AMD Zen Diagram Leaked and Analysis

Subject: Processors | April 27, 2015 - 06:06 PM |
Tagged: Zen, Steamroller, Kaveria, k12, Excavator, carrizo, bulldozer, amd

There are some pretty breathless analysis of a single leaked block diagram that is supposedly from AMD.  This is one of the first indications of what the Zen architecture looks like from a CPU core standpoint.  The block diagram is very simple, but looks in the same style as what we have seen from AMD.  There are some labels, but this is almost a 50,000 foot view of the architecture rather than a slightly clearer 10,000 foot view.

There are a few things we know for sure about Zen.  It is a clean sheet design that moves away from what AMD was pursuing with their Bulldozer family of cores.  Zen gives up CMT for SMT support for handling more threads.  The design has a cluster of four cores sharing 8 MB of L3 cache, with each core having access to 512 KB of L2 cache.  There is a lot of optimism that AMD can kick the trend of falling more and more behind Intel every year with this particular design.  Jim Keller is viewed very positively due to his work at AMD in the K7 through K8 days, as well as what he accomplished at Apple with their ARM based offerings.

View Full Size

One of the first sites to pick up this diagram wrote quite a bit about what they saw.  There was a lot of talk about, “right off the bat just by looking at the block diagram we can tell that Zen will have substantially higher single threaded performance compared to Excavator and the Bulldozer family.”  There was the assumption that because it had two 256-bit FMACs that it could fuse them to create a single 512 bit AVX product.

These assumptions are pretty silly.  This is a very simple block diagram that answers few very important questions about the architecture.  Yes, it shows 6 int pipelines, but we don’t know how many are address generation vs. execution units.  We don’t know how wide decode is.  We don’t know latency to L2 cache, much less how L3 is connected and shared out.  So just because we see more integer pipelines per core does not automatically mean, “Da, more is better, strong like tractor!”  We don’t know what improvements or simplifications we will see in the schedulers.  There is no mention of the front-end other than Fetch and Decode.  How about Branch Prediction?  What is the latency for the memory controller when addressing external memory?

Essentially, this looks like a simplified way of expressing to analysts that AMD is attempting to retain their per core integer performance while boosting floating point/AVX at a similar level.  Other than that, there is very little that can be gleaned from this simple block diagram.

Other leaks that are interesting concerning Zen are the formats that we will see these products integrated into.  One leak detailed a HPC aimed APU that features 16 Zen cores with 32 MB of L3 cache attached to a very large GPU.  Another leak detailed a server level chip that will support 32 cores and will be seen in 2P systems.  Zen certainly appears to be very flexible, and in ways it reminds me of a much beefier Jaguar type CPU.  My gut feeling is that AMD will get closer to Intel than it has been in years, and perhaps they can catch Intel by surprise with a few extra features.  The reality of the situation is that AMD is far behind and only now are we seeing pure-play foundries start to get even close to Intel in terms of process technology.  AMD is very much at a disadvantage here.

Still, the company needs to release new, competitive products that will refill the company coffers.  The previous quarter’s loss has dug into cash reserves, but AMD is still stable in terms of cash on hand and long term debt.  2015 will see new GPUs, an APU refresh, and the release of the new Carrizo parts.  2016 looks to be the make or break year with Zen and K12.

Edit 2015-04-28:  Thanks to SH STON we have a new slide that has been leaked from the same deck as this one.  This has some interesting info in that AMD may be going away from exclusive cache designs.  Exclusive was a good idea when cache was small and expensive, as data was not replicated through each level of cache (L1 was not replicated in L2 and L2 was not replicated in L3).  Intel has been using inclusive cache since forever, where data is replicated and simpler to handle.  Now it looks like AMD is moving towards inclusive.  This is not necessarily a bad thing as the 512 KB of L2 can easily handle what looks to be 128 KB of L1 and the shared 8 MB of L3 cache can easily handle the 2 MB of L2 data.  Here is the link to that slide.

View Full Size

The new slide in question.

Source: AMD

April 27, 2015 | 07:57 PM - Posted by Master Chen (not verified)

Phenom III, lol.

