News AMD's Press Conference - my SFF perspective

I'll be live tweeting the AMD CES 2019 stream on SFF Workshop's Twitter - https://twitter.com/SffWorkshop starting at 9AM Las Vegas time. I'm not in Las Vegas with the SFFn team, but I'm watching from the comfort of my own home!

SFF Workshop is my new project, as I move away from content production for SFFn - see my signature for more information. A full announcement coming in the next week.

Also, I'll be in my Discord channel throughout the week discussing the event - feel free to join

 

[confusis]

5 minutes! - https://twitter.com/SffWorkshop

 

[Legion]

 

[confusis]

TLDR;

EPYC is epic, still - 64 cores, 128 threads per socket, Zen 2 @ 7nm
AMD Radeon VII (7.. for 7nm) is a 2080 equivalent for $100 cheaper, but still plays 2nd fiddle to the 2080ti
Ryzen 2 performs similarly to the 9900K in cinebench at 30% lower power consumption
Usual partner asskissing
AMD in Chromebooks now
AMD pushing ultralight notebooks that can game

 

[NateDawg72]

I'm very very curious on what Ryzen 3 AM4 performance/latency is going to be like with that single 8 core chiplet and IO die. Eagerly looking forward to seeing benchmarks on that

 

[Legion]

I'm very very curious on what Ryzen 3 AM4 performance/latency is going to be like with that single 8 core chiplet and IO die. Eagerly looking forward to seeing benchmarks on that

#

That's what I'm thinking as well and I hope they improve the IMC considerably over what 2xxx is capable of for faster DDR4 speeds!!

 

[Boil]

AMD Radeon VII (7.. for 7nm) is a 2080 equivalent for $100 cheaper, but still plays 2nd fiddle to the 2080ti
Ryzen 2 performs similarly to the 9900K in cinebench at 30% lower power consumption

Not the ultimate high-end AMD GPU that eats 2080Tis for breakfast, but also cheaper; one could probably get the rest of a full system for the price difference...

I'm very very curious on what Ryzen 3 AM4 performance/latency is going to be like with that single 8 core chiplet and IO die. Eagerly looking forward to seeing benchmarks on that

I am curious about that bit of empty space there below the Zen 2 chiplet...

Obvious space for a Navi GPU chiplet when they push out the Zen 2 based APUs; but might it also be reserved for a second Zen 2 CPU chiplet, thereby giving us up to 16 core / 32 threads, like the rumors...?

 

[Michael]

Even though the MSRP is lower than the 2080 - 2080s have already dipped below the $699 price point several times. Also no new APU announcment - so no follow up to the 2400G? I'm disappointed

 

[GuilleAcoustic]

Also no new APU announcment - so no follow up to the 2400G? I'm disappointed

The leaked informations about Ryzen 3000 APU say 3rd quarter of 2019. As Lisa said, more will come later when closer to production. Lets wait...

 

[NateDawg72]

APU's will be later as usual

We haven't yet had an APU that came out with a new architecture, for either CPU or GPU portion. They all lagged behind their respect CPU/GPU architecture releases

(I'm not counting Kaveri/Carizzo since they used Bulldozer iterations that never got used elsewhere)

 

[Michael]

APU's will be later as usual

We haven't yet had an APU that came out with a new architecture, for either CPU or GPU portion. They all lagged behind their respect CPU/GPU architecture releases

(I'm not counting Kaveri/Carizzo since they used Bulldozer iterations that never got used elsewhere)

Wasn't Raven Ridge announced January 8th last year?

 

[NateDawg72]

Wasn't Raven Ridge announced January 8th last year?

