AMD Raven Ridge 'Zen APU' Thread

[moinmoin]

I have no idea if it is legit or not

https://gfxbench.com/compare.jsp?benchmark=gfx40&did1=53156431&os1=Windows&api1=gl&hwtype1=iGPU&hwname1=AMD+Radeon(TM)+Vega+10+Mobile+Graphics&D2=AMD+A12-9800E+RADEON+R7,+12+COMPUTE+CORES+4C+8G

Going by the model number that would be a very unexpected beats. U designates standard mobile APUs, the performance level of 7 in 2700 should line up with Ryzen 7 1700 so it would be an 8c 16t chip. So far we have only heard of 1xCCX/4c 8t mobile APUs. I'd say not legit.

 

[AtenRa]

Going by the model number that would be a very unexpected beats. U designates standard mobile APUs, the performance level of 7 in 2700 should line up with Ryzen 7 1700 so it would be an 8c 16t chip. So far we have only heard of 1xCCX/4c 8t mobile APUs. I'd say not legit.

By that logic, Intel Mobile 2C 4T Core i7 7500U should be 8C 16T because of the Core i7 7820X.

 

[moinmoin]

By that logic, Intel Mobile 2C 4T Core i7 7500U should be 8C 16T because of the Core i7 7820X.

Not sure how you arrive at that, shouldn't the former be 7820U then?

I honestly don't follow Intel's crazy model number scheme which they seem to redefine every single gen, mostly around prices instead features. For Ryzen the scheme is rather well defined:

You could argue AM4 doesn't apply for mobile parts, but then listing them under power suffix would be superfluous. You could also argue that the performace level of APUs will be based on CPU and iGPU combined instead CPU alone, making Ryzen CPUs and APUs not directly comparable by model number alone, but I sure hope they don't do that.

 

[DeeJayBump]

As the image above shows of the AM4 naming scheme, the 7 (in 2700) in the alleged Ryzen 7 2700U leak simply denotes that it would be the Enthusiast-level (in performance) part in the mobile APU product stack, not that it would have the same core+thread count as it's AM4 desktop-counterpart.

Enthusiast-level performing parts in the Ryzen APU mobile stack are expected to be 4 core + 8 threads in CPU, as you noted.

 

[moinmoin]

And a hypothetical R7 2700G would double the core count?

 

[jpiniero]

And a hypothetical R7 2700G would double the core count?

The desktop models presumably will use 5 or 3.

 

[DeeJayBump]

Unlikely. The G in that case would mean the 2700G has a higher (Desktop-level) TDP, so higher CPU + GPU base/turbo/core clocks and/or higher GPU CU count would be expected in those (DT) APU parts versus the mobile APU offerings, rather than anything beyond 4C+8T. At least that seems the logical way to read the above graphic on naming scheme, IMO. Also, none of AMD's APU product roadmaps for desktop + mobile APU offerings to this point have indicated SKUs with more than 4 cores + 8 threads.

 

[moinmoin]

Well, as I wrote before I'd hope they wouldn't do that, but as all we have seen officially so far are standard desktop and X high performance chips it's anyone's guess how they'll handle the other power suffixes.

 

[DeeJayBump]

The R7 desktop parts offer the analagous comparison here. Consider the mobile 4C+8T APU part the 1700. Same core + thread count as the 1700X & 1800X with lower base/turbo clocks + less GPU CUs to fit into the lower TDP. Then the desktop 4C+8T APUs would be the 1700X & 1800X in this analogy--higher clocks + more GPU CUs. SImple.

 

[Khanan]

Sorry if I'm asking again if it's already been asked but: is a CU on this one deactivated? If yes, is it for yield reasons? I think overall it will be a great APU, best ever - not that it is a hard thing to accomplish seeing past APUs had inferior CPUs and lacked L3 cache.

 

[Seronei]

http://browser.geekbench.com/v4/cpu/compare/3770816

Geekbench, first info on Clock speed and Cache afaik. Looks like it has half the l3 cache of a normal CCX(4mb instead of 8) and 2.2ghz base clock.

 

[AtenRa]

http://browser.geekbench.com/v4/cpu/compare/3770816

Geekbench, first info on Clock speed and Cache afaik. Looks like it has half the l3 cache of a normal CCX(4mb instead of 8) and 2.2ghz base clock.

Southbridge : AMD Carrizo FCH 51 ?????

Fake ??

 

[FIVR]

Southbridge : AMD Carrizo FCH 51 ?????

Fake ??

Probably fake. Ryzen 4 core should have a much higher MT/ST ratio than this. Single thread at 3500 is comparable to top-binned 2C/4T Skylake-Y but the MT is 30% lower. Doesn't make sense for a 4C/8T chip

 

[BeepBeep2]

Southbridge : AMD Carrizo FCH 51 ?????