April 27, 2015 | 09:06 PM - Posted by JL (not verified)

-.- Worthy of being called Phenom III than Zen.

April 28, 2015 | 04:09 AM - Posted by dragosmp (not verified)

Phenom III on 14nm actually. A betting man may say the L3 cache in inclusive for the L2, the 0.5/core to 8MB L3 points to it.

Like Intel did with Core 1, went back to P3 and put a good architecture on a newer process node, it should work.

April 28, 2015 | 12:03 PM - Posted by Anonymous (not verified)

You are completely and utterly wrong. This is not even remotely K8(/K10).

Just because you want it to be Phenom@14nm doesn't make it true.

It looks to be a beefed up Jaguar with Bulldozer Elements and maybe SMT on top.

April 28, 2015 | 12:41 PM - Posted by Josh Walrath

It does look like a very beefed up Jaguar at this level.  When that particular CPU was released and the power/performance ratio it had... I remember many questioning why AMD continued along the Bulldozer route.  Pretty sure a lot of Jaguar DNA will make its way to Zen.

April 27, 2015 | 09:40 PM - Posted by Anonymous (not verified)

Definitely a simplified block diagram, and hopefully by the time of the Hot Chips Symposium in August 2015 there will be a presentation with a little more than a simple block diagram. Certainly there will have to be more execution pipelines if the processor wants to take full advantage of SMT, especially if more than 2 processor threads are supported per core. Implementing SMT will add more complexity to each core but will allow better execution resources utilization and higher IPC.

April 27, 2015 | 10:07 PM - Posted by Anonymous (not verified)

Issue Processors
Width
1 UltraSPARC T1
2 UltraSPARC T2/T3, Scorpion, Cortex-A9
3 Pentium Pro/II/III/M, Pentium 4, Krait, Apple A6, Cortex-A15/A57
4 UltraSPARC III/IV, PowerPC G4e
4/8 Bulldozer/Piledriver
5 PowerPC G5
6 Athlon, Athlon 64/Phenom, Core 2, Core i*1 Nehalem, Core i*2/i*3 Sandy/Ivy Bridge, Apple A7/A8
7 Denver
8 Core i*4 Haswell, Steamroller

April 28, 2015 | 12:35 AM - Posted by Anonymous (not verified)

We need AMD to succeed or the world will be royally screwed.

April 28, 2015 | 09:37 AM - Posted by Mobile_Dom

if there is no competitor isn't that just an anti-trust lawsuit waiting to happen?

kinda like MS and Apple in the 90's, MS bailed them out so they wouldnt be a monopoly and have to go through anti-trust lawsuits

April 28, 2015 | 11:17 AM - Posted by Anonymous (not verified)

AMD hasn't been relevant in the CPU space to anyone except die hard AMD fans since Core 2 Duo, and yet, the overwhelming majority of the computing population enjoys fast and affordable Intel CPUs.

Unpossiblez!

April 28, 2015 | 11:23 AM - Posted by Mobile_Dom

yeah but AMD still exist and sell processors (albeit not all that many)

without any competition *at all* it just seems like a lawsuit waiting to happen, at the moment ou have the choice between the Intel and AMD CPU, there are reasons for picking both.

AMD goes away you *have* to use an Intel chip in your desktop

April 28, 2015 | 04:58 PM - Posted by Anonymous (not verified)

Again, what is the problem with *having* to use an Intel chip when most do by choice already? lol.

April 29, 2015 | 12:21 PM - Posted by Anonymous (not verified)

Um, because the second intel feels it can stop cooperating with AMD and nVidia discrete graphics with no threat of losing CPU share to AMD... it will cripple those connections to sell its own crap 3D cards.

You want a world where HD4000 is as good as it gets for graphics? Then by all means sneer and lie and fanboi intel into a CPU monopoly. We all know what monopoly they will seek next, and not by performing.

May 2, 2015 | 10:24 AM - Posted by collie

well, you are mostly right. Intel does want the gpu market, and I think they have enough money trees to eventually make it happen. But it's not HD4000 that they are gona bring to the party. They will, eventually, be it a discreet card, part of chipset, or on the CPU, they will eventually have different tiered graphics products that will compete with each and every product from the other guys. They would also have the freedom to undercut the competition if they so choose.

