AMD's Next Horizon Event is Live

AlphaAtlas

[H]ard|Gawd
Staff member
Joined
Mar 3, 2018
Messages
1,713
AMD's Next Horizon event is live. AMD is posting slides of the event on slideshare, and on Twitter, while some sites at the event are independently covering it. Currently, AMD's event page doesn't have a live stream, but you can check out the promotional video they just uploaded here.


Today, AMD demonstrates its total commitment to datacenter computing innovation, at AMD’s Next Horizon event in San Francisco, by detailing its upcoming 7nm compute and graphics product portfolio designed to extend the capabilities of the modern datacenter.
 
live stream

Thanks. I saw that, but the live stream wasn't there a few minutes ago.


EDIT: And just so you know, that guy is screen capping Alex on Tech's stream, and doesn't have the whole presentation.
 
Last edited:
nice zen 2 will have hardware fix for spectre.

now they're getting to the good shit.
 
Can't watch this at work :(

Will have to rely on comments here to keep up with what is going on.

Hoping for some good performance news for Zen2 3000 CPU's.
 
it's mostly going to be server related and maybe enterprise vega 7nm.

this guy(forgot his name already) is shitting all over intel without ever directly mentioning them.. friggin hilarious.
 
That's cool. Chiplet designs basically avoid certain microwave circuit design limitations. It was nothing but theory back in the early 2000s (I have some books on the topic), and now you can buy 7nm chiplet compute cores. Smart move.
 
Yeah, I just saw that. This is a Datacenter event. But certain things can be used as predictors for the consumer/enthusiast/pro-sumer parts.
 
It's rare for me to encounter stuff I don't have any clue what it is, but that happened today. R5a, M5a and T3a. Never heard of them before, and google isn't helping much :p
 
Zen 2 news? Any non-datacenter info on this?

nothing yet for non datacenter stuff but i'd expect the chiplet design to be used with zen 2 am4 which could be a massive improvement over zen+ but i think threadripper would see the biggest improvement.
 
I like this chart:

20181106_173414_HDR_575px.jpg


I mean, It has a useless level of detail, but still nice to see that they are optimistic!
 
Zen 2 news? Any non-datacenter info on this?
Not really. Now they're talking about gpu density in the datacenter for remote computation. They've got a new, very small form factor Radeon card. Considering that the best small form factor computers are based on datacenter type components (mini-PCI-E), this is good news for some. ;)
 
I like this chart:

View attachment 117975

I mean, It has a useless level of detail, but still nice to see that they are optimistic!

he actually goes further in the discussion saying that yes it's 7nm but it's basically the same critical design size as intel's 10nm but also says they will still get it to market before their "competition" :p

Rome and new MI60 gpu will use pcie 4.0
 
he actually goes further in the discussion saying that yes it's 7nm but it's basically the same critical design size as intel's 10nm.

I think we expected that. TSMC has always been slightly larger at the same node name compared to Intel. It's a shame everyone can't just use the standard ITRS definitions. Either way, Intel's 10nm has serious problems, but TSMC seems to have their process working, so that's an advantage right there.

Also, notice that the axis on the chart is perf/watt, so it's not simply a matter of charting the node size.
 
"We believe in open source because we don't have the means nor the staff to do it ourselves"....
 
The conversation regarding Infinity fabric and PCIe4 seems pretty interesting.

Connected in a ring, no scaling or switching, 100GB/s per link GPU-GPU or CPU-GPU.

Multidirectional GPU-CPU is 64GB/s.

No mention of latency though.
 
"We believe in open source because we don't have the means nor the staff to do it ourselves"....
That's the sentiment on all fronts right now. "Ain't nobody got time for this" until AI takes over.
 
"We believe in open source because we don't have the means nor the staff to do it ourselves"....

The only problem I have with them touting their Linux support is that I recently built a Raven Ridge Ryzen 5 2400G APU for my fiance.

It's been on the market for 9 months, based on a CPU infrastructure released almost 2 years ago, and a GPU infrastructure launched over a year ago, and it still isn't running properly with current Kernels and MESA drivers.

They really need to get this shit working right, and shortly after launch if they want to claim good open source chops.

IMHO, if you don't have full hardware support in current distributions by 2 months after launch, you are done. PC product cycles may be longer today than they were 20 years ago, but even so, no one wants to buy hardware and have it depreciate in front of their eyes before they can even fully use it.
 
  • Like
Reactions: Jza
like this
No MI60 performance figures.... interesting...

basically just talking about features and simulation stuff right now for machine learning. i'm officially lost on this part of it since the guys hard to hear on the stream.
 
mi60 7nm 7.4 TFlops FP64, 14.7 TFlops FP32, 118 TOPS INT4(?). 1TB/s HBM2 will use PCIE 4.0 and infinity fabric link. will also support virtualization of up to 16 VM's per GPU or 1 VM virtualized across up to 8 GPU's.
 
I like this chart:

View attachment 117975

I mean, It has a useless level of detail, but still nice to see that they are optimistic!


So this would mean that MI60 would be faster than Volta/Turing at similar power comsuption which I don't think is anywhere near possible

or

MI60 is slower but with much less power draw than Volta/Turing. Which while more plausible I don't really think AMD can pull it off either.

And that's not even taking into account nvidia at 7nm.
 
So this would mean that MI60 would be faster than Volta/Turing at similar power comsuption which I don't think is anywhere near possible

or

MI60 is slower but with much less power draw than Volta/Turing. Which while more plausible I don't really think AMD can pull it off either.

And that's not even taking into account nvidia at 7nm.

I read this chart to apply to the CPU side, vs Intel, not the GPU side vs Nvidia, but I could be wrong.
 
  • Like
Reactions: Nobu
like this
confirmed rome will be upto 64 cores 128 threads.

drop in support for current socket same with Milan in 2020.

also thank god for Lisa Su, i was seriously starting to fall asleep until she went back on stage..
 
  • Like
Reactions: Jza
like this
Alex on Tech says he'll be streaming the demo stuff in about 50 minutes or so.. sadly i need to go to sleep so i'll have to watch the vod.
 
Back
Top