AMD Raven Ridge 'Zen APU' Thread

[ahimsa42]

"As for its embedded GPU, with the company's introduction of Polaris 10, enthusiasts can expect the Zen APU to have a performance close to an RX 460 card but with fewer stream processors"

http://www.christiantoday.com/artic...rst.zen.based.apu.to.arrive.in.2017/90516.htm

 

[Sweepr]

"As for its embedded GPU, with the company's introduction of Polaris 10, enthusiasts can expect the Zen APU to have a performance close to an RX 460 card but with fewer stream processors"

http://www.christiantoday.com/artic...rst.zen.based.apu.to.arrive.in.2017/90516.htm

RX 460 will be a 896 SPs part clocked above 1.1 GHz (>2 TFLOPs) @ 112 GB/s. AMD wishes they had an APU that could come close to this.

 

[frozentundra123456]

Depends on what you call "close" I guess, unless they are talking about some benchmark that can use HSA. With less shaders, lower bandwidth an tdp, both shared with the cpu, I dont see how it could be more than 60 or 70 % of the performance of a 460, assuming your specs are correct for the 460.

 

[Sweepr]

Depends on what you call "close" I guess, unless they are talking about some benchmark that can use HSA. With less shaders, lower bandwidth an tdp, both shared with the cpu, I dont see how it could be more than 60 or 70 % of the performance of a 460, assuming your specs are correct for the 460.

And don't forget RX 460 will be broadly available at low prices by then, as well as GP107 VGAs from NVIDIA. 14nm/16nm raises the bar dramatically for low-end dGPUs.

 

[cbn]

Regarding the 35W situation OEMs have been struggling with, I am assuming with the reported ECC support for Raven Ridge AMD is going to bring back FirePro.

So that is one factor that should help establish the needed 35W APU only chassis (which can also be shared with non FirePro 35W Raven Ridge consumer APU laptops).

With that mentioned, I wonder how many Mobile workstations are actually used on battery power. I would think the predominant usage is AC power (ie, plugged into the wall).

Maybe to make the chassis more compact and/or to provide extra airflow to the VRMs and NAND (which I assume will only get closer and closer to the processor over time) there could be a "transformer" type "35W Notebook (without battery)" that can turn into a 15W Workstation with the addition of a battery in place of a supplementary cooler/fan system.

Notebook without battery (and supplementary cooler in place) = 35W cTDP

Notebook with battery (and supplementary cooler removed) = 15W cTDP (possibly with lower memory clocks and tighter timings on the DDR4 if this would boost CPU performance at this lower TDP)

So, in a nutshell, having the replaceable battery/cooler allows two types of laptops:

1. More compact than usual 35W for folks that mostly use or purely use AC power (but still have the option of 15W battery power for mobile usage if it becomes necessary)

2. More quiet than usual 35W for folks that mostly use or purely use AC power (but still have the option of 15W battery power for mobile usage if it becomes necessary).

Of the two options above, I am thinking number 2 would be more popular.

P.S. The only down side I can think of would be lack of built-in UPS for those using the laptop in 35W mode (ie, without battery). Maybe if necessary there could be a compromise solution where there is a type supplementary cooler that has a very small battery in it (Maybe even just two rechargeable AA batteries).

A third transforming option would be 35W SFF slim tower (with the notebook lid closed and the machine positioned vertically on some type of stand with feet).

So these three options in total:

1. 35W SFF slim tower with battery removed and supplemental cooler in place (mounted in some type vertical docking stand).

2. 35W notebook with battery removed and supplemental cooler in place.

3. 15W notebook with battery installed and supplemental cooler removed.

Adding removable this, that and the other will increase size, increase cost, and reduce robustness and build quality.

The idea behind having the removable cooler (which swaps with battery) is actually to reduce size and/or increase quietness for laptop running 35W on AC power (most, if not all, the time).

With that mentioned, it is true that 35W laptops with built-in batteries can be overall smaller at the same noise/same Watt-hour rating as a 35W laptop with a removable battery. But this comparison is in a different category because it is not a true 35W laptop. It is 15W laptop able to cheat to the 35W mode via the sacrifice of the battery pack. (EDIT: For an example of the power management options available in the past for AMD mobile, take a look at the info in this post regarding Carrizo).