April 28, 2015 | 04:59 PM - Posted by ppi (not verified)

This is rather a reason, why eventual buyer of AMD (should Zen fail) may be able to convince Intel to let AMD keep the x86 license despite change of conrol.

May 2, 2015 | 10:35 AM - Posted by collie

well, any lawer could argue that VIA still produces x86. They could also claim that Intel is "Looking for a viable partner to purchase AMD's former licence," be it true or not.

The think that I think people overlook in this area is not everyone WANT to be a x86 producer.

A custom ARM soc is only one chip, and it's purpose is to be better in ones own products. x86 makers have to worry about tiers, markets, process shrinks, binning, you are making chips to compete with your other chips, bla bla bla.

April 29, 2015 | 03:18 AM - Posted by Martin Trautvetter

Last time AMD was on top in the CPU business, they kept dual-core prices high (entry-level desktop dual-cores north of $300) and even raised the top-end prices to $1236. (that's almost $1.5k in today's dollars)

Intel has dominated for going on 9 years now, their high-end desktop CPU is still under $350, and the extreme CPUs top out at $1k.

April 29, 2015 | 06:52 AM - Posted by Anonymous (not verified)

Wow, that's quite a selective memory you got there.

Ever heard of the Core 2 Extreme QX9775?

No?

This was the last Extreme SKU that had an asking price of 14"Ninety-Nine" Dollar. This SKU launched in March 2008. So that was about 7 years ago. Only with Bloomfield half a year later did they change the pricing structure topping out at $999. (technically not true, some mobile SKUs still cost more than $1000)

So if Intel does it it's justified, but when AMD tried it once, they shall forever be known as the price gougers from hell.

April 29, 2015 | 09:56 AM - Posted by Anonymous (not verified)

He's right though.

When AMD had the lead with industry-leading performance with Athlon 64, you couldn't find a 64-bit AMD processor for less than $400. Now, you can't find an AMD processor for more than $400.

My times have changed! I for one have been welcoming the Intel monopoly for the last 9 years, they keep providing me with great, affordable CPUs right around $300.

April 29, 2015 | 11:39 AM - Posted by Anonymous (not verified)

LOL. I can't actually imagine spending $300 for any CPU. Whole systems cost that much based on AMD APUs which perform very well for users doing email, web browsing, video and audio streaming, etc.

Intel has better single threaded performance, AMD better graphics and better parallel performance. Any whole system under about $600-700 should be based on AMD APU not intel, that's been so since APUs got started. Your money is far better spent on an SSD or more DRAM or more graphics DDR5 RAM than on faster CPU on any sub-$1000 system.

May 3, 2015 | 02:09 PM - Posted by Mike S. (not verified)

No, he's wrong and you are making a false comparison. In 2015 for day to day tasks a $70 CPU is fine, so if Intel prices too high they lose sales. In 2006 even everyday users would see a benefit in every step they climbed the CPU price list. Intel pulled the same high prices with early Pentiums.

May 3, 2015 | 02:09 PM - Posted by Mike S. (not verified)

No, he's wrong and you are making a false comparison. In 2015 for day to day tasks a $70 CPU is fine, so if Intel prices too high they lose sales. In 2006 even everyday users would see a benefit in every step they climbed the CPU price list. Intel pulled the same high prices with early Pentiums.

August 29, 2015 | 04:59 PM - Posted by Anonymous (not verified)

"When AMD had the lead with industry-leading performance with Athlon 64, you couldn't find a 64-bit AMD processor for less than $400."

simply not true. while there was a VERY short time when only FX dual-channel CPU's were available (basically server cpu's sold on the desktop), the 754 socket athlon64's were available as well for well under 400 dollars.

April 28, 2015 | 01:44 AM - Posted by Anonymous (not verified)

Now can we see the English translated version of this please JoshTekk?

