Speculation: Ryzen 4000 series/Zen 3

[DisEnchantment]

Zen Eng Sample in July/Aug 2016 --> Launch March 2017 (8 months)
Zen2 Eng Sample in Aug/Sep 2018 --> Launch July 2019 (8 months)

Zen3 taped out last year Q2. And in the presentation for the Archer2 award in Oct 2019 an AMD rep said they were already sampling Milan to the Hyperscalers.
Also from last Oct, Milan was always planned for Q3 and probably Ryzen in Q2, but things changed a bit. So we could expect Ryzen for Q3 which is barely 4-5 months away.
Zen3 is really a blackhole this time, It is either very very good or it is not being tested out there in the wild

For a 'later this year' launch as mentioned in the FAD there is really a dearth of ENG samples compared to past history.
But at least this time, APISAK will have more company with rogame and uzzi

 

[moinmoin]

I'm not entirely following that logic. The previous/current gen had 8 core CPUs, yet we got dual core Zen chips. I'd like to be wrong, but I'd guess we'll still see even quad core chips.

The dual core Zen chips are the lowest level APUs. The question is not so much whether AMD will offer the full range of cores amounts, they clearly will, but what part will be occupied by the Zen 3 Vermeer CPUs on desktop. Since the Zen 2 APUs Renoir will go up to 8c/16t those could cover that, with Vermeer going to 12c and 16c chips. So the question is to what extend AMD intends to allow Renoir and Vermeer to overlap on desktop and compete against each other in the DIY market.

 

[uzzi38]

Zen Eng Sample in July/Aug 2016 --> Launch March 2017 (8 months)
Zen2 Eng Sample in Aug/Sep 2018 --> Launch July 2019 (8 months)

Zen3 taped out last year Q2. And in the presentation for the Archer2 award in Oct 2019 an AMD rep said they were already sampling Milan to the Hyperscalers.
Also from last Oct, Milan was always planned for Q3 and probably Ryzen in Q2, but things changed a bit. So we could expect Ryzen for Q3 which is barely 4-5 months away.
Zen3 is really a blackhole this time, It is either very very good or it is not being tested out there in the wild

For a 'later this year' launch as mentioned in the FAD there is really a dearth of ENG samples compared to past history.
But at least this time, APISAK will have more company with rogame and uzzi

Oh, I don't find the benchmarks... that got a little tiring pre-Zen 2, and all I used to do was search GB and UB haha. _rogame does all the benchmark finding, and I don't want to take any credit for that stuff. All goes to him haha.

I follow the rumour mill a little tighter and and I help him dig through any more technical stuff, Github leaks etc.
Speaking of which, the 30th March embargo date for Renoir? Yeah, that's probably legit, heard it last week after the person that told me got my hopes up as they thought it that very day they told me.

Tomorrow's the day. We're just hours away from seeing stupid level power efficiency in a CPU.
And I can't wait for it.

 

[exquisitechar]

https://www.tomshardware.com/reviews/asus-rog-zephyrus-g14

That battery life and performance...Renoir is great.

 

[tamz_msc]

https://www.tomshardware.com/reviews/asus-rog-zephyrus-g14

That battery life and performance...Renoir is great.

That's strange, Notebookcheck got very high idle power consumption(30W) on their review sample. According to Compterbase, Asus says that lower than expected performance in silent mode is a bug. And this YouTube review shows that the battery life is lower than the previous gen 3750H+1660 Ti Max-Q with the same battery size.

Right now, I would take all battery life figures with a big grain of salt.

 

[RetroZombie]

That's strange, Notebookcheck got very high idle power consumption(30W) on their review sample

One cpu that can have full load TDP of only 15W gets 30W idle? That's strange.

 

[moinmoin]

One cpu that can have full load TDP of only 15W gets 30W idle? That's strange.

4900HS is a 35 TDP chip.
Still the Notebookcheck report as well as that video seem to be the odd ones.

 

[RetroZombie]

4900HS is a 35 TDP chip.

He said idle and the chip is the same as the 4800u which does 15W full load, will it do 30W idle?

 

[moinmoin]

He said idle and the chip is the same as the 4800u which does 15W full load, will it do 30W idle?

No, U chips are 15W, H chips 45W and HS chips (tested here) 35W. Other than that it's very clear by now that something about Notebookcheck's test and/or setup is broken.

 

[misuspita]

Wow, that is impressive. The diffeence to last gen mobile is like Zen1 vs... whatever they had mobile before

 

[Hans Gruber]

Does anybody here think including an i-gpu in Zen3 would be good for the 4600 and 4700X. Intel isn't going away. AMD needs to add value to their CPU's. I am assuming that Zen3 will finally top Intel in gaming performance.

 

[inf64]

Does anybody here think including an i-gpu in Zen3 would be good for the 4600 and 4700X. Intel isn't going away. AMD needs to add value to their CPU's. I am assuming that Zen3 will finally top Intel in gaming performance.

Zen3 APU is coming, but I doubt they will launch it with regular Zen3 desktop chips. Renoir on desktop will do just fine as it's miles ahead of intel's 2020 offerings.

