AMD Raven Ridge 'Zen APU' Thread

[Topweasel]

Vega suffers from being FUCKING UNFINISHED.

Nah it's was finished. Drivers could have been better and they still had higher clock speed goals. But it is a finished product. But the point being that Vega requires a lot of power at its clock speed, it's generally due to trying to create a really efficient design. If you try to push it out of it's comfort zone for that efficiency power usage sky rockets. Hell you could have 4 of these CPU's in the amount of power that Vega64 leaks at stock. 2-3GHz is the strong point for Zen and it's obvious that at 2.0GHz+ that even with some iGPU load Zen is insanely efficient if it can pull off 15w with a 4c part.

 

[PhonakV30]

LOL!

AMD PR !

 

[Yotsugi]

Nah it's was finished.

WHILE MISSING AT LEAST TWO CRUCIAL UARCH FEATURES?
ARE YOU SHITTING ME PAL?
SOMEONE GOTTA BOMB FORMER ATi CAMPUS FOR THAT.

 

[PhonakV30]

Full Slide can be found on this :

https://www.anandtech.com/show/1196...-vega-and-updated-zen?utm_source=notification

 

[Topweasel]

Soldered RAM, on Ryzen Mobile? Is it considered that much of a low-end platform by OEMs? Heck, half of my Atom-based laptops, don't even have soldered RAM... :|

It's about packaging not about being cheap. Sometimes being cheap also applies. But realistically adding slots adds several mm to the design.

 

[krumme]

It's about packaging not about being cheap. Sometimes being cheap also applies. But realistically adding slots adds several mm to the design.

Yes and as nobody changes ram besides 0.1% better looks win.

 

[Topweasel]

Yes and as nobody changes ram besides 0.1% better looks win.

Exactly, Heck even as an enthusiast my dream is to get this into a Surface like Tablet to replace my SP3. Hoping by the beginning of the year next year I will see one, or I might get the ROG 8c laptop as a semi portable workstation. But really I am hoping for a true Tablet option a 4c replacement for my Surface is more important than the half dozen of times I would want the 8c on the laptop next year.

 

[The Stilt]

I expected Raven Ridge to rule the roost, but @ 15w? I did not see that coming. AMD is not messing around.

I'd love to know how long the Raven mobile SKUs are allowed to / can maintain the 25W boost, until they (gracefully) taper down to 15W.
At those power limits even the lower-end systems shouldn't be too badly thermally limited.

 

[Glo.]

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.

 

[Jan Olšan]

Dunno about you but IMHO it was clear the 2500U and 2700U were going to be 15W.
The U-suffix was clearly chosen to match Intel and they have different letters for 35-45W chips. And we know there are ES chips with higher base clock in FP5 package which will fit that TDP likely.

 

[Topweasel]

Dunno about you but IMHO it was clear the 2500U and 2700U were going to be 15W.
The U-suffix was clearly chosen to match Intel and they have different letters for 35-45W chips. And we know there are ES chips with higher base clock in FP5 package which will fit that TDP likely.

I get what you are saying and 15w is probably something people could have inferred from the U name (though it could have been a designator for Mobile Ryzen and now a TDP rating). But with this you are seeing Clock speeds competitive (through lower) on a 4 core CPU with a fairly large iGPU using as much power as a space typically reserved for 2 core CPU's. That is the revelation. "Knowing" that it was 15w is only half the battle.

 

[DrMrLordX]

You have ceased looking at this objectively, as I told you before, take a step back and rethink this.

Only a tiny percentage (read: a few "enthusiasts") actually upgrades a laptop. Most buyers will never pop it open, much less think about the DIMM configuration.

An HP Envy has dual channel RAM up to 8GB, so make sure your monkey's paw has the same number of fingers held up that it used to:

Odd that they don't have a 16GB option, but it's good that they went dual-channel. Someone got the memo.

So just below the GTX 950M for a 15w part thats really impressive.

I agree. Like TheStilt, I find myself wondering how long these chips are allowed to operate outside of their TDP range before throttling back. Should be interesting to see battery life tests on these.

The power management really, REALLY makes me wanna jizz.

It's those DLDOs at work.

Yes and as nobody changes ram besides 0.1% better looks win.

See above, and thank you for confirming what I've been saying for awhile. Soldered SODIMMs are practical for the majority of buyers. I'm not 100% sure I like the obvious RAM limits, but at least they're using a dual-channel config. They are probably saving some money going with a soldered solution, and the end-user doesn't have to worry about upgrading RAM to get full memory performance at the rated speed.

