already have an SM951 and want to leave the other free for the future.
If you need a couple of DVD drives*, that means you've only ports for 2 SSD/HD?

*they are still used and needed due to the inertia of others, so a lecture on the advantages of the cloud or usb sticks will not help.

I suppose a PCIe SATA card can solve it, but thats not ideal.

It does have 6... if you are talking about the MSI board pics.

In addition we see six SATA ports, a U.2 port and an internal USB 3.1 connector. Internal USB connectors appear to become rather popular with the latest generation of motherboards since more and more vendors are adapting to this trend. Since this model should be the one overclockers want to use, there are also features like onboard buttons and a debug display.

 

[Atari2600]

Cheers Elixer - although that does appear to be coming at the cost of one of the SATAe ports (?)

 

[ozzy702]

I have to admit, only 4 SATA ports doesn't really float my boat...


Personally:
Not willing to give up a SATAe port to bring that up to 6, I already have an SM951 and want to leave the other free for the future.
If you need a couple of DVD drives*, that means you've only ports for 2 SSD/HD?

*they are still used and needed due to the inertia of others, so a lecture on the advantages of the cloud or usb sticks will not help.

I suppose a PCIe SATA card can solve it, but thats not ideal.

Agreed completely. Why wouldn't they build more in? Especially for high end boards. I'd have to think that the CPU can handle a ton of sata devices since they'll be marketed to servers right?

 

[C@mM!]

On a sidenote, have we had confirmation that AM4 supports ECC? Its always been a welcome unsung feature of most AMD boards, and it'd be lovely to see it again.

 

[ShintaiDK]

I have to admit, only 4 SATA ports doesn't really float my boat...


Personally:
Not willing to give up a SATAe port to bring that up to 6, I already have an SM951 and want to leave the other free for the future.
If you need a couple of DVD drives*, that means you've only ports for 2 SSD/HD?

*they are still used and needed due to the inertia of others, so a lecture on the advantages of the cloud or usb sticks will not help.

I suppose a PCIe SATA card can solve it, but thats not ideal.

You have to add it together. 2 SATA ports from the CPU, and 2 or 4 from the chipset.

For M.2/U.2 however its going to be low. Most likely only a single one. And it wont be full speed from the looks of it. (1GB/sec).

 

[psolord]

I hope Asrock builds a new Extreme6 for Zen, like my old P67 Extreme6 that I use for years now.

10 onboard sata man, in 2011, for 150 euros. Ok ok, they are not the fastest satas, but they get the job done. I don't want crazy speed for all storages and I'd prefer to avoid add in cards.

I swear to god, sometimes I think mobo development is going backwards, ffs.

 

[majord]

You have to add it together. 2 SATA ports from the CPU, and 2 or 4 from the chipset.

For M.2/U.2 however its going to be low. Most likely only a single one. And it wont be full speed from the looks of it. (1GB/sec).

I guess it will depend what is onboard summit Ridge . Bristol ridge was released with different config to that earlier preliminary chart, so is a bit up in the air for SR also.

Even it is only a single M.2, it should be at least x2 PCIe 3, which is 2GB/s

 

[StinkyPinky]

Sorry if this is a repeat question but it is a 142 page discussion....

When are AMD going to release more info? Tomorrow at CES?

 

[Elfear]

AMD hasn't released this information to public, but you can probably do the math by looking at the photos of those AM4 boards.

It depends what CPU you use.
But these configs seems to match:

Thanks.

Am I reading that right that there will be 24 lanes between the chipset and processor (16x G3 + 8x G2)?

 

[majord]

Nope, 4x G3 as per Top row ("chipset pcie link")

 

[PPB]

Why not?
Because that's exactly the reason.
It is entirely possible to make a board with 2-phase secondary plane to support Raven properly, however the components required to do that (e.g. IR PowIRStage) cost around 6x the cost of a convential phase. So even when you double conventional phases, it will be 3x cheaper.

