Report: Ryzen "Raven Ridge" APU Not Using HBM2 Memory

[darkbasic]

Not a very smart move to buy intel based expensive toys. Intel&Vega Apu laptop will cost much more that HPs 600 usd Raven Ridge laptop. You do not do very much serious jobs with 15-45 watts.

Price is not a problem for a work machine, portability and closed source drivers are. Give me a decent quad core, 32GB+ of ram and Vega + HBM2 and I'll be happy.

[OneBitUser]

Because the Intel+Vega APU will have HBM2. That means I'll have to buy Intel once again: not a very smart move from AMD.

The Intel+Vega Kaby Lake-G is not an APU!
It is an Intel Kaby Lake-H die (complete with an active IGP) and a Vega dGPU in the same package.
So this is like saying that you won't buy a Ryzen 3 because Intel has introduced 6 core/12thread Core i7s.
The two products target completely different markets and users, and have nothing to do with each other.

Also, Raven Ridge will be targeting a much lower price point than any product based on KL-G. Raven Ridge will also be available for the AM4 desktop platform, while Kaby Lake-G will most likely be a highly integrated, small-platform size mobile-only offering.
Plus let us not give up hope of seeing that Raven Ridge "Iris Pro Edition"!

[Nille_kungen]

Sound pretty normal to me.
To include dedicated HBM2 graphics memory in an APU would make it expensive and not a generic 15W APU.
I doesn't even find the graphics performance a disappointment but rather what i would expect.
I think it's the "i dream away and make my own expectations and then cry out loud on the internet that it's a disappointment when it doesn't live up to my fantasy performance" syndrome.

[marek]

Thanks Marek for the update. Can you share at least how is RR APU GPU performance stacks vs the green/blue mobile solutions.

I can't say anything that hasn't been published. Common sense tells me that a GPU with HBM should be faster. As for Raven, you'll have to search for benchmark articles yourself.

[TitanFighter]

Slow down guys.

Currently we have HP Envy x360 15z which SUPPORTS dual channel and max RAM can be 16GB (check official hp store http://store.hp.com/us/en/ConfigureV...318&quantity=1).
Also Acer is going to release Swift 3 with dual channel. Lenovo - IdeaPad 720S with SINGLE channel. Details -> https://www.anandtech.com/show/11964...nd-updated-zen (ignore info about max RAM).

About graphic test of HP x360 15z which have Ryzen 5 2500U and just 8 CUs (512 Shaders) instead of 10CUs (640 Shaders) in Ryzen 7 2700U: https://hothardware.com/reviews/ryze...mance-analysis . Tests show this APU wins GeForce 940MX.

If someone wants to check benchmark list https://www.notebookcheck.net/Mobile...ist.844.0.html

[TheLexMachine]

What we were told back in the day, was that Raven Ridge would have HBM. They didn't specify that the mobile would have it, they just said Raven Ridge would have HBM. Obviously, they have not implemented that on these low-end products, which are clearly the low-end mobile APUs without HBM, that were reported in the recent leaks. Whether or not the desktop's high-end APUs will have HBM with the yields reportedly being very low and Intel buying up HBM for it's upcoming CPU/GPU sammich, remains to be seen. All we do know is that AMD is going to have HBM on it's upcoming high-end video cards. I for one, want a Raven Ridge APU with HBM, in a compact NUC-like system for Windows gaming.

[blackshard]

What we were told back in the day, was that Raven Ridge would have HBM. They didn't specify that the mobile would have it, they just said Raven Ridge would have HBM. Obviously, they have not implemented that on these low-end products, which are clearly the low-end mobile APUs without HBM, that were reported in the recent leaks. Whether or not the desktop's high-end APUs will have HBM with the yields reportedly being very low and Intel buying up HBM for it's upcoming CPU/GPU sammich, remains to be seen. All we do know is that AMD is going to have HBM on it's upcoming high-end video cards. I for one, want a Raven Ridge APU with HBM, in a compact NUC-like system for Windows gaming.

Well that's quite a nonsense putting HBM. It should be quite simple to understand that if you need more bandwidth you go with HBM, but you still have to remember that the memory bandwidth must match the processing power (in terms of shader processors) you got.

Now, if you think that VEGA 64 (which has HBM2) has 4096 shader processors and these APUs got 640 shaders at best, you can realize that using HBM2 is a complete waste. GDDR5 also has the big problem that it requires a quite complex memory controller (increases power budget) and the chips consume more power by themselves, so cheap DDR4 is the natural choice for a local buffer of cache to hide latencies of main system memory. If AMD engineers did the thing well, the iGPU should not suffer too much the single channel interface against system memory

[orome]

Now, if you think that VEGA 64 (which has HBM2) has 4096 shader processors and these APUs got 640 shaders at best, you can realize that using HBM2 is a complete waste. GDDR5 also has the big problem that it requires a quite complex memory controller (increases power budget) and the chips consume more power by themselves, so cheap DDR4 is the natural choice for a local buffer of cache to hide latencies of main system memory. If AMD engineers did the thing well, the iGPU should not suffer too much the single channel interface against system memory

even iGPU can use more bandwidth than DDR4 can provide. Most AMD GCN dGPU designs have ~7-9 GB/s bandwidth per CU. raven ridge (supposedly up to 11CUs) would ideally have 77-99GB/s dedicated bandwidth, which is more than even 2ch ddr4 can give, not to mention it's shared with CPU.
that said HBM2 (256/GBs per stack) is probably overkill, HBM1 would be enough to satisfy both CPU cores and iGPU.

[wizard69]

Im not sure what all the whining is about here. This single channel memory is not AMDs fault. Further this is a 15 watt platform that appears at this moment to be doing very well performance wise against other 15 watt platforms.

As for HBM memory i still expect AMD to offer such an APU. This simply due to every APU type processor being bandwidth starved to date. In other words they need faster RAM subsystems to really get any value out of todays faster GPUs.

The real change will come when enough HBM can be embedded to serve all system needs. Imagine a chip with 8 or 16 GB of HBM and no external RAM bus. That is the future!

As for cost and availability it is a simple matter of demand driving production. The more hardware using HBM the better for its long term survival in the market. HBM will go the way of the Dodo if demand isnt there, so in a way AMD needs to expand HBM usage to prevent it from becoming a niche and expensive solution.

[tg--]

I'm sorry, but I don't know how else to put it: the FUDzilla "test" is pure BS.
Seriously, who expected that the early, cheap, Raven Ridge APUs would come with HBM? Nobody, that's who.
These APUs don't compete against high-power nVidia discrete mobile GPUs, the compete against Intel's integrated GPUs.

They seriously compare a 15W CPU+GPU part against a (older) high-end 33W dGPU with it's own dedicated memory.
And even though they mock the lack of HBM, they are not even competent enough to mention the RAM configuration they "tested".
The Envy was from the start criticised to repeat the old mistake and ship only single-channel memory, that's very likely what they tested.

If you add the ~25% performance boost common when going to dual-channel, you suddenly have a 15W part beating out a 3 year old 2-chip solution with a 15W CPU part and a 33W GPU part by a comfortable margin.

To me that sounds rather impressive.

To clarify: we're talking about a USD 630 laptop from HPs higher-priced consumer line, not about a USD 1000 gaming machine.