Fake ??

No, Bristol Ridge also says Carrizo FCH I believe, and so does the Ryzen Desktop platform - they are just reusing the southbridge across all platforms.

Probably fake. Ryzen 4 core should have a much higher MT/ST ratio than this. Single thread at 3500 is comparable to top-binned 2C/4T Skylake-Y but the MT is 30% lower. Doesn't make sense for a 4C/8T chip

The multi threaded test just looks bugged, some scores are lower than the single threaded scores.

I posted my take on the Geekbench leak on reddit:
https://www.reddit.com/r/Amd/comments/6wos2z/intel_i7_8650u_es_vs_amd_r7_2700u_es_geekbench_4/

 

[DisEnchantment]

Patches in Radeon SI for RR.
[Mesa-dev] [PATCH 3/3] radeonsi/gfx9: implement primitive binning
https://lists.freedesktop.org/archives/mesa-dev/2017-September/168270.html

This increases performance, but it was tuned for Raven, not Vega.
We don't know yet how Vega will perform, hopefully not worse.

[Mesa-dev] [PATCH 00/14] AMD GCN tile swizzle
https://lists.freedesktop.org/archives/mesa-dev/2017-August/164897.html

Make of this what you will...

 

[ahimsa42]

this is exactly the kind of laptop i hope will soon be available with a quad core RR apu:

https://www.newegg.com/Product/Product.aspx?Item=N82E16834234721

 

[prtskg]

Patches in Radeon SI for RR.
[Mesa-dev] [PATCH 3/3] radeonsi/gfx9: implement primitive binning
https://lists.freedesktop.org/archives/mesa-dev/2017-September/168270.html

"This increases performance, but it was tuned for Raven, not Vega.
We don't know yet how Vega will perform, hopefully not worse."


[Mesa-dev] [PATCH 00/14] AMD GCN tile swizzle
https://lists.freedesktop.org/archives/mesa-dev/2017-August/164897.html

Make of this what you will...

Why would they tune it for Raven and not vega? Was there no time for vega? Reminds me Raja's statement that programming for vega was an extensive task. Any eta for raven?

 

[msroadkill612]

I have read many posts here, but no mention of whats truly novel about rr apu?

Dunno, but guestimates would be gpu >< cpu interlinks are normally ~14GB/s over the usual pcie 16 lanes after overheads.

Afaik, infinity uses 40GB/s processor links, and may use multiple links. (in theory, the same bandwidth link used by zeppelin die to link its 2x mmx should apply to RR)

http://www.anandtech.com/show/11551...7000-series-cpus-launched-and-epyc-analysis/2

or

https://www.google.com.au/url?sa=i&rct=j&q=&esrc=s&source=images&cd=&cad=rja&uact=8&ved=0ahUKEwik3oa_6ovWAhXIW7wKHZnVCJcQjRwIBw&url=http://www.anandtech.com/show/11551/amds-future-in-servers-new-7000-series-cpus-launched-and-epyc-analysis/2&psig=AFQjCNFM-lpT2nQhew8eGaQF97FoQiSAOQ&ust=1504624809624617

Whatever, the available interconnect speeds between the zen ccx & vega are unprecedented.

Its not often such an important ingredient in the IT mix increases so exponentially.

Coders abhor a vacuum.

 

[Insert_Nickname]

I have read many posts here, but no mention of whats truly novel about rr apu?

In a word? Integration. Right now you have choose between poor CPU / excellent GPU (AMD APUs) or excellent CPU / mediocre GPU (Intel). With RR you can have both. Well, almost...

Further, Carrizo and newer APUs have on die southbridges (PCH/FCH), so you don't really need a chipset for them. Which cuts down on complexity, and should result in cheap mainboards.

Dunno, but guestimates would be gpu >< cpu interlinks are normally ~14GB/s over the usual pcie 16 lanes after overheads.

Afaik, infinity uses 40GB/s processor links, and may use multiple links. (in theory, the same bandwidth link used by zeppelin die to link its 2x mmx should apply to RR)

AMDs APUs already have a massively overkill bus (I think 30-50GB/s) between IGP and memory controller, so there shouldn't be bottlenecks there other then pure memory performance. I wouldn't be surprised either if its somehow related (variation of HyperTransport?) to Infinity Fabric.

 

[PeterScott]

In a word? Integration. Right now you have choose between poor CPU / excellent GPU (AMD APUs) or excellent CPU / mediocre GPU (Intel). With RR you can have both. Well, almost...

That remains to be seen.

It's too bad RR is still MIA, it's giving Intel a chance to preemptively boost mobile core counts. I notice a whole bunch of laptops jumping from intel dual cores to quad cores at the same price point, before RR arrives. One potential advantage gone.