April 28, 2015 | 03:21 AM - Posted by Martin OnnA A. (not verified)

Now we have to wait a Year to put this baby into our RIG's :D
512Bit thats cool -> Now we need at least 256bit OS.
Win 11 256bit DX12/13 sounds Great.
Next-Gen

April 28, 2015 | 04:05 AM - Posted by Anonymous (not verified)

Just because it might support 512-bit AVX doesn't mean it's a 512-bit CPU and that we'll need more than 64-bit operating systems, you moron. Learn the difference between operating on integers or addresses and operating on data.

April 28, 2015 | 03:26 AM - Posted by JohnGR

AMD in 2015 will also see some gains because of DirectX 12. That will help Bulldozer to look less of a huge failure and APUs more capable in games.

April 28, 2015 | 07:46 AM - Posted by Anonymous (not verified)

For your consideration:

Benefits of DirectX 12 go to Nvidia too.. And on top of that it's going to take a very long time indeed before it even gets to be used in the actual games. Processor wise it wont make Bulldozer any less of a failure than it already was a very long time ago. Actually, this wont help much their APU's neither, cause they are generally quite weak and thus any real gains in overall performance are hardly going to blow anyone's socks off. Those APU's running anything modern in fullhd in the year 2015? Well, frankly, nope..

April 28, 2015 | 12:12 PM - Posted by Anonymous (not verified)

But the CPU benefits of using DX12, or Vulkan, for AMD's processors in the multicore environment, via better multicore support in the respective graphics APIs, will be less dependency on single threaded performance than there was in the past. Single threaded performance is always good for any code than can not be executed in parallel, but a lot of gaming code can be executed in parallel if the graphics APIs are made to take advantage of multicore/multi-processor treaded CPU cores.

So expect any AMD ZEN based CPU cores to not be at such a disadvantage, and the Zen cores with SMT, and the proper execution pipelines to support efficient SMT, and plenty of low latency cache with improved caching algorithms, will be better at single threaded tasks, and provide as many processor threads as the Intel competition. Any AMD APU will have better graphics, and if the newer Zen based APUs have even better integration of the CPU with the GPU, expect the CPU to be able to dispatch more of its floating point workloads directly to the GPU, with even lower latency than previous generation AMD APUs. AMD is gradually fusing the CPU with the GPU with each successive generation of APUs released. An AMD APU with some on module HBM will be able to have an even faster than GDDR5 memory pool of 4 or 8 gigs of HBM memory, in which to stage both code, and frame buffers, with the regular RAM (GDDR whatever) as a last level memory store for the larger data sets that may not be immediately needed by the game.

AMD could make a high end gaming APU that would not necessarily need a discrete GPU for high end gaming. It just depends on the amount of GPU resources that can be put on a single die with 4 full fat CPU cores, and there is northing stopping both AMD and Intel from offering BIG/Little style x86 core based SOCs with plenty of GPU resources on a single die. The Fat CPU cores for better single threaded performance, the little CPU cores for more parallel execution resources in less space.

April 29, 2015 | 11:46 AM - Posted by Anonymous (not verified)

AMDs higher end APUs already are overkill for any Internet gaming. Which is more and more of the gaming world. Remember if you cannot get more than 20-30 position updates per second (33-50ms lag/latency) there is no game play value in having more than about double that in FPS. 70fps with 100ms is drawing 7 frames extrapolated on one start and end point, not that helpful unless playing against bots.

Net gamers should be spending their money on better routers with open source OS (Gargoyle, Tomato, DD-WRT etc.) and their time on learning to tweak those. Not tweaking for more FPS with crappy net latency.

April 29, 2015 | 02:12 PM - Posted by Anonymous (not verified)

Who cares about only internet gaming, and WTF does internet gaming have to do with having a more powerful APU, just about every processor is overkill for internet gaming, and any internet games are just streaming from the internet game's servers, or they are web based simplified games for single player use. People use processors for other things, and I'm hoping for at least a 6 core Zen APU for the laptop form factor. I do a lot of other things on my PC/laptops and especially rendering for ray tracing workloads needs as many cores as possible. Hopefully there will be affordable workstation SKUs with a 16 core workstation APU Zen variant that are reasonably priced compared to the Xeon SKUs. A 16 core workstation APU with beefy graphics may be great for a low cost graphics workstation, and having 8GB of HBM attached to AMD's Greenland graphics and Zen Cores would make a great more affordable workstation part.

