Thursday, December 14th 2017

Mystery AMD APU with 1,792 SP Shows Up on SANDRA Database

A mysterious AMD APU showed up on SiSoft SANDRA online database, featuring a massive integrated graphics. The chip reports itself to SANDRA as "AMD Fenghuang Raven," and is likely a semi-custom chip being tested by an AMD engineer in the course of its development. SANDRA reports the integrated graphics component as "AMD 15FF Graphics," featuring 1,792 stream processors across 28 compute units, 555 MHz engine clock, and 2 GB of video memory with 182.15 GB/s memory bandwidth. The result doesn't put out too many details about the CPU component, except its 2.40 GHz clock speed. The iGPU scored 98 points on SANDRA graphics tests with Direct3D 11 API, and 39.99 GB/s observed score.
Add your own comment

26 Comments on Mystery AMD APU with 1,792 SP Shows Up on SANDRA Database

#1
OSdevr
Mobile MCM I'll bet.
Posted on Reply
#2
FrustratedGarrett
Shouldn't they be focusing on fixing the performance issues plaguing Ryzen and Vega/GCN? They haven't Ryzen's subpar gaming performance and Vega is a failure from an engineering standpoint.
Posted on Reply
#3
prtskg
It's on sever/workstation platform as pointed out in 2nd pic. Adding that in article would have been good.
Posted on Reply
#4
kruk
FrustratedGarrett said:
They haven't Ryzen's subpar gaming performance and Vega is a failure from an engineering standpoint.
Both seem to perform just fine at 1080p. If you play at 720p (CPU limit) or 4k (GPU limit) then feel free to buy the competition ¯\_(ツ)_/¯
Posted on Reply
#5
prtskg
kruk said:
Both seem to perform just fine at 1080p. If you play at 720p (CPU limit) or 4k (GPU limit) then feel free to buy the competition ¯\_(ツ)_/¯
Come on! he's Frustated...:p
Posted on Reply
#6
Mussels
Moderprator
FrustratedGarrett said:
Shouldn't they be focusing on fixing the performance issues plaguing Ryzen and Vega/GCN? They haven't Ryzen's subpar gaming performance and Vega is a failure from an engineering standpoint.
ryzen doesnt have subpar gaming performance, unless you're excited about 300 FPS at 720p on low graphics
Posted on Reply
#7
silentbogo
I've just noticed this newspost on overclockers.ua and they speculate that this is a new compute accelerator for AI/Deep learning.
What gave it up, is a massive memory bandwidth (read "HBM2").
So, either it's a "beefed-up" version of Core+Vega, but for compute or workstations (the one announced for laptops earlier only has 24CUs), or AMD is working on their own version with Ryzen core and better Vega... Probably to empower Baidu in the conquest to take over the world. :D
Posted on Reply
#8
notb
FrustratedGarrett said:
Shouldn't they be focusing on fixing the performance issues plaguing Ryzen and Vega/GCN? They haven't Ryzen's subpar gaming performance and Vega is a failure from an engineering standpoint.
No.

AMD has to concentrate on the large volume segments: mobile and server. So any "APU" news is important.
As of December 2017 - almost a year of Ryzen and few months of Vega - the first one is almost non-existent in notebooks and the best implementation of the latter is in currently developed Intel's MCM.
And assuming TPU is publishing a press release for every large client that chooses EPYC over Xeon, there are clearly not many of them as well...

Desktop high-end parts (both CPU and GPU) are niche products. Seriously, no one beside gaming/nerd communities cares about them.
And BTW: do you remember how few months ago people here got excited by market share estimations on PassMark? It looked like AMD share is going to explode. The more optimistic AMD fans expected it to reach 30% by the end of the year. And what happened?
https://www.cpubenchmark.net/market_share.html
I assume you're aware of this as well:
http://store.steampowered.com/hwsurvey/processormfg/
Posted on Reply
#9
evernessince
FrustratedGarrett said:
Shouldn't they be focusing on fixing the performance issues plaguing Ryzen and Vega/GCN? They haven't Ryzen's subpar gaming performance and Vega is a failure from an engineering standpoint.
Plaguing? Hyperbole for sure. Ryzen only has less performance in gaming than Intel due to it's lower clocks, which will improve when they release zen plus in a month or two.

Vega is just screwed anyway you look at it. Raja touted a bunch of features that ended up doing nothing. People can defend the man all they want but this is the biggest dud of a video card since Nvidi'a Fermi.
Posted on Reply
#10
Athlonite
It's an R9-285 with HBM2 memory LOL
Posted on Reply
#11
Vya Domus
FrustratedGarrett said:
Shouldn't they be focusing on fixing the performance issues plaguing Ryzen and Vega/GCN? They haven't Ryzen's subpar gaming performance and Vega is a failure from an engineering standpoint.
Nice trolling.
Posted on Reply
#12
RejZoR
Fenghuang = Fenghuang are mythological birds of East Asia that reign over all other birds.

