New Zen microarchitecture details

[cbn]

He's got a long history of bouncing from company to company. There's been some criticism of AMD's internal structure and management these last several years. Maybe things were a lot worse than they were when he left the first time and he was disillusioned with the company.

Well, Keller did not take the job with Tesla till four months after he left AMD.

So I'm thinking he was frustrated with K12 future prospects.

(But I don't think it has anything to do with the Zen core)

 

[Exophase]

A K12 APU would not work well without good Linux drivers even if had the best CPU cores. So that is the importance of the GPU as relates to K12.

As far a Linux, I think we are going to see it grow in importance. Just look at all the game developers porting to Linux now. How much of this is being done because they want to break free of x86 Windows vs. How much of this because they want to run ARM in the future?

Anyway, I bring this up because I don't think Zen or Zen's performance is the problem.

Linux is seeing a lot more game ports now because Steam was made available for Linux and SteamOS is based on Linux. 100% of Steam games supporting Linux also support Windows and very close to 100% support MacOS X. I don't think Linux support has much to do with wanting to break free of Windows much less x86 entirely, it's simply a matter of reaching a larger audience.

The x86 Linux userbase is small but pretty constant and has existed for decades. The Linux ARM equivalent doesn't really exist outside of devices like RPi which are in an entirely different league when it comes to gaming (not least of all because they only support OpenGL ES) There's much more that has to happen to get this ecosystem really going and be interesting from a market point of view, making sure driver ports are good is not going to be this hugely critical place to start. Right now I'm not aware that K12 APUs are even on a roadmap.

Anyway if we're going to talk about weak Linux GPU drivers surely that'd hurt Zen APUs more.

 

[ShintaiDK]

3.7 - 3.8GHz base & 4.1 - 4.2GHz turbo is extremely wishful thinking for a 8C/16T CPU at 95W TDP.

I would be shocked to see anything higher than =< 3.2GHz baseclock @ 95W, if that for the full chip.

If it was the case, including 40% IPC. Intel could pretty much just close shop. And AMD would be a 25+ billion $ yearly revenue company in 2017.

So Agree, more than wishful thinking.

 

[ShintaiDK]

The hype train around Zen is practically unreal, no doubt helped along by Lisa Su and Mark Papermaster. If Zen is going to be so great & be the nail-in-the-coffin for Intel that some of the AMD supporters claim, then why did Rory Read and Jim Keller bail before Zen's launch?

Didn't they want to bask in the glory that this amazing super new CPU core was going to bring them?

Exactly. The CPU that would make the stock climb 25x or more and give them endless money via their stock options. Yet they dont want to be part of it.

Its the same show every time, the hyperbole just gets bigger and bigger without any chance of delivering.

 

[ShintaiDK]

Intel's enthusiast series has no competition, that's why they jack the price sky high. I wouldn't expect intel to drop their prices much, but AMD will likely be far more affordable.

Assuming that AMD can compete, what do you think you get?

Wakeup call, history shows us you would just get 1000$+ Zen CPUs.

Nothing changes.

 

[ShintaiDK]

Absolutely they are. The bits and chips guy is claiming 3.7-3.8GHz base, 4.1-4.2GHz turbo, 8 cores/16 threads in a 95W power envelope. If this is true, and Zen hits around Haswell/Broadwell perf/clock, this is an Intel HEDT killer plain and simple.

Intels entire server division would be jobless the next day more or less while revenue would plummet into the ground

 

[NTMBK]

Intels entire server division would be jobless the next day more or less while revenue would plummet into the ground

It's not like the Opteron killed Intel, I'm sure they would survive that I would be more concerned about the mobile division- more likely that would get shut down, with engineers and resources devoted to making server competitive again.

But this is all fantasy, of course I'm not jumping on the hype train until we get Zen reviews.

 

[guskline]

Who does AMD source to make the test MBs for the @ 8 Zens out there?

I agree with NTMBK about the hype train. Sounds like at least 4-6 months at a minimum until we see reviews.

On a final note, I bet Joe Macri of AMD won't be on stage during the Zen roll out!:sneaky:

 

[scannall]

Well, Keller did not take the job with Tesla till four months after he left AMD.

So I'm thinking he was frustrated with K12 future prospects.

(But I don't think it has anything to do with the Zen core)

I doubt that. Jim Keller has always had wanderlust throughout his career. Does great work, then moves on to the next big thing that catches his interest. Not a bad way to do things if you're good enough to do it.

 

[cbn]

Well, Keller did not take the job with Tesla till four months after he left AMD.

So I'm thinking he was frustrated with K12 future prospects.

(But I don't think it has anything to do with the Zen core)

I doubt that. Jim Keller has always had wanderlust throughout his career. Does great work, then moves on to the next big thing that catches his interest. Not a bad way to do things if you're good enough to do it.

It was announced Jim Keller was leaving AMD on September 18th, 2015 and then on January 29th, 2016 Benchlife announced Keller had joined Tesla as the Vice President of Autopilot Hardware Engineering.

So it was a bit over four months.

It was not a planned transition.