Hypothesizing on size (with noise at full load designed to be equal) here is how I think the laptop categories would rank....

Smallest= 15W laptop running 35W via battery removed and supplemental cooler added.

Larger= 35W laptop with built-in battery

Largest= 35W laptop with removable battery
35W laptop

P.S. To reduce cost, the manufacturer could always ship without the battery as well. (Including only the supplemental cooler).

 

[Phynaz]

A total of 2 people would by that laptop, which why an OEM would never make it.

 

[cbn]

A total of 2 people would by that laptop, which why an OEM would never make it.

And yet people buy small desktops like NUC and Brix all the time though (despite the high cost for the specialized form factor)

And what I am describing is so much more capable than NUC or Brix.

 

[cbn]

Specifically, there is a class of 35W desktop that about the size of NUC and Brix:

http://www.dell.com/us/business/p/optiplex-7040-micro-desktop/pd?ref=PD_OC

http://store.hp.com/us/en/pdp/hp-el...i-pc-(energy-star)?jumpid=Desktops_Finder_PDP

http://www.bhphotovideo.com/bnh/con...ap=y&m=Y&c3api=1876,92051677442,&A=details&Q=

So the transforming desktop concept is an alternative to what I pictured above (but has the added option of being also used as a 15W laptop)

 

[NTMBK]

Specifically, there is a class of 35W desktop that about the size of NUC and Brix:

http://www.dell.com/us/business/p/optiplex-7040-micro-desktop/pd?ref=PD_OC

http://store.hp.com/us/en/pdp/hp-el...i-pc-(energy-star)?jumpid=Desktops_Finder_PDP

http://www.bhphotovideo.com/bnh/con...ap=y&m=Y&c3api=1876,92051677442,&A=details&Q=

So the transforming desktop concept is an alternative to what I pictured above (but has the added option of being also used as a 15W laptop)

With the added cost of a battery, a laptop screen, a keyboard, a trackpad, a docking solution, a highly integrated removable cooling system... It's going to be >2X more expensive than those little desktops.

 

[cbn]

With the added cost of a battery, a laptop screen, a keyboard, a trackpad, a docking solution, a highly integrated removable cooling system... It's going to be >2X more expensive than those little desktops.

They don't need to come with a battery or docking station.

Also keep in mind laptops are a higher volume item. In fact, a common complaint I read about NUC and Brix is that they are often more expensive than what an equivalent spec'd laptop would cost a person (despite the laptop having extra parts in the form of battery, keyboard/trackpad , LCD screen, etc).

 

[NTMBK]

They don't need to come with a battery or docking station.

Also keep in mind laptops are a higher volume item. In fact, a common complaint I read about NUC and Brix is that they are often more expensive than what an equivalent spec'd laptop would cost a person (despite the laptop having extra parts in the form of battery, keyboard/trackpad , LCD screen, etc).

And then in order to actually use the thing, they need to go out and buy them! If you think this thing if going to be high volume, you're mad!

If you want to continue this fantasy, please take it to a new thread.

 

[cbn]

They don't need to come with a battery or docking station.

Also keep in mind laptops are a higher volume item. In fact, a common complaint I read about NUC and Brix is that they are often more expensive than what an equivalent spec'd laptop would cost a person (despite the laptop having extra parts in the form of battery, keyboard/trackpad , LCD screen, etc).

And then in order to actually use the thing, they need to go out and buy them!

No, an AC adapter and a simple plastic stand (if wanting to use purely as SFF desktop, rather than laptop) is all that is needed to get it running.

 

[cbn]

If you think this thing if going to be high volume, you're mad!

Here are the three modes the "transformer" PC is capable of running:

1. Slim SFF 35W desktop (when used with a inexpensive plastic stand with feet)
2. 35W laptop (capable of running only on AC power)
3. 15W conventional laptop.

That is super broad application for a device (ie, not a niche application).

A example of niche application would be something like a completely fanless 35W desktop using a enclosure like Akasa Euler. I wrote about that in this thread---> http://forums.anandtech.com/showthread.php?t=2476471 (see posts #6, #10 and #11)

 

[superstition]

I think a laptop should be made for high-performance gaming that doesn't have a battery at all, to keep cost and weight down — plus, most importantly, to enable more space to be devoted to cooling.