Internal interconnects are a non issue. It's external memory bandwidth that is a HUGE issue for for APU/IGPs. That is the real bottleneck that will hold them both back.

Hurry up Raven Ridge. It seems like AMD always leaves us waiting.

 

[Insert_Nickname]

It's too bad RR is still MIA, it's giving Intel a chance to preemptively boost mobile core counts. I notice a whole bunch of laptops jumping from intel dual cores to quad cores at the same price point, before RR arrives. One potential advantage gone.

I must admit, I was looking from the desktop side of things. RR is very attractive package, because it can do almost anything in one neat bundle. With the exception of high-end gaming. But even then you can add a discrete card easily, and enjoy relatively good CPU performance. Something you can't on BD APU derivatives.

For a general usage cheap desktop PC, RR is a perfect fit.

Mobile might be a different story, due to the reasons you list, but at least competition is back. Those new 15W Intel quads have also taken quite a hit on base frequency to hit that TDP, from what I've seen. Enough that RR might not be at a significant disadvantage. The GF 14nm process is very efficient at lower frequency/power settings. It also has that excellent IGP to fall back on. With far, far better drivers then Intels.

 

[PeterScott]

Mobile might be a different story, due to the reasons you list, but at least competition is back. Those new 15W Intel quads have also taken quite a hit on base frequency to hit that TDP, from what I've seen. Enough that RR might not be at a significant disadvantage. The GF 14nm process is very efficient at lower frequency/power settings. It also has that excellent IGP to fall back on. With far, far better drivers then Intels.

Relative power usage, also remains to be seen. Intels focus for most of the last decade has really been on laptop chips and getting power consumption down.

We are talking about theoretical benefits of an unreleased chip. IMO history shows us that expectations usually exceed practical reality when the product is delivered.

Intel's chips are also a great fit for a general usage PC. In fact something over 60% of the PC market are just that(IIRC). Intel IGP chips with no dGPU.

RR is largely an unknown quantity right now. It should be a large improvement on the previous APUs, but having a better GPU for a big audience that never cared enough to get a dGPU doesn't seem like it will make that big a deal to most people.

It's nice beat Intel's IGP, but this is NOT a dGPU replacement for hardly anyone. Most of the people who care about gaming, enough to care about GPU performance, will still be wanting some kind of dGPU.

 

[Insert_Nickname]

Relative power usage, also remains to be seen. Intels focus for most of the last decade has really been on laptop chips and getting power consumption down.

AMDs power management is every bit as sophisticated as Intels. Perhaps even more so.

We are talking about theoretical benefits of an unreleased chip. IMO history shows us that expectations usually exceed practical reality when the product is delivered.

No arguments there.

Intel's chips are also a great fit for a general usage PC. In fact something over 60% of the PC market are just that(IIRC). Intel IGP chips with no dGPU.

60%? Try 90+%.

But Intel IGPs are what they are. Not bad, but not good either. They are something you live and make do with, but don't expect anything from. Worse, their driver support is lacking, and tend to get cut after a new generation is released.

I'm not arguing against you. I just think overall, AMD might have a better package at a better price. Based on what we've seen so far. But that remains to be seen.

RR is largely an unknown quantity right now. It should be a large improvement on the previous APUs, but having a better GPU for a big audience that never cared enough to get a dGPU doesn't seem like it will make that big a deal to most people.

It's nice beat Intel's IGP, but this is NOT a dGPU replacement for hardly anyone. Most of the people who care about gaming, enough to care about GPU performance, will still be wanting some kind of dGPU.

There is a lot more to GPUs then just 3D gaming performance. Beating Intel in games is really just a bonus.

Don't forget the hugely improved CPU performance compared to Excavator. RR's Vega IGP should perform better because of that fact alone. Excavator is not suited as a gaming CPU. Never has been, never will be.

 

[LTC8K6]

If they have 90% of the market, it seems hard to criticize it.

I wonder if the one chip great CPU/great GPU solution is still a bridge too far?

 

[PeterScott]

If they have 90% of the market, it seems hard to criticize it.

I wonder if the one chip great CPU/great GPU solution is still a bridge too far?

It is while sticking with 2 channel DDR memory.

Look at Xbox One X, it is a fantastic single chip solution, but it has a 384 bit GDDR interface to memory, vs 128 bit DDR on modern APUs.

Bandwidth is 326 GB/s on the XB1X, more like 1/10th of that, so about 30-40 GB/s for an APU. This is a massive bottleneck which makes boosting the GPU section almost pointless.

We aren't going to see much happen until the memory bottleneck is dealt with. But if you move to Triple/Quad channels, or GDDR, then you don't have a mainstream solution anymore.