 

[Nothingness]

Anyway if we're going to talk about weak Linux GPU drivers surely that'd hurt Zen APUs more.

That definitely is an issue.

AMD Linux drivers suck (I had an AMD GPU at work which had to be replaced due to Catalyst drivers leaking memory). And despite Vulkan being supposedly based on AMD API, guess who was first to provide drivers? Nvidia (and they released these drivers for devices such as Linux/Android Tegra ones).

I can understand AMD can't invest a lot of money on such things and have to prioritize Windows, but that also implies an AMD GPU is no option for me.

OTOH that doesn't mean I won't consider getting a ZEN CPU if AMD deliver, but before getting a ZEN APU I would wait a few months until it's proven there's no Linux issue (I mean for initial boot, as I don't see myself getting rid of dGPU yet).

 

[Maxima1]

Wakeup call, history shows us you would just get 1000$+ Zen CPUs.

While it sounds too far-fetched, obviously the consumers would win. What you just said is asinine bias. Case in point, somehow (considering what you just said and prior statements) Intel would lower prices if AMD was gone. This doesn't even get into all your other assumptions that are always favorable to Intel and/or Nvidia....

Its the same show every time, the hyperbole just gets bigger and bigger without any chance of delivering.

Well, Intel certainly isn't delivering now with their tic-tac-toe strategy now and mobile is shambles, which you had based many assumptions on.

 

[Exophase]

That definitely is an issue.

AMD Linux drivers suck (I had an AMD GPU at work which had to be replaced due to Catalyst drivers leaking memory). And despite Vulkan being supposedly based on AMD API, guess who was first to provide drivers? Nvidia (and they released these drivers for devices such as Linux/Android Tegra ones).

I can understand AMD can't invest a lot of money on such things and have to prioritize Windows, but that also implies an AMD GPU is no option for me.

OTOH that doesn't mean I won't consider getting a ZEN CPU if AMD deliver, but before getting a ZEN APU I would wait a few months until it's proven there's no Linux issue (I mean for initial boot, as I don't see myself getting rid of dGPU yet).

I too had to switch out an AMD GPU because the Linux drivers entered an utterly broken and unfixable state. It was a very old and very low end GPU, but I didn't replace it for performance reasons (although having something much better doesn't hurt)

The situation with an nVidia card isn't perfect either, I get bugs like some of my windows turning black randomly after my computer's been on a while - possibly due to some issue with some particular piece of software doing something, maybe WINE. But at least it nominally works.

 

[ShintaiDK]

While it sounds too far-fetched, obviously the consumers would win. What you just said is asinine bias. Case in point, somehow (considering what you just said and prior statements) Intel would lower prices if AMD was gone. This doesn't even get into all your other assumptions that are always favorable to Intel and/or Nvidia....

Well, Intel certainly isn't delivering now with their tic-tac-toe strategy now and mobile is shambles, which you had based many assumptions on.

AMD even tried 800$ for a 220W turd of a slow CPU. And you think they are going to sell you bargain chips if they can actually deliver performance? Naive doesn't even describe it.

 

[Maxima1]

AMD even tried 800$ for a 220W turd of a slow CPU. And you think they are going to sell you bargain chips if they can actually deliver performance? Naive doesn't even describe it.

So? They weren't able to do it, so what's your point? Why don't you assume the same of Intel? It's also quite rich because you assume the worst of AMD with a strong competitor, yet think Intel will be so benevolent as to lower prices if AMD is gone. It's illogical and inconsistent. So why would I think the performance/$ remains unchanged regardless of whether Zen is high/low clocks or high/low IPC? Something would give, but I certainly agree that it wouldn't be a $100-200 bonanza.

 

[deasd]

So? They weren't able to do it, so what's your point? Why don't you assume the same of Intel? It's also quite rich because you assume the worst of AMD with a strong competitor, yet think Intel will be so benevolent as to lower prices if AMD is gone. It's illogical and inconsistent. So why would I think the performance/$ remains unchanged regardless of whether Zen is high/low clocks or high/low IPC? Something would give, but I certainly agree that it wouldn't be a $100-200 bonanza.

Low-end AM4 board which also support Bristol Ridge would between 100-200$, and I don't think it wouldn't support CPU which more than 95w TDP.
It's funny you all only talk about CPU price but not whole platform cost.

 

[cbn]

Does anyone happen to know the die size of Broadwell-DE (the 8 core variant)?

160mm2.

This includes the eight cores, dual channel DDR4 controller, 28 PCIe lanes, 12MB L3 cache and usb 3.0.

The SATA is on a separate chip located on package.

 

[cbn]

Linux is seeing a lot more game ports now because Steam was made available for Linux and SteamOS is based on Linux. 100% of Steam games supporting Linux also support Windows and very close to 100% support MacOS X. I don't think Linux support has much to do with wanting to break free of Windows much less x86 entirely, it's simply a matter of reaching a larger audience.

For Valve it is all about breaking free of Windows (and the MS app store model).

However, for the game developers I suspect it has more to do with the anticipation of ARM CPUs.

