AMD Raven Ridge 'Zen APU' Thread

[Yotsugi]

Pretty sure KBL-R still has the upper hand quite easily when it comes to CPU performance.

It all depends on the chassis.
Are you new to mobile SKUs or what?

 

[krumme]

This is a cherry picked benchmark and we have to wait for some 2700U benchmarks from a neutral source based on a production notebook. Take time to think about.


https://www.computerbase.de/bildstrecke/80377/7/

In this slide Cinebench runs way slower on 2700U. AMD isn't trustable when it comes to such marketing slides.

Mikk you were like 200-300% way off for the tdp. Marketing, throtling what not it doesnt really matter much.

Its just flat out beat the crap out of the competing product and its a surprise its that low tdp and the reason is the new power tech employed. Next to nobody expected that. Let it rest. Be happy.

 

[USER8000]

Mikk you were like 200-300% way off for the tdp. Marketing, throtling what not it doesnt really matter much.

Its just flat out beat the crap out of the competing product and its a surprise its that low tdp and the reason is the new power tech employed. Next to nobody expected that. Let it rest. Be happy.

He also forgets the Ryzen 5 2500U was tested in an Acer Swift3:

https://cdn.arstechnica.net/wp-cont...-Press-Deck-LEGAL-FINAL-page-025-1440x810.jpg

If you look at the Anandtech article,it has this little tidbit:

https://www.anandtech.com/show/1196...md-apus-for-laptops-with-vega-and-updated-zen

The Swift 3 is a little different than the others – we were told that Acer has built this chassis to dissipate 25W of processor power rather than 15W, meaning that Acer is going to be taking advantage of longer turbo modes and better performance numbers than other Ryzen Mobile parts.

Hence,the Core i5 8550U testbed was in a 25W chassis.

From a link earlier in the thread from computerbase.de there was pictures of the Acer Swift3 with the Ryzen 5 2500U:

https://www.computerbase.de/bildstrecke/80377/19/

Notebookcheck also compared the Ryzen 5 Swift 3 with Core i5 Swift 3 models:

https://www.notebookcheck.net/AMD-Ryzen-Mobile-Raven-Ridge-Back-to-the-top.260136.0.html

In the same chassis,the Ryzen 5 2500U was scoring higher than the Core i5 models.

This is not the top Ryzen 7 2700U which is a better bin and should in practice boost higher.

 

[HurleyBird]

This looks really good, and Raven Ridge is probably set to be the best laptop chip in terms of Perf/W, but given the sheer number of ways to subtly fudge the numbers when it comes to laptop testing, I'm still going to consume these initial numbers with a pinch of salt. Even allowing for some bias in AMD's testing, I'm pretty excited. I hope a version with an HBM cache is in the cards for the future.

 

[Yotsugi]

I hope a version with an HBM cache is in the cards for the future.

RR die has no HBM PHY.
And why would you waste HBM2 on 704SP chip?

 

[Topweasel]

Regarding that in a slide released back when Ryzen was launched:

I guess I missed the day that 15w became the standard mobile power level.

 

[Roland00Address]

I guess I missed the day that 15w became the standard mobile power level.

More or less its been this way since Haswell.

The transition started slowly in Sandy, but from Sandy to Ivy to Haswell to the almost non existent Broadwell followed quickly by Skylake and Kabylake, it has been a gradual change with more and more skus being in the sub 20 watt range of 15 to 18w.

 

[Topweasel]

More or less its been this way since Haswell.

The transition started slowly in Sandy, but from Sandy to Ivy to Haswell to the almost non existent Broadwell followed quickly by Skylake and Kabylake, it has been a gradual change with more and more skus being in the sub 20 watt range of 15 to 18w.

Could be. Thought it was still roughly 35w.

 

[moinmoin]

I guess I missed the day that 15w became the standard mobile power level.