It should be a thick profile model instead of sacrificing performance and quiet cooling for unnecessary thinness. Who cares if the laptop is a few inches thick if it is light enough to carry and has very good performance? A non-gaming version would also be useful as a portable workstation, given the appropriate hardware.

A high contrast MVA non-PWM GB-LED panel with a defeatable strobing backlight would be good.

If AMD could offer an HBM2 APU I suppose such a CPU would work in this type of laptop. But, I think more typical desktop parts (CPU and discreet graphics) would be adapted given the larger form factor and lack of need for battery power.

 

[cbn]

I think a laptop should be made for high-performance gaming that doesn't have a battery at all, to keep cost and weight down — plus, most importantly, to enable more space to be devoted to cooling.

It should be a thick profile model instead of sacrificing performance and quiet cooling for unnecessary thinness. Who cares if the laptop is a few inches thick if it is light enough to carry and has very good performance?

How thick were you thinking?

Could it possibly use a thin mini-itx board (6.7" x 6.7") with the cooler off to the side of the board like the one shown below?

With that mentioned, I am not sure what headers or connections would be needed on the board to connect up the keyboard and trackpad? Or even the LCD screen? (Maybe the keyboard and trackpad could be connected up via the internal usb header and the LCD via a very short external data cable (if necessary)?).

 

[cbn]

A non-gaming version would also be useful as a portable workstation, given the appropriate hardware.

+1, I like the idea of having a person being able to use a FirePro APU.

 

[cbn]

Regarding the discussion of HBM2 (for future APUs) which started early in this thread, I asked the following question in the memory and storage forum on possible applications (beyond iGPU):

http://forums.anandtech.com/showthread.php?t=2480064

P.S. Just to put things in perspective, each stack of HBM2 has a 1024 bit memory interface while a SO-DIMM or DIMM has a 64 bit memory interface. So a stack of HBM2 running at the same level of bandwidth as a SO-DIMM hypothetically would run 16x slower DRAM. And at faster speeds (such a 102 GB/s) for a single 8GB HBM2 stack the speed of the DRAM works out to be 800 level (the same speed as comonly found DDR2). With two 8GB HBM2 stack @ 102 GB/s all that would be needed is DRAM running at 400 level (the speed of commonly found DDR1).

That, if possible, should offer potential for some very tight timings.

P.S. (As an example) Slowest DDR3 speeds I have seen were 800 Mhz (the level of commonly found DDR2) and the slowest DDR4 speeds I have seen are 1600 (the level of commonly found DDR3) so I do know that DRAM has the ability to scale downward to some degree. How slow the DRAM found in HBM2 can slow down I don't have any idea though.

In post #2 (of that thread I linked above) I asked the question about HBM2 (used as system RAM) for SSD host memory buffer (NVMe 1.2).

Here is some preliminary info provided by Anandtech on DRAM-less SSD and host memory buffer:

http://www.anandtech.com/show/10546/toshiba-announces-new-bga-ssds-using-3d-tlc-nand

Toshiba has shared some details about how they plan to make use of HMB and what its impact on performance will be. The BG series uses a DRAM-less SSD controller architecture, but HMB allows the controller to make use of some of the host system's DRAM. The BG series will use host memory to implement a read cache of the drive's NAND mapping tables. This is expected to primarily benefit random access speeds, where a DRAM-less controller would otherwise have to constantly fetch data from flash in order to determine where to direct pending read and write operations. Looking up some of the NAND mapping information from the buffer in the host's DRAM—even with the added latency of fetching it over PCIe—is quicker than performing an extra read from the flash.

Toshiba hasn't provided full performance specs for the new BG series SSDs, but they did supply some benchmark data illustrating the benefit of using HMB. Using only 37MB of host DRAM and testing access speed to a 16GB portion of the SSD, Toshiba measured improvement ranging from 30% for QD1 random reads up to 115% improvement for QD32 random writes.