The x86 Linux userbase is small but pretty constant and has existed for decades. The Linux ARM equivalent doesn't really exist outside of devices like RPi which are in an entirely different league when it comes to gaming (not least of all because they only support OpenGL ES) There's much more that has to happen to get this ecosystem really going and be interesting from a market point of view, making sure driver ports are good is not going to be this hugely critical place to start. Right now I'm not aware that K12 APUs are even on a roadmap.

SteamOS is on ARM already via the Steam Link. (Marvell Armada 1500-mini (88DE3005) SoC).

And while the processor is only powerful enough for decoding, I imagine the day will come when Valve can run all of its titles off some type of ARM APU. Example: I can play CS:GO (Valve's most hardware challenging title) at FPS over 60 FPS (the vast majority of the time) in Linux Mint 17.3 with just a stock speed Q6700 and GT 730 GDDR5 @ 1080p low.

So not much in the way of CPU is really needed to play those Valve Titles.

Anyway if we're going to talk about weak Linux GPU drivers surely that'd hurt Zen APUs more.

Yes, that is a serious issue. I will probably make a thread on this in the future to address concerns.

 

[JDG1980]

8C/16T on Intel 14nm is 3.2GHz base/3.7GHz max single core turbo.

All that means is that Intel is conservative with its default clock speeds on E/EP/EX-series chips. It says nothing about what the silicon is actually capable of. The 22nm i7-5960X has a base clock of 3.0 GHz, but as the link I posted previously indicates, it can be increased to 3.5 GHz while still adhering to the 140W TDP. Anandtech's initial Haswell-E review had both 5960X samples hitting over 4.0 GHz, though one needed more voltage and thus drew more power than the other. I've never heard of anyone having a 5960X that couldn't do 4.0 GHz; the only question is how much power it takes to get there.

Now take into account that 14LPP probably has lower power consumption than Intel 22nm (it definitely results in a smaller die size), plus the fact that Zen is a leaner design than Haswell-E (dual-channel versus quad-channel, and separate 128-bit paths combined for AVX instead of a dedicated 256-bit path). 4.0 GHz at a 140W TDP seems quite plausible in that light.

 

[JDG1980]

I don't think these expectations are reasonable.

I don't take these new claims at face value, though they certainly would be nice if true. For my part, I've said before that I expect IPC roughly on par with Sandy Bridge (better in some applications, worse in others), and factory base clocks of somewhere between 3.0-3.5 GHz for the top 8-core SKU (with higher clocks available for lower core counts). This is assuming the 95W TDP figure we've been hearing for some time; a higher TDP would let AMD push the clocks harder if they wanted.

What are your predictions for Zen?

 

[JDG1980]

If it was the case, including 40% IPC. Intel could pretty much just close shop.

Intel didn't close shop when they were selling Netburst and Itanic crap against AMD's far superior K8 lineup.

 

[ShintaiDK]

So? They weren't able to do it, so what's your point? Why don't you assume the same of Intel? It's also quite rich because you assume the worst of AMD with a strong competitor, yet think Intel will be so benevolent as to lower prices if AMD is gone. It's illogical and inconsistent. So why would I think the performance/$ remains unchanged regardless of whether Zen is high/low clocks or high/low IPC? Something would give, but I certainly agree that it wouldn't be a $100-200 bonanza.

It didn't do so in the past, yet it will in the future. No.

 

[ShintaiDK]

All that means is that Intel is conservative with its default clock speeds on E/EP/EX-series chips. It says nothing about what the silicon is actually capable of. The 22nm i7-5960X has a base clock of 3.0 GHz, but as the link I posted previously indicates, it can be increased to 3.5 GHz while still adhering to the 140W TDP.

Your link is based on a water cooling setup.

 

[ShintaiDK]

Intel didn't close shop when they were selling Netburst and Itanic crap against AMD's far superior K8 lineup.

Because they had the Pentium-M. Something easily forgotten it seems. So the laptop segment was still Intel.

And AMD sold every single CPU they could make. They was capacity limited, not sales limited. I thought this was common knowledge.

 

[Soulkeeper]

That definitely is an issue.

AMD Linux drivers suck (I had an AMD GPU at work which had to be replaced due to Catalyst drivers leaking memory). And despite Vulkan being supposedly based on AMD API, guess who was first to provide drivers? Nvidia (and they released these drivers for devices such as Linux/Android Tegra ones).

I can understand AMD can't invest a lot of money on such things and have to prioritize Windows, but that also implies an AMD GPU is no option for me.

OTOH that doesn't mean I won't consider getting a ZEN CPU if AMD deliver, but before getting a ZEN APU I would wait a few months until it's proven there's no Linux issue (I mean for initial boot, as I don't see myself getting rid of dGPU yet).

As a long time linux user i'm very happy with the open source drivers for AMD.
They have really stepped it up the past 2yrs, i'll only buy AMD video cards for this reason now.
FGLRX is less relevant these days, their plan is to develop in the open and have the open drivers eventually replace it completely. Nvidia still puts all their effort towards their closed driver which has traditionally been the best for linux gaming, while their open drivers suck because they have to be reversed engineered with little or no official support. The open source drivers are what users get on a fresh linux install without modification ie: just works.