From the looks of it the Zen 4 full core units could be made on a single monolithic DIE that contains 4 such ZEN FULL CPU core Units and an on die AMD freedom fabric that could deliver the faster speeds and coherency between the Zen 4 full core units, something like that will make the OEM workstation market take notice.

Silicon interposers with many thousand wide interconnect/memory BUS traces are being utilized to attach HBM memory chips to AMDs GPUs, so what would be the problem with adding 4 Zen units each supporting 4 full Zen cores around a beefy GPU die right along with the HBM memory chips and having a complete system on an Interposer module.

Nvidia will be doing something similar with IBM Power8's and Nvlink connecting the Power8 CPU to a Nvidia GPU accelerator on a mezzanine module for IBM server systems, and Nvidia will be doing that for the government supercomputers with IBM power9s, and Nvidia's Volta GPU accelerators. AMD could use Its Freedom Fabric, which is a competing technology to Nvidia's Nvlink, so what's is the problem with having a coherent connection fabric traced out on an interposer substrate, and the various full core Zen units connected to the interposer and connected to HBM memory all around the periphery of a large GPU die, The interposer would be essentially a little motherboard of its own in a limited way, and that's what a interposer module/package is in the first place, and interposers could have tens of thousands of tiny traces etched out on multiple layers, something that would be impossible to do with the resin based technology used in current motherboard manufacture.

Too many people are looking at this slide and thinking that AMD is only making dies with 4 full Zen cores in a grouping called a unit and that's that, But there is nothing stopping AMD from making a monolithic die with 2 of these 4 fat core Zen units, or more, the engineering in not impossible to make large DIEs with as many units as could fit. But going with the base unit of 4 full Zen cores and placing the units on an interposer along with HBM and Graphics, on a silicon Interposer, with the coherent connection fabric traced out in an ultra wide BUS on the interposer will be a more affordable solution that would allow the separately fabbed functional blocks(CPU, GPU, HBM) to be fabbed on different processes more suited to each die's functionality.

So do not assume that AMD does not have other options, these units are basically modular building blocks to make into larger more powerful systems. Once AMD has the Zen 4 full core units designed these units can be paired with each other and other functional blocks on a single monolithic DIE, or separately for inclusion on a interposer, for many different uses. That's how SOCs and CPUs are designed in the first place, for modularity and scalability, by all the CPU/SOC industry.

April 28, 2015 | 11:21 AM - Posted by Anonymous (not verified)

On low-end configs sure, but they will still be outpaced by Intel CPUs at every step since Intel CPUs will also stretch their legs with DX12 CPU enhancements. Intel has also been trumpeting the reduced power consumption in cases where CPU utilization is lower in DX12.

April 28, 2015 | 05:02 PM - Posted by ppi (not verified)

On all configs that are rather GPU limited, DX12 will help AMD's CPUs.

April 29, 2015 | 09:57 AM - Posted by Anonymous (not verified)

You seem to be confused, friend.

April 28, 2015 | 03:58 AM - Posted by tnt2 palit (not verified)

' One leak detailed a HPC aimed
APU that features 16 Zen cores with 32 MB of L3
cache attached to a very large GPU.'

the devillss 32 mb!!! remember the palit tnt2 days??. has any gpu eber had a 32 mb buffer...what if...Lol...when i could..16...whatever...

April 28, 2015 | 03:59 AM - Posted by tnt2 palit (not verified)

' One leak detailed a HPC aimed
APU that features 16 Zen cores with 32 MB of L3
cache attached to a very large GPU.'

the devillss 32 mb!!! remember the palit tnt2 days??. has any gpu eber had a 32 mb buffer...what if...Lol...when i could..16...whatever...

April 28, 2015 | 08:37 AM - Posted by akaronin (not verified)

"...of a single leaked block diagram..." Do you know you leaked it ? If you don't how do you dare to write "Source: AMD" !?!?!?

April 28, 2015 | 10:07 AM - Posted by Josh Walrath

Don't know who leaked it, but the rumor pertains to AMD.  I think initially it was leaked on a forum, but WCCFTech picked it up really quickly.