Table from Anandtech link above called "Performance improvement from enabling HBM:

Randon Read QD1:30%, QD32: 65%
Random Write QD1: 70% QD32: 115%

While it looks like HMB can do a lot to alleviate the worst performance problems of DRAM-less SSD controllers, the caveat is that it requires support from the operating system's NVMe driver. HMB is still an obscure optional feature of NVMe and is not yet supported out of the box by any major operating system, and Toshiba isn't currently planning to provide their own NVMe drivers for OEMs to bundle with systems using BG series SSDs. Thus, it is likely that the first generation of systems that adopt the new BG series SSDs will not be able to take full advantage of their capabilities.

I wonder how much HBM2 would reduce the latency over PCIe bus compared to DDR4? And how would this performance compare not only to DDR4, but also to an SSD using a dedicated DDR3 or LPDDR3 DRAM buffer?

 

[Spartak]

Why in the hell would they do such thing?

There are (were) actually a few good reasons for disabling one or two cores. First one is the most obvious: bad bins with a faulty core can still be sold as a tricore or dual core.

Another very important reason is you can better mix and match your product portfolio at certain power budgets. For example a quad running at 2.7 GHZ might not hit the desired TDP budget, but as a tri-core it could hit 3 GHz. And as a dualcore it could even hit 3.5 GHz but then you are throwing away an awful lot of transistors. Granted, turbo addresses this issue nowadays.

A third reason would be more L3 cache per core, further improving per core performance.

The tricore Phenom / Athlon II were great bang for your buck back then, with faster clocks in a certain TDP budget. So for SFF enthusiasts like me, tricore was the way to go with AMD. Really hoping we will see tricore models again with ZEN that will be both faster in single core performance, and cheaper within a given TDP budget.

.

 

[The Stilt]

There are (were) actually a few good reasons for disabling one or two cores. First one is the most obvious: bad bins with a faulty core can still be sold as a tricore or dual core.

Another very important reason is you can better mix and match your product portfolio at certain power budgets. For example a quad running at 2.7 GHZ might not hit the desired TDP budget, but as a tri-core it could hit 3 GHz. And as a dualcore it could even hit 3.5 GHz but then you are throwing away an awful lot of transistors. Granted, turbo addresses this issue nowadays.

A third reason would be more L3 cache per core, further improving per core performance.

The tricore Phenom / Athlon II were great bang for your buck back then, with faster clocks in a certain TDP budget. So for SFF enthusiasts like me, tricore was the way to go with AMD. Really hoping we will see tricore models again with ZEN that will be both faster in single core performance, and cheaper within a given TDP budget.

.

In is generally more efficient to run many cores at lower frequency and significantly lower voltage, than fewer cores at higher frequencies. On any design manufactured on 14nm LPP it most definitely is. Take a look at R9 Nano or Intel HCC xwell Xeons or Xeon-D for example.

 

[Sweepr]

New roadmap:

 

[DrMrLordX]

The good news is that there is no more Carrizo-L, which means all boards manufactured for AMD's mobile platforms must at least be designed with Stoney Ridge in mind. That might be an improvement. Not sure why they listed Carrizo-L as being 12-35W when it was 10-25W?

 

[The Stilt]

The good news is that there is no more Carrizo-L, which means all boards manufactured for AMD's mobile platforms must at least be designed with Stoney Ridge in mind.

How will that change anything when Stoney Ridge is single channel too

 

[cbn]

New roadmap:

For AIOs, It would great to see AMD make some headway into TVs.

These days all major brand 4K TVs are smart TVs , so why not have TV with PC operating system instead of an ARM based one? This enabled by Microsoft bringing back Media Center and having the All-in-One PC TV booting straight into that?

AIO PC + TV tuner = "All-in-Wonder" (Remember ATI's (now AMD) All-in-Wonder hardware)

 

[DrMrLordX]

How will that change anything when Stoney Ridge is single channel too

Stoney Ridge is single-channel too? What the . . . ugh. AMD screws the pooch again. Cmon guys, you have to make your mobile APUs REQUIRE dual-channel configs to stop OEMs from being idiots.

 

[NostaSeronx]

Stoney Ridge is single-channel too? ....

Well.. Stoney Ridge is a single Excavator core and a single CU cluster. So, of course it will be tied to a single channel. Kaveri -> Bristol Ridge => Two Bulldozer-derived cores and two CU clusters. Obviously, means Dual-channel.

Raven Ridge is four Zen cores and four CU clusters. Qua...wait what its dual channel again?!