AMD Raven Ridge 'Zen APU' Thread

[NTMBK]

Or how about LPDDR4?

https://news.samsung.com/global/samsung-rolls-out-industrys-first-8gb-lpddr4-dram-package

At 4266 speed it is 2x as fast as DDR4 2133.....and it comes in 8GB packages.

DDR4 is already up to similar speeds: http://techreport.com/news/32553/corsair-pumps-vengeance-lpx-ddr4-ram-to-4600-mt-s But we don't yet have any memory controllers that support it without overclocking.

LPDDR4 would be cool for mobile parts though. Hopefully Banded Kestrel will support it.

 

[ahimsa42]

an intel laptop with LPDDR4-certainly hope it will be available with mobile RR too:

http://www3.lenovo.com/sg/en/laptops/ideapad/ideapad-miix-series/IdeaPad-Miix-510/p/88IPMX50836

 

[cbn]

Anyone want to take a guess (or speculate) on the double precision floating point of Raven Ridge?

Think it will have a lot of double precision (like Bristol Ridge)?

 

[NTMBK]

Anyone want to take a guess (or speculate) on the double precision floating point of Raven Ridge?

Think it will have a lot of double precision (like Bristol Ridge)?

To be honest I hope it has 1/16 DP performance, but very efficient SP. DP just isn't that useful in consumer workloads!

 

[The Stilt]

Anyone want to take a guess (or speculate) on the double precision floating point of Raven Ridge?

Think it will have a lot of double precision (like Bristol Ridge)?

I fail to understand what relevance does the DPFP capacity hold on an APU?

 

[blublub]

Is it possible that RR is already based on Zen+ with higher IPC?

 

[NTMBK]

Is it possible that RR is already based on Zen+ with higher IPC?

There would be precedent- Trinity used Piledriver, and skipped Bulldozer.

 

[PeterScott]

Is it possible that RR is already based on Zen+ with higher IPC?

Dream on.

 

[maddie]

Or how about LPDDR4?

https://news.samsung.com/global/samsung-rolls-out-industrys-first-8gb-lpddr4-dram-package

At 4266 speed it is 2x as fast as DDR4 2133.....and it comes in 8GB packages.

Not bad. 15mm X 15mm per device. Two of these giving 128 bit bus and 68 GB/s. Bandwidth/shader would be almost at the RX 560 level assuming 11 CU on die. With better compression we would have effective parity.

 

[PeterScott]

Not bad. 15mm X 15mm per device. Two of these giving 128 bit bus and 68 GB/s. Bandwidth/shader would be almost at the RX 560 level assuming 11 CU on die. With better compression we would have effective parity.

It doesn't matter what LPDDR4 speed is available. It matters what is supported. It is probably safe to assume RR laptops won't support anywhere near that speed of LPDDR4.

 

[jpiniero]

Does RR even support LPDDR4? Note that it is not the same thing as 'regular' DDR4.

 

[blublub]

Dream on.

Well isn't naming scheme is 2500 instead if 1500 - that's what made me wonder if it's already another version.

 

[PeterScott]

Well isn't naming scheme is 2500 instead if 1500 - that's what made me wonder if it's already another version.

That's pretty weak. There are tons of leaked AMD slides, and they show Zen cores, not Zen+.

Are Zen+ cores even a thing?

I think it goes:
Zen-14nm
Zen-14nm+
Zen2-7nm
Zen2-7nm+

 

[Arachnotronic]

That's pretty weak. There are tons of leaked AMD slides, and they show Zen cores, not Zen+.

Are Zen+ cores even a thing?

I think it goes:
Zen-14nm
Zen-14nm+
Zen2-7nm
Zen2-7nm+

Zen - 14nm
Zen - 14nm+
Zen2 - 7nm
Zen3 - 7nm+

 

[Gideon]

Are Zen+ cores even a thing?