I dont know, it seems more sensible to leave the higher spectrum of boards for CPU focused designs. Doubling the secondary phase without doing the same on the CPU one considering it's a 1337 xtr3m3 OC board seems iffy. That is why more people thought it was a 8+2 rather than a 6+4 design in first place.

EDIT: To add more context, even with DDR4 4000 mhz sticks on this board and with RR, how useful is to OC a 768 shader part without incurring in bandwidth bottlenecks? Thinking about future products obviously as RR is +6 months away.

 

[zinfamous]

Sorry if this is a repeat question but it is a 142 page discussion....

When are AMD going to release more info? Tomorrow at CES?

AFAIK,the only thing confirmed for tomorrow, so far, is a Vega architecture presentation. But it would be weird to have CES pass without anything more regarding Zen. Maybe they'll pull one of those Apple-like "one more thing" deals at the end of the Vega session...

 

[Dresdenboy]

LP curve will ALWAYS be lower in Clocks/Power than a HP curve. That's exactly why they are created and thus aptly named 'Low Power Plus'. IBM isn't naive to develop SHP for its highest performance, highest clocking chips

Traditionally, LP has 100-1000x lower leakage at the same voltage/temp, lower on the frequency curve.

Intels 45nm HKMG brought 1000x lower leakage than 65nm!

That NEVER means 100-1000x lower CPU power

Example with Intels 22nm transistor curves:

LP process is typical for mobile parts, or used if a HP is not yet available, or if it offers the required power without sacrificing too much on the clocks. But there's always a hit in clocks from an equivalent MFGs HP process, even if the HP is subnormal.

I'm no process expert, but I think the 14LPP process has a bunch of device options one can choose from to select desired performance/leakage levels at the FinFET level alone (e.g. Lg). On top of that are two CPP options. Here we seem to actually be discussing whether there is an option deserving the name "high performance".

RE TheStilt: he gave some predictions, and it seems he's been more right than most here.

But I did say there is no way A0 will launch. B1 was what I and some others thought will be launch step, if there were no major bugs. If the rumored uop/SMT bug is chip level, then I'd expect B2 to be launch stepping.

Those bug rumours seem to be related to A0 steppings, like the one Canard PC tested ("E4").

 

[Abwx]

Those bug rumours seem to be related to A0 steppings, like the one Canard PC tested ("E4").

Not everyone is aware that there s some water that did flow in the rivers since june, statistically there s a delay for at least one fan out of four, and it s not BJT2, for instance, who would contradict this established number...

Anyway hope that there will be some raw numbers in the coming days to clear all those waters...

 

[The Stilt]

A theoretical question: If Ryzen has 3.4GHz (all core, 8C/16T) stock frequency and consumes 95W of power, what kind of power consumption in your opinion would be reasonable when it is overclocked to e.g 4.0GHz? I'm not asking what you think it will consume, but what you would find reasonable before you are ready to say that it has been pushed too far?

 

[.vodka]

We're used to power consumption usually doubling from stock with a good overclock + overvolt when pushed too far, for example, on a 4c8t mainstream i7... I'd be fine with something similar for Ryzen, maybe dialing it back a bit.

I mean, we used to have Nehalems and their 130w TDPs + x58 itself guzzling another 22w on top of that, powerful little furnace of a platform back in the day. It's a desktop, I want some grunt when it's needed and I'll get it cooled, plus some more load at least in my case brings it all into my PSU's more efficient range.

Seeing how Ryzen and BW-E both overclock similarly from what we know, maybe power scales similarly when OC'd + OV'd between both, save for cases where AVX loads inherently consume more power on the Intel platform.

 

[Abwx]

According to AMD s demo it consume 93 x 0.8 = 74.4W/3.4GHz with Blender as load, and we know that it was overvolted, say by 5%, to ensure total stability , that would make 68W/3.4GHz for the retail chip, at 4GHz it would consume 95W if the process is still very good at this frequency or as much as 115W if it s already a mediocre part of the curve.

Those numbers are of course with Blender, with Prime 95 it would get to 95W at launch base frequencies, that is 3.4-3.5GHz, and in a 120-145W range at 4GHz , depending of said process curves.