April 28, 2015 | 11:39 AM - Posted by Anonymous (not verified)

The source is planet3dnow.de

The member is a new member, so it's uncertain if it's the real or not.

http://www.planet3dnow.de/vbulletin/threads/421433-AMD-Zen-14nm-8-Kerne-...

he already posted another slide.

http://www.planet3dnow.de/vbulletin/threads/421433-AMD-Zen-14nm-8-Kerne-...

pls don't credit wccftech, they steal and steal and steal and don't credit when it's due.

April 28, 2015 | 11:58 AM - Posted by Josh Walrath

Thanks for the clarification!

April 28, 2015 | 09:59 AM - Posted by funandjam

“Da, more is better, strong like tractor!”

@JoshTekk, If you are going to type out an Ahhhhnold S. accent, you need to change it to say:

" Da, more is betta, strong like tracta!”

April 28, 2015 | 10:26 AM - Posted by Josh Walrath

Supposed to be more Russian sounding.  In Russia, more pipes better, strong like tractor!

April 28, 2015 | 10:42 AM - Posted by Josh Walrath

And for reference, I believe that quote was originally posted many years ago on RWT.  Can't remember who posted that first...

April 28, 2015 | 10:58 AM - Posted by funandjam

it was a Russian accent? lol, fair enough, only as long as it is the opulence guy doing the voice-over - https://www.youtube.com/watch?v=rkB9OT2XVvA

April 28, 2015 | 11:22 AM - Posted by Anonymous (not verified)

AMD is FINALLY ditching CMT, thankfully. Great place to start, at least they can admit they made a mistake going with CMT config but they really should have just scrapped the whole Contruction codename core years ago instead of trying (unsuccessfully) to fix it. Broken design, was obvious years ago, Zen should go a long way in fixing single-threaded performance.

April 28, 2015 | 12:07 PM - Posted by Anonymous (not verified)

They didn't have an alternate architecture they could just pull out of their bottoms and send to the fabs. Creating a new architecture takes years. AMD took a huge gamble on the Bulldozer clustered method, hoping that developers would write to take advantage of more cores, but their gamble failed as Intel just juiced IPC and developers just kept writing the way they always have for single-threads. So they were stuck with an architecture for a few years while they got to work designing its replacement.

They did the best they could trying to improve it in stages in the meantime. But to claim that AMD should've scrapped the clustered approach years ago (without anything to replace it) is ridiculous.

April 28, 2015 | 04:57 PM - Posted by Anonymous (not verified)

No what was ridiculous was their unsuccessful attempts to repeatedly polish a turd. Sure a new architecture takes years, but if they scrapped Bulldozer when it initially flopped, we'd already have that new architecture today!

Instead we got Pileocrap, Bobcrap, Steamycrap etc.

They should've just ramped PhII as far as they could take it and develop a new arch in the meantime, but it was obvious there was politics behind it and they simply couldn't admit Bulldozer and CMT was a huge failure. They finally have, but we are now 5+ years more of non-competitive AMD arch as a result.

April 29, 2015 | 03:35 AM - Posted by Martin Trautvetter

I think they were just hoping to hit a Northwood somewhere along the line of turds. Never materialized though, so I agree, they should've just given up on Bulldozer when they had the sims down and knew for a fact that only a major software miracle could save them.

April 29, 2015 | 10:46 AM - Posted by Josh Walrath

I will be very curious how Excavator actually fares as compared to previous core iterations.  I don't expect miracles, obviously, but I wonder how much of a jump in perf and power it really holds?  I guess we will find out soon!

April 29, 2015 | 05:10 AM - Posted by Anonymous (not verified)

FYI, servers have been taking advantage of more cores forever. The problem wasn't developers. The problem was the team of managers and engineers beyond Bulldozer and the unbalanced desing that they gave us.

April 28, 2015 | 12:59 PM - Posted by Shambles (not verified)

I can't wait until 2016-2017 to build my home server which is a shame because I'm sure AMD will be offering a much better core count than Intel is at certain price points. Maybe when it gets replaced in 2020. Hopefully AMD still exists and is somewhat competitive by then.