Interesting :)
Posted on Reply
#14
jabbadap
T4C Fantasy said:
Vega 28
Nah, that is not how it will be written. It might be Radeon RX Vega²⁸, though it's most probably APU with hbm2:s(That internal memory bandwidth 182.15 GB/sec can't be from main memory) so it might have it own differentiation from desktop graphics card(Like Mobile vega is Radeon Vega¹⁰/Vega⁸ Processor Graphics).
Posted on Reply
#15
T4C Fantasy
CPU & GPU DB Maintainer
jabbadap said:
Nah, that is not how it will be written. It might be Radeon RX Vega²⁸, though it's most probably APU with hbm2:s(That internal memory bandwidth 182.15 GB/sec can't be from main memory) so it might have it own differentiation from desktop graphics card(Like Mobile vega is Radeon Vega¹⁰/Vega⁸ Processor Graphics).
i dont see an APU ever having 1792 shaders, not with 12nm even. meh you could be right somehow though

Elite APU? this would be most powerful APU of all time with 28 vega CUs
Posted on Reply
#16
jabbadap
T4C Fantasy said:
i dont see an APU ever having 1792 shaders, not with 12nm even. meh you could be right somehow though

Elite APU? this would be most powerful APU of all time with 28 vega CUs
Well yeah it might get quite too claustrophobic to fit such a thing on am4 platform. On that Sisoftware page it is under server workstation platform so maybe it's some bga chip for datacenters? Structure might have some similarities with kaby lake G, which have 24 polaris CUs.
Posted on Reply
#17
notb
jabbadap said:
Radeon Vega¹⁰/Vega⁸
How I hate this superscript naming... What did they think?!

BTW: I needed some info for work and I had to go through AMD's pages. They are a mess! Almost worse than those of Taiwanese motherboard manufacturers...
@ https://gaming.radeon.com/en/product/vega/
Posted on Reply
#18
OSdevr
T4C Fantasy said:
i dont see an APU ever having 1792 shaders, not with 12nm even. meh you could be right somehow though

Elite APU? this would be most powerful APU of all time with 28 vega CUs
jabbadap said:
Well yeah it might get quite too claustrophobic to fit such a thing on am4 platform. On that Sisoftware page it is under server workstation platform so maybe it's some bga chip for datacenters? Structure might have some similarities with kaby lake G, which have 24 polaris CUs.
As I said earlier, MCM (multi-chip module). Intel is working on one with an Intel CPU, Radeon GPU and HBM all in one package why wouldn't AMD be doing the same?
Posted on Reply
#20
escapeclause
Datacenter APU for GPU accelerated compute? AMD did promise such a thing with Vega and Zen.
Posted on Reply
#21
Vya Domus
escapeclause said:
Datacenter APU for GPU accelerated compute? AMD did promise such a thing with Vega and Zen.
Would it be useful though ? Why would a datacenter employ APUs rather than full on discrete GPUs.
Posted on Reply
#22
OSdevr
Vya Domus said:
Would it be useful though ? Why would a datacenter employ APUs rather than full on discrete GPUs.
This is exactly what I'm confused about. I'm not convinced that the whole "server/workstation" designation isn't just because this is early silicon.
Posted on Reply
#23
T4C Fantasy
CPU & GPU DB Maintainer
This would be cool a apu with r9 380 like gpu in it
Posted on Reply
#24
escapeclause
Vya Domus said:
Would it be useful though ? Why would a datacenter employ APUs rather than full on discrete GPUs.
Well it's a question of power-consumption vs computation on a platform level. Datacenters have very big costs getting rid of heat and also with the price of electricity. So if this purported compute APU delivers the FLOPS for less energy it will be attractive.
Posted on Reply
#25
notb
escapeclause said:
Well it's a question of power-consumption vs computation on a platform level. Datacenters have very big costs getting rid of heat and also with the price of electricity. So if this purported compute APU delivers the FLOPS for less energy it will be attractive.
Nope.
In a typical modern datacenter, which mixes CPUs and GPUs (or coprocessors like Xeon Phi), the two types of processors complement each other. CPUs are better at some types of calculations, GPUs are better at others.
If you need more GPU potential, you just add a GPU. There's really no point in adding an APU, which is inherently slower than a PCI-card variant on the same architecture.
Moreover, from a practical standpoint, it's fairly important to have identical GPUs in the system.

And BTW: CPUs also cover the "maintenance" tasks - like moving data around (disks <-> RAM <-> GPU), so replacing them with APUs could actually bottleneck the whole system, not make it faster.
Posted on Reply
Add your own comment