 

[moinmoin]

I get what you are saying and 15w is probably something people could have inferred from the U name (though it could have been a designator for Mobile Ryzen and now a TDP rating).

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

 

[bsp2020]

I just want to bounce this idea.

At 15W, it seems to perform really well. I also assume that this chip has full x16 PCIe pins? Knowing that 2 socket EPYC is connected using PCIe pins (dual purpose IO pins), what is the chance that AMD will allow dual APU laptop?

I mean, all the components are in place now. We know that IF runs over PCIe pins. 2 socket EPYC system has proven that AMD can build system that scales well using 2 sockets connected using PCIe pins. It would add so much more performance for the engineers who crave CPU power than adding a discrete GPU in a notebook.

Basically, such setup would offer 1.5X performance of current Macbook Pro with 2.9GHz quad-core 7th-generation Intel Core i7 processor + Radeon Pro 560 while consuming less power than just one of the chips (CPU is 35W TDP right? GPU is 35W TDP https://www.notebookcheck.net/AMD-Radeon-Pro-560-GPU-Benchmarks-and-Specs.226608.0.html) The board should not be any bigger than the current board since we are talking placing another APU in place of discrete GPU. 1.5X performance at half the power. We are talking 3X perf/watt.

All the pieces of required technology is there. Whether AMD has used the same PCIe/IF phy or not is the only question. If they did, AMD should be able to build such system with only firmware/software tweaks.

Anybody want to comment?

 

[Yotsugi]

Anybody want to comment?

Why the hell would one want NUMA in laptops? That and mGPU.

WHY?

 

[Excessi0n]

Full Slide can be found on this :

https://www.anandtech.com/show/1196...-vega-and-updated-zen?utm_source=notification

So Raven Ridge has per-core frequency and voltage control... if that's configurable in the BIOS then we could see per-core overclocking on future desktop chips! Wonder if anyone at AMD has thought about that?

 

[bsp2020]

Why the hell would one want NUMA in laptops? That and mGPU.

WHY?

The same reason you want MCM/chiplet design in bigger systems. More performance/watt at lower price. You are reusing the chip that you already made to build higher end systems.

 

[Yotsugi]

The same reason you want MCM/chiplet design in bigger systems. More performance/watt at lower price. You are reusing the chip that you already made to build higher end systems.

You're suggesting NUMA setup in a laptop when OEMs are too greedy for damned dual channel?
Ayy lmao.
Besides NUMA in laptop?
Don't be so silly.

 

[OTG]

The same reason you want MCM/chiplet design in bigger systems. More performance/watt at lower price. You are reusing the chip that you already made to build higher end systems.

The perf/watt comes from low clock speed + power management tricks.
They could easily just clock a R7 1700 at 2 ghz and do the same with pretty much any discrete GPU.
All the benefits, none of the multi-cpu/gpu headaches.
In a laptop it'll be limited by cooling anyhow, just like a 2xAPU would be.

 

[bsp2020]

Of course, it won't be for every one. I'm thinking more along the lines of "Ryzen Threadripper 2950X Mobile Platform". Only the most confident OEMs will be allowed to build it.

 

[krumme]

So Raven Ridge has per-core frequency and voltage control... if that's configurable in the BIOS then we could see per-core overclocking on future desktop chips! Wonder if anyone at AMD has thought about that?

I thought i had that opportunity on the 1700 ryzen i bought launch day. Lol
I dont know if if there is anyone on amd that havnt thought of it.

 

[mikk]

Coffeelake Kabylake Refresh: That part is my bad

You said 550 for 8550U. 2700U gets ~700. That's 30%. 28% is about something else. Please take time to read.

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.

 

[Yotsugi]

This is a cherry picked benchmark

This is simply your usual 30s nuclear boost.
Intel does the same as long as chassis is not made of garbage.

In this slide Cinebench runs way slower on 2700U

Can you read? These are sustained results, 5minute loop of running CB15 with thermal throttling and the likes of it.

 

[krumme]

Deleted

 

[mikk]

This is simply your usual 30s nuclear boost.
Intel does the same as long as chassis is not made of garbage.

Can you read? These are sustained results, 5minute loop of running CB15 with thermal throttling and the likes of it.

It's paints a completely different picture because i7-8550U scores 540-550 points @15W sustained in an Aspire 5:
https://www.notebookcheck.com/Test-...-i7-8550U-MX-150-Full-HD-Laptop.258204.0.html


2700U isn't any better in this cherry picked benchmark. Pretty sure KBL-R still has the upper hand quite easily when it comes to CPU performance.