April 28, 2015 | 05:04 PM - Posted by Bianchi4me (not verified)

Can't hurt, but even massive performance improvements may not be enough to save AMD. Even when AMD was making better performing AND cheaper chips, most people were still lining up to buy the early pentiums. Regardless of performance, Intel has the well established cache of being the premium brand in their segment.

April 29, 2015 | 12:53 PM - Posted by Anonymous (not verified)

"Premium brand" buyers are usually just suckers or fanboi. Apple for instance has released an inferior product in iPhone 6 and 6s, compared to Samsung's and others. The former Apple users who actually use their phones for performance challenging stuff like running external 1080p displays will tell you why they switched.

Computing/communication devices are a commodity, they are basically just net terminals now, focus has shifted to Internet performance and 4K+ displays, with just enough CPU and GPU to keep the frames smooth.

April 29, 2015 | 02:52 AM - Posted by Martin Trautvetter

Josh, I don't think AMD had exclusive caches in Bulldozer, either. The move appears to be away from their previous hodge-podge to a straight-forward fully inclusive design, simplifying snoop-traffic across the board. (no more probing semi-local or distant L2s)

April 29, 2015 | 10:39 AM - Posted by Josh Walrath

Oh man, it has been a long time since I looked over Bulldozer tech specs about exclusive vs. inclusive caches.  Thanks for the note.  Yes, it seems simplicity is best in this instance, especially when process tech allows you to have a ton of cache on a chip without inflating die sizes.

April 29, 2015 | 07:17 AM - Posted by Shortwave (not verified)

APU/On-board graphics are really exciting to me. I LOVE the feeling of PC gaming off of devices I can run off portable solar panels basically. Heh. Or soon hitching off micro-wives maybe?..

I'll be building one of these ASAP for testing!
I figure once I can get 85% of my steam library running at 1080 60fps with respectable settings I'll actually keep it for a while.
Getting VERY close on the desktop environment side for the under 100 watt range.

DX12/Vulcan will make all of this interesting for sure..

I suppose if it'll have to surpass a nice i5>i3's integrated graphics.. I have a feeling it probably will with the right system configuration.. We'll see!

Mobile currently Intel holds the crown with their Cherrytrail/X series. Something a lot of people are missing out on are how you can configure the bios often to improve gaming performance. It's jizz-worthy. I can't wait to get my hands on one, I've already been amazed after tweaking Bay-trail devices.

I but you every time I've commented here it's been on a different system.. HA!!

April 29, 2015 | 07:36 AM - Posted by razor512

when they talk about increased IPC, are they referring compared to the Phemon II chips, or are they referring to the low end APU's?,

It still seems like they are focusing on having a ton of cores, which is still troubling since many applications are not going to make use of something like 16 cores.

For a dual xeon system (12 cores, 24 logical), Photoshop barely uses 5, and premiere pro does not use all unless you are encoding video.

AMD should be focused on getting IPC up to the levels of at east the current Gen core i5 chips.

April 29, 2015 | 10:42 AM - Posted by Josh Walrath

I think that is their target.  Getting anywhere near to even older Sandy Bridge and Ivy Bridge levels of IPC would be a big win for them.  Their 4 core module concept seems interesting, but we have yet to see how well those interconnects work when fusing multiple modules.  I guess they had a lot of practice there with Bulldozer modules...

April 29, 2015 | 11:59 AM - Posted by Anonymous (not verified)

AMD has to include ARM cores in its APUs. Period. It's the only way to really cut power draw to less than Haswell's when idling but on... and the only way to ensure long term compatibility with mobile code based on that less power-hungry RISC model.

Having a nearly-competitive architecture on a larger process should remain AMD's goal. Heat dissipation has always been superior since it was AMD's biggest problem with those wider pathways. Accordingly as smaller 14nm and 12nm process becomes a commodity, AMD can blow past intel simply by shrinking. Competing in single threaded code is a mistake, trying to get performance at expense of more power draw is a mistake, not exploiting the fact that intel cannot for corporate reasons include ARM cores is a mistake, and chasing the $1000-3000 desktop market with anything other than graphics cards is a mistake.