I'm not sure, it might have been. I was looking at release dates

(desktops)
First Ryzen part: March 2017
First Raven Ridge part: February 2018

I imagine zen 2 APUs won't lag as much as first gen but it gives some perspective on it

 

[parlinone]

it looks like these Zen2 IO dies will be real power hogs. The threadripper IO uses about 25-30W around idle and 45W from 16-32 threads. As the ES CB15 demo displayed around 20W higher system idle compared to the 9900K (67W vs 46W) it looks like the IO die of Ryzen3 won't perform much better. So while the 8core chiplet seems to have outstanding performance/power, the IO die by design (IF) not so much…

Another disappointment is that the empty space will only be used for a second CPU chiplet, so no Navi GPU.

Looks like my next CPU will be either a Sunny Cove or Renoir, but since early Geekbench results have Sunny Cove around 20% higher IPC I dont think Renoir will beat that.

 

[chx]

Except, of course, Sunny Cove is vaporware. We won't see any Intel 10nm products, only 7nm

 

[Therandomness]

it looks like these Zen2 IO dies will be real power hogs. The threadripper IO uses about 25-30W around idle and 45W from 16-32 threads. As the ES CB15 demo displayed around 20W higher system idle compared to the 9900K (67W vs 46W) it looks like the IO die of Ryzen3 won't perform much better. So while the 8core chiplet seems to have outstanding performance/power, the IO die by design (IF) not so much…

Another disappointment is that the empty space will only be used for a second CPU chiplet, so no Navi GPU.

Looks like my next CPU will be either a Sunny Cove or Renoir, but since early Geekbench results have Sunny Cove around 20% higher IPC I dont think Renoir will beat that.

There will be a different PCB used for the Navi APUs coming later in the year, or so I've heard. And overall, the supposed R5 3600X system was drawing less power than the 9900k system during the run, don't forget that, and when the 9900k was allowed to run stock without any limitations.

 

[NateDawg72]

Except, of course, Sunny Cove is vaporware. We won't see any Intel 10nm products, only 7nm

How far away do you think Intel 7nm is? I can't imagine it is soon.

 

[chx]

https://en.wikichip.org/wiki/7_nm_lithography_process#Intel

On February 8 2017 Intel announced a $7B investment in Arizona's Fab 42 which will eventually produce chips on a 7 nm process. In May of 2017 Mark Bohr, Intel's Senior Fellow and Director of Process Architecture and Integration, confirmed that Intel's 7 nm node has entered development phase and that the company's research focuses on the 5 nm and 3 nm nodes. Details of their 7 nm node have not been disclosed yet. CEO Brian Krzanich mentioned a 2020 timeframe in an investor conference in June.

So yeah, next year. I could see that happening instead of this botched 10nm.

 

[Phuncz]

it looks like these Zen2 IO dies will be real power hogs. The threadripper IO uses about 25-30W around idle and 45W from 16-32 threads. As the ES CB15 demo displayed around 20W higher system idle compared to the 9900K (67W vs 46W) it looks like the IO die of Ryzen3 won't perform much better. So while the 8core chiplet seems to have outstanding performance/power, the IO die by design (IF) not so much…

Another disappointment is that the empty space will only be used for a second CPU chiplet, so no Navi GPU.

AMD's footnotes say:

AMD engineering sample silicon, AMD reference motherboard
vs
Intel i9-9900K, Gigabyte Z390 Aorus​

I would think an engineering sample CPU isn't optimized for idle power consumption. I'm not expecting it to be massively better than Zen 1 either but to me it seems realistic to expect lower idle power consumption. I also wouldn't expect a chiplet to be swappable with an iGPU that easily, also it leverage them almost no advantage cost-wise versus just doing a proper APU later.

 

[parlinone]

Except, of course, Sunny Cove is vaporware. We won't see any Intel 10nm products, only 7nm

Sunny Cove is already spotted in the wild as Ice Lake ES, with strong statements it will launch this year in high volume, and has been demoed. It's no less vaporware than Zen2 at this point. Nice try at trolling though.

 

[NateDawg72]

2020 for Intel 7nm sounds extremely optimistic. Maybe for a product similar to last years 10nm dual core parts at the end of 2020, in low volume, for a specific market...