Edit ; While we are at it :

http://www.anandtech.com/show/10990...oards-for-am4-laying-the-groundwork-for-ryzen

At the CES 2017 show today, AMD is lifting the lid on some of the upcoming AM4 motherboard designs coming to end users for Ryzen/Summit Ridge and Bristol Ridge. The sixteen boards being displayed come from the major manufacturers, and include overclocking and gaming oriented models.

 

[ozzy702]

A theoretical question: If Ryzen has 3.4GHz (all core, 8C/16T) stock frequency and consumes 95W of power, what kind of power consumption in your opinion would be reasonable when it is overclocked to e.g 4.0GHz? I'm not asking what you think it will consume, but what you would find reasonable before you are ready to say that it has been pushed too far?

I would be perfectly happy with 175 watts @4ghz. Power is cheap and it would rarely be pushed to that level of power consumption so meh. I think it will consume a lot less though.

 

[ozzy702]

I hope Asrock builds a new Extreme6 for Zen, like my old P67 Extreme6 that I use for years now.

10 onboard sata man, in 2011, for 150 euros. Ok ok, they are not the fastest satas, but they get the job done. I don't want crazy speed for all storages and I'd prefer to avoid add in cards.

I swear to god, sometimes I think mobo development is going backwards, ffs.

Agreed. I want a mobo with 10 sata ports and ability to run m.2 drives without sacrificing sata ports. Since this is a new design, why not be forward thinking with your design, once again, especially given server applications?

 

[richaron]

Agreed. I want a mobo with 10 sata ports and ability to run m.2 drives without sacrificing sata ports. Since this is a new design, why not be forward thinking with your design, once again, especially given server applications?

Because having one Gibson to do it all is an extremely outdated design. What they would essentially be doing is adding extra silicone and pins onto the chipset for a tiny portion of the market which already has a number of options. Namely having a motherboard supplier adding a 3rd party PCIe SATA controller onboard, or (something everyone could do), the user just adds a PCIe SATA card

Like I said having one PC do it all is backwards thinking, and if all your PCIe ports are already full then I'd suggest you're living in the past & are in love with the idea of a "mainframe".

Edit: That being said... I still really hope these puppies will run ECC RAM.

 

[bjt2]

A theoretical question: If Ryzen has 3.4GHz (all core, 8C/16T) stock frequency and consumes 95W of power, what kind of power consumption in your opinion would be reasonable when it is overclocked to e.g 4.0GHz? I'm not asking what you think it will consume, but what you would find reasonable before you are ready to say that it has been pushed too far?

Look the red curve: if at 3.4GHz draw 95W, that is about 4 on this graph, at 4GHz will draw less than 150W, that is about 6 on this graph... Actually will draw 150W (6) at about 4.2-4.3Ghz...

 

[coercitiv]

Look the red curve: if at 3.4GHz draw 95W, that is about 4 on this graph, at 4GHz will draw less than 150W, that is about 6 on this graph... Actually will draw 150W (6) at about 4.2-4.3Ghz...

Wait a second, it wasn't long ago you were telling us that Zen will clock at 4Ghz stock. Now you casually mention 4Ghz is for somehting as high as 150W TDP?

Just so we're clear, a minimum of -50% drop in power would put Zen 8C at 100W for 4Ghz. Do you stand by this prediction?

Yes. This was ever my prediction, even in other forums.

 

[AtenRa]

You have to add it together. 2 SATA ports from the CPU, and 2 or 4 from the chipset.

For M.2/U.2 however its going to be low. Most likely only a single one. And it wont be full speed from the looks of it. (1GB/sec).

There is already an MSI TITANIUM AM4 Board that has 2x M.2 ports and U.2

 

[Gideon]

From here.

early performance comparison of Zen versus Kaby Lake have been mostly in AMD's favor, winning most clock for clock results and expected to be a lower price overall to allow customers to add other component upgrades to their build for similar overall spend

What the hell .... I think the guy, who wrote the newsletter, is too hyped up, as this doesn't seem to match with the Broadwell comparisons, but if true ...