Maybe not, But Pinnacle Ridge will have to do something better than Summit Ridge, maybe it's just clock speeds and process improvements, but hopefully there are at least some uncore tweaks as well

 

[PeterScott]

Maybe not, But Pinnacle Ridge will have to do something better than Summit Ridge, maybe it's just clock speeds and process improvements, but hopefully there are at least some uncore tweaks as well

Pinnacle ridge is Zen 2.

 

[Bouowmx]

Is AMD Zen targeting yearly product cycles?

Zen: 14 nm, 2017 Q1
Zen+: 14 nm +, 2018
Zen 2: 7 nm, 2019
Zen 3: 7 nm +, 2020

Raven Ridge mobile comes this year, but desktop comes 2018. So for the desktop, it might be the +-version, or that the desktop gets 2017-Zen Raven Ridge, and Summit Ridge +-version.

 

[PeterScott]

Is AMD Zen targeting yearly product cycles?

Zen: 14 nm, 2017 Q1
Zen+: 14 nm +, 2018
Zen 2: 7 nm, 2019
Zen 3: 7 nm +, 2020

Raven Ridge mobile comes this year, but desktop comes 2018. So for the desktop, it might be the +-version, or that the desktop gets 2017-Zen Raven Ridge, and Summit Ridge +-version.

There is no Zen+.

AMD doesn't do a lot of tape-outs. RR Mobile and Desktop will be the same part. It will just take a long time to saturate the OEM channel. The same thing you see with the A12 APU, just recently released for desktop usage.

 

[prtskg]

The policy AMD used with Bulldozer cores won't be followed with zen as AMD knew BD cores were turd while zen is money maker. Zen apu will get more attention from AMD then BD apus did.

 

[Gideon]

Pinnacle ridge is Zen 2.

Really? I'm sure I read somewhere something in the lines of:

• Pinnacle Ridge (Zen1; optimized; 14LPP+): Q1/2018 (one year after Summit Ridge)
• Whatever Ridge (Zen2; 7nm FinFET): Q4/2018 or Q1/2019 (one year after Pinnacle Ridge)

If Pinnacle Ridge is indeed Zen 2 then it must be 7nm, therefore all those rumors of it launching in Q1/2018 must be utter BS (considering where 7nm production is at in GlobalFoundries), It couldn't reasonably be much sooner than Q4/2018.

Now, I don't believe for a minute that AMD plans to just limp by with it's current lineup till the end of 2018, while Intel will release both Coffee-Lake and Ice-Lake, they will lose a lot of their thunder if that's the case

EDIT: Here ya go, straight from the horses mouth:

During the follow-up call this afternoon, AMD confirmed that it would be building another generation of Ryzen processors. However, next time around, they will be based on refined ’14nm+’ technology, likely paving the way for efficiency improvements and perhaps better clock speeds.

After that ‘incremental’ update, AMD will begin focusing on ‘Zen 2’, which is confirmed to be built on the 7nm process at Globalfoundries. Current estimates point to a 2019 date for 7nm. This will then later be followed up by ‘Zen 3’, so AMD is set on the CPU front for the next several years.

Considering that every single roadmap leak we've had, says that Pinnacle Ridge is coming in 2018 and it has 8 "Zen" not "Zen 2" cores, i think it safe to say that it will be the very same zen core (with minor tweaks) on an improved 14nm+ process

 

[PeterScott]

Really? I'm sure I read somewhere something in the lines of:

• Pinnacle Ridge (Zen1; optimized; 14LPP+): Q1/2018 (one year after Summit Ridge)
• Whatever Ridge (Zen2; 7nm FinFET): Q4/2018 or Q1/2019 (one year after Pinnacle Ridge)

If Pinnacle Ridge is indeed Zen 2 then it must be 7nm, therefore all those rumors of it launching in Q1/2018 must be utter BS (considering where 7nm production is at in GlobalFoundries), It couldn't reasonably be much sooner than Q4/2018.