Point was there were three Ryzen Mobile designators from the beginning with U being of a middle TDP rating, most of us just were wrong to think it it's for 25-35W and M for 15W. U being for 6-25W makes it interesting to see what M will be used for (2c4t Banded Kestrel alike silicon still coming?).

 

[HurleyBird]

RR die has no HBM PHY.
And why would you waste HBM2 on 704SP chip?

I don't think you should be too confident on the first point. The obvious memory controller area in Raven Ridge looks decently different from Summit Ridge and we don't have a real die shot of Vega - at least to the best of my knowledge, to use as a reference. I'll admit that it's unlikely and there doesn't appear to be any smoking gun on the die shot, but I also wouldn't say with 100% certainty that there is no HBM2 PHY tucked in there somewhere.

As for why one would want such a part? Because it would provide a huge boost to graphics performance with negligible power cost, and if it behaved anything like the 5775c, a solid boost to some CPU tasks as well. Something like a 7700HQ + 960M, or maybe slightly shy of that inside a 15-ish Watt power envelope? That would be a major and lucrative niche, and would be able to meet certain performance and form factor goals that an Intel + Nvidia solution has little chance of. An ultra thin, fanless, 2-in-1 with an HBM2 equipped mobile Ryzen would be awesome.

 

[PeterScott]

I don't think you should be too confident on the first point. The obvious memory controller area in Raven Ridge looks decently different from Summit Ridge and we don't have a real die shot of Vega - at least to the best of my knowledge, to use as a reference. I'll admit that it's unlikely and there doesn't appear to be any smoking gun on the die shot, but I also wouldn't say with 100% certainty that there is no HBM2 PHY tucked in there somewhere.

People need to stop fantasizing about HBM on an APU anytime soon. A mainstream part like this doesn't justify HBM from a cost or performance perspective.

 

[Topweasel]

Point was there were three Ryzen Mobile designators from the beginning with U being of a middle TDP rating, most of us just were wrong to think it it's for 25-35W and M for 15W. U being for 6-25W makes it interesting to see what M will be used for (2c4t Banded Kestrel alike silicon still coming?).

That was my point people I am not sure by the slide that it was apparent that these were going to be 15w chips and on top of that have these kind of clocks for 4 cores.

 

[Topweasel]

People need to stop fantasizing about HBM on an APU anytime soon. A mainstream part like this doesn't justify HBM from a cost or performance perspective.

Agreed as cool as a HBM chip would be a a SoC, there is only one design win worth the development cost. Honestly its probably several generations before it's really worth it in packaging, cost, and performance.

 

[Bouowmx]

Perhaps the next iteration of DDR would make integrated-GPU gaming viable: 2 channels, 4.8 GT/s, 76.8 GB/s, reducing need for HBM.

 

[HurleyBird]

It may also be possible that there's a GDDR5 controller in there somewhere. The only thing holding back Raven Ridge from >= PS4 gaming performance is bandwidth.

 

[raghu78]

Imagine, what product would AMD have if they would just slap there Interposer and HBM2 on the die package.

About Raven Ridge chips, available, It is actually quite huge surprise with the TDP, of this product. Nobody expected this.

HBM2 memory and packaging technology should be mature in a couple of years time. imo we should see AMD bring HBM2 to APUs when 7nm parts arrive in 2020. There is no point adding any more cores to AMD APUs with bandwidth limitations.

Perhaps the next iteration of DDR would make integrated-GPU gaming viable: 2 channels, 4.8 GT/s, 76.8 GB/s, reducing need for HBM.

DDR5 will not be coming before 2020 and even then will take years to become mainstream. 7nm APUs in 2020 will be an even bigger jump than 14nm as it brings a 55% transistor density reduction and a >55% reduction in power over 14nm at same performance (or >40% increase in performance at same power).

and DDR5 will not be coming before 2020 and even then will take years to become mainstream.

It may also be possible that there's a GDDR5 controller in there somewhere. The only thing holding back Raven Ridge from >= PS4 gaming performance is bandwidth.