April 29, 2015 | 01:22 PM - Posted by Josh Walrath

The latest roadmaps have been leaked as well.  Low power mobile is now based on K12.  AMD is moving away from x86 for the ultra low power market.  Zen will cover the rest.

April 29, 2015 | 03:00 PM - Posted by Anonymous (not verified)

K12 for affordable tablets, and why should AMD expend the engineering resources trying to get the CISC x86 architecture shoehorned into the very low power usage envelope of the RISC design ARMv8a custom parts and an intrinsically lower power using RISC designs. The whole idea of AMD's "ambidextrous" approach is to have a Custom ARMv8a ISA based design for low power using mobile/tablet usage, while saving the x86 Zen for the PC/laptop/Server markets. Intel is spending billions trying to get its x86 designs into a low power using mobile products range, and still having trouble competing with some ARM designs that are fabbed on a larger process node.

When everybody gets to the 14nm node, Intel will be at an even larger disadvantage trying to get to lower power usage parity with the custom designed ARMv8a ISA based cores from Apple, Nvidia, AMD(when K12 is released), and others. If the process node is equivalent, the designs with the less amount of transistors needed to implement a CPU's core logic are the ones that will use the less power per core. For the mobile market, x86 based designs will always face an up hill battle for the low power usage metric that can make or break a mobile design in the marketplace, that and price(without contra revenue) will work against Intel's x86 mobile offerings.

Graphics on the low cost Mobile market SKUs in not an Intel strong point either, as is GPU power efficiency or graphics power compared to AMD's, Nvidia's, or Imagination's PowerVR designs. Intel's in house graphics for its Mobile SKUs is seriously lacking, compared to the competition, and even when Intel licenses graphics IP, it's never the latest or greatest. AMDs K12, if it has SMT baked in to its custom cores, is really going to give Apple some IPC and performance/watt competition, and AMD's graphics in a custom K12 will be something that even Apple will have trouble matching.

April 29, 2015 | 05:48 PM - Posted by Anonymous (not verified)

interesting.
i personally can't tell if it is real or fake.

hmm so just how far is single-thread performance will be stretching out ?
i think back then amd made the right decision to focus on multi-thread performance instead of single thread, but programmers killed off the FX series.
battlefield does show that the old dinosaur FX chips with the -by nowdays - clumsy 28 nm process still have the juice for it.
if i was a billionare and could afford to sepnd billions on stupid things i would really make an 14 nm vriant of the FX series just for the lulz. probably it would not be all that mutch better, but i would still be interested to see what it would yield.Fx does scale quite linear up to the 5 ghz that is a valid setting in my personal experience from 4 crore to 8 core..

anyways back to the zen core.
for the love of god, i would expect far more cache memory.
128kb L1 is not going to cut it. double it up, make a monster. along with L" and L3 cache memory.
thisone would be feasable and i would really love it.

AMD should use quad channel memory, manufactured onyl by them. make every stick essentially a quad channel memory.
use a custom socket.
thatway they could at least gain some profit on that as well.
so was thatone...

do what microsoft did, offer developers some money if they at least consider to take advantage of your platform.
release a compiler that can help the devs produce stuff running better on your system.
thisone would be unlikely... sadly.

and get rid of the lunatic IHS.
it was a real pain to remove it on FX series.
hte gains made it worthy, but still. why have it when its soo useless...
now thisone would be a musthave.

December 6, 2015 | 11:41 AM - Posted by Anonymous (not verified)

In this article speculation in the form of a vague block diagram from a non confirmed source, I have no comment on it until I see the official diagrams.

In response to the article: Any sufficiently advanced troll is indistinguishable from a knowledgeable authority. G.D.M.

Post new comment

The content of this field is kept private and will not be shown publicly.
  • Lines and paragraphs break automatically.
  • Allowed HTML tags: <a> <em> <strong> <cite> <code> <ul> <ol> <li> <dl> <dt> <dd> <blockquote><p><br>
  • Web page addresses and e-mail addresses turn into links automatically.

More information about formatting options

CAPTCHA
This question is for testing whether you are a human visitor and to prevent automated spam submissions.