Now, I don't believe for a minute that AMD plans to just limp by with it's current lineup till the end of 2018, while Intel will release both Coffee-Lake and Ice-Lake, they will lose a lot of their thunder if that's the case

So you really think AMD is going to have 2 new architecture next year? Zen+ and Zen2?

Even getting one new one out next year will be a big deal. A process tweak (14nm+) in 2018 1H might boost clock speed a bit. Zen 2 with IPC and 7nm in 2H. That is actually an aggressive timeline IMO. Here is the slide:

 

[Gideon]

So you really think AMD is going to have 2 new architecture next year? Zen+ and Zen2?

I have to agree that's unlikely (and i never mentioned zen+, just 14nm+ for the exact slide you just brought up). And yes, realistically Zen 2 is released in 1H/2019 rather than the end of 2018. But as they are porting zen to 14nm+ anyway ( because of Raven Ridge AFAIK) I see little reason not to update the current lineup with 14nm+ before 2019 ...

 

[PeterScott]

I have to agree that's unlikely (and i never mentioned zen+, just 14nm+ for the exact slide you just brought up). And yes, realistically Zen 2 is released in 1H/2019 rather than the end of 2018. But as they are porting zen to 14nm+ anyway ( because of Raven Ridge AFAIK) I see little reason not to update the current lineup with 14nm+ before 2019 ...

Lets not get hung up on what is called what.

This is whole recent conversation is about whether Raven Ridge will have Zen+ cores with improved IPC.

There is no sign that there is any such thing as Zen+ with improved IPC. The next core design is Zen 2. It doesn't make sense that there will be a different core update before Zen. So RR and Ryzen will have the same IPC.

14nm+ will likely show in 1H next year on Ryzen first IMO. It will improve clockspeed/power a tiny bit. Such refreshes are of close to barely noticeable. Wether the first RR chips have it or not, will hardly matter, it simply won't be that significant.

 

[exquisitechar]

Zen+ doesn't exist, it was the previous code name for the Zen successor which is now called Zen 2.

Zen 2 is 7nm and isn't coming anytime soon. Pinnacle Ridge will feature Zen cores. Same goes for Raven Ridge.

 

[cbn]

Anyone want to take a guess (or speculate) on the double precision floating point of Raven Ridge?

Think it will have a lot of double precision (like Bristol Ridge)?

To be honest I hope it has 1/16 DP performance, but very efficient SP. DP just isn't that useful in consumer workloads!

You definitely make a good point about the energy efficiency.

If looking at the Nvidia Quadro GP100 (GP100, 3584sp, 235W) vs. the Quadro P6000 (GP102, 3840sp, 250W) the FP32 of Quadro GP100 is only 86% (10.3 TFLOPs vs. 12 TFLOPs) despite having 93% of the Stream processors (3584sp vs. 3840sp) and 94% of the TDP (235W vs. 250W).

GP100= Nvidia's GPU specialized in FP64 (aka double precison floating point)
GP102= Nvidia's GPU specialized in FP32 (aka single precision floating point)

I fail to understand what relevance does the DPFP capacity hold on an APU?

It is a niche area for certain.

HPC obviously was the (desired) reason AMD chose to equip so much DPFP in Bristol Ridge/Carrizo......but then I am also wondering about workstation loads that rely heavily on DPFP

http://www.anandtech.com/show/11102/nvidia-announces-quadro-gp100

Meanwhile the second market for the Quadro GP100 is the traditional high-end CAD/CAE market. For those more specialized users who need a workstation card with fast FP64 performance and ECC memory for maximum accuracy and reliability, the Quadro GP100 is the first Quadro card since the K6000 to offer that functionality. Arguably this is a bit of a niche, since most CAD users don’t need that kind of reliability, but for those who do for complex engineering simulations and the like, it’s critical

In fact, I am wondering if Autodesk CFD would be one such program? The solver for CFD does use double precision, but does it work with AMD GPUs or just CPUs?