Actually whats holding back AMD APUs is the immaturity of HBM2 memory and packaging technology in terms of yields and cost. I expect in a couple of years we will see that part of the puzzle get solved. AMD's 25x20 Vision and the original vision of AMD Fusion will be fully realized with 7nm APUs with HBM2 in 2020.

 

[happy medium]

How would this Zen APU compare to a Intel i7 with MX150 graphics in real world gaming @ 1080p?

 

[itsmydamnation]

How would this Zen APU compare to a Intel i7 with MX150 graphics in real world gaming @ 1080p?

At what TDP? at the 15watt 2700u its no contest because the i7 + MX150 cant run that low. At 25watt its still going to be the 2700u. At 50watt TDP its prob the i7 + MX150 but that is a completely different class of laptop.

 

[happy medium]

At what TDP? at the 15watt 2700u its no contest because the i7 + MX150 cant run that low. At 25watt its still going to be the 2700u. At 50watt TDP its prob the i7 + MX150 but that is a completely different class of laptop.

I guess the problem is the Intel has integrated graphics on the cpu eating up power and a MX 150 for gaming @ 1080p. Or is the power usage for the Intel gpu disabled while gaming with the Mx150?

Is there a source that shows the Zen APU using 15 watts when gaming? How does the APU's TDP work?
Does it throttle while gaming?
Does it choke @ 1080p due to bandwidth limitations?
I wonder how it will compare to the new Nvidia MX 130/ core i7 laptops.
I think there is many questions to be answered before we get too excited, but it does look promising.

 

[Yotsugi]

How does the APU's TDP work?

They literally spent like three slides explaining new and shiny power management to you. It's actually really awesome.

Does it throttle while gaming?

If course it does, down to cTDP levels.

Does it choke @ 1080p due to bandwidth limitations?

A little bit. Vega is much more effective as using bandwidth, courtesy of DSBR.
Still needs NGG Fast Path to eke more performance out of every ALU.

 

[Bouowmx]

With NVIDIA Optimus, Intel integrated GPU is just a display controller, idling ~0.1 W, when dedicated GPU is rendering game.

AMD Raven Ridge power limit supposedly works like Intel's: Turbo Boost short (temporary, for the Cinebench score for headlines) and long (sustained) power limit.

 

[coercitiv]

I guess the problem is the Intel has integrated graphics on the cpu eating up power and a MX 150 for gaming @ 1080p. Or is the power usage for the Intel gpu disabled while gaming with the Mx150?

That's not the problem, iGPU power is near zero while the dGPU is active. The difference is the sum of chip TDP: 15W for i7 + 25W for MX 150 results in a total platform power consumption of about 60W. By contrast a platform with 15W TDP CPU and no dGPU will use around 25-30W, around half the power.

The only fair comparison would be i7 + MX150 versus Raven Ridge with 35-45W cTDP setting and equivalent cooling.

 

[majord]

Forget HBM..I think it's more disapointing lpddr4 isnt supported.

This would be far more realistic and have huge advantages that outweigh the inabilty to upgrade.

 

[Yotsugi]

Forget HBM..I think it's more disapointing lpddr4 isnt supported.

This would be far more realistic and have huge advantages that outweigh the inabilty to upgrade.

How many OEMs do actually bother with LPDDR3/4 in the laptops besides Apple?

 

[Elixer]

Imagine, what product would AMD have if they would just slap there Interposer and HBM2 on the die package.

About Raven Ridge chips, available, It is actually quite huge surprise with the TDP, of this product. Nobody expected this.

In the end, the only thing that counts is performance, and, as we have seen in the past, AMD's partners are just not going to have a product out there that fully takes advantage of what AMD brought to the table.
I wouldn't be surprised to find the first batch of releases being a thermal throttling mess and/or a GPU that is starved because lack of dual channels, and that will basically kill AMD's momentum.

Yes, HBM2 would have been really interesting, but the costs are way too high, and the complexity is still a problem for AMD.
Realistically, HBM won't really be feasible for a few more years.