 

[HurleyBird]

I don't think any kind of beefy APU is necessary, but some kind of minimal display adaptor would be great.

There are two main buckets who will be buying these processors: those that need some actual GPU horsepower, and those that just need to drive displays. The rest of the market gets covered by a Renoir analog.

That said, I think it's a limitation of AM4 that you need to lose some PCIe lanes when you utilise a display adaptor, so it's not something I would expect for Zen 3.

Also, eventually packaging tech should improve to the point where AMD won't make a different mobile focused monolithic die anymore.

 

[moinmoin]

Does anybody here think including an i-gpu in Zen3 would be good for the 4600 and 4700X. Intel isn't going away. AMD needs to add value to their CPU's. I am assuming that Zen3 will finally top Intel in gaming performance.

If AMD keeps the model numbering scheme 4600 and 4700 (possibly also 4800) will be available as APUs since those go up to 8c.

Actually it's funny how AMD unnecessarily backed itself into a corner with those model numbers, Threadripper is already restricted to numbers above x950, now normal Ryzen CPUs are going to be restricted by the rising amount of cores in APUs. Time to radically revamp the model numbering scheme.

 

[soresu]

If AMD keeps the model numbering scheme 4600 and 4700 (possibly also 4800) will be available as APUs since those go up to 8c.

Actually it's funny how AMD unnecessarily backed itself into a corner with those model numbers, Threadripper is already restricted to numbers above x950, now normal Ryzen CPUs are going to be restricted by the rising amount of cores in APUs. Time to radically revamp the model numbering scheme.

The letters add a different dimension to the SKU naming schemes - U,G,H, and HS are all mobile/desktop APU SKU's whatever the number preceding them.

 

[moinmoin]

The letters add a different dimension to the SKU naming schemes - U,G,H, and HS are all mobile/desktop APU SKU's whatever the number preceding them.

Mobile has its own independent numbering scheme. On desktop there's only G for APUs, and in the last two gens AMD avoided pitching APUs agains CPUs in the desktop DIY market.

 

[DrMrLordX]

The diffeence to last gen mobile is like Zen1 vs... whatever they had mobile before

Carrizo.

 

[soresu]

Carrizo.

The fact that they actually had a partner make a VR headset with that makes me wonder what they could do with Renoir or Cezanne on the way.

 

[DrMrLordX]

The fact that they actually had a partner make a VR headset with that makes me wonder what they could do with Renoir or Cezanne on the way.

A heck of a lot more.

 

[Tup3x]

The letters add a different dimension to the SKU naming schemes - U,G,H, and HS are all mobile/desktop APU SKU's whatever the number preceding them.

They managed to make same naming mistake as Intel. Random guy doesn't have any idea what those letters mean. Higher number isn't necessarily better. Larger number yet same generation. I don't really care though.

 

[soresu]

They managed to make same naming mistake as Intel. Random guy doesn't have any idea what those letters mean. Higher number isn't necessarily better. Larger number yet same generation. I don't really care though.

Random guy walks into a PC store or an electronics store like Currys/PC World and picks something off the shelf, often something the salesperson pushed at him - my Nan is random guy squared.

 

[amrnuke]

They managed to make same naming mistake as Intel. Random guy doesn't have any idea what those letters mean. Higher number isn't necessarily better. Larger number yet same generation. I don't really care though.

Really frustrating. Obfuscating your market doesn't make any sense unless you're trying to push units that cost you less to make, at a higher price, by naming them similarly to better products.

Heck, even the salesperson half the time has no clue what the differences are.

 

[Saylick]

So AdoredTV claims to have new info on Zen 3 and even a bit on Zen 4.

I know there's some people here who don't trust his sources (as they should) but I figured I drop the summarized version here:

**Zen 3**
Late 2020 launch, or possibly Q1 2021
SMT2
10-15% ST IPC gain
Shared 32 MB L3 across 8 core CCX
47 cycle latency for the shared L3 (vs 40 cycles for the 16 MB L3 in Zen 2)
New ISA and better security
Power efficiency improvements
Higher boost clocks

**Milan**
A0 sampling with no SMT
B0 silicon delayed until September

**Zen 4**
5nm derivative of Zen 3
DDR5
1 MB L2 (2x vs Zen 2 and Zen 3)
AVX512 support
2021 launch, but possibly late

 

[moinmoin]

So AdoredTV claims to have new info on Zen 3 and even a bit on Zen 4.

I know there's some people here who don't trust his sources (as they should) but I figured I drop the summarized version here:

**Zen 3**
Late 2020 launch, or possibly Q1 2021
SMT2
10-15% ST IPC gain
Shared 32 MB L3 across 8 core CCX
47 cycle latency for the shared L3 (vs 40 cycles for the 16 MB L3 in Zen 2)
New ISA and better security
Power efficiency improvements
Higher boost clocks

**Milan**
A0 sampling with no SMT
B0 silicon delayed until September

**Zen 4**
5nm derivative of Zen 3
DDR5
1 MB L2 (2x vs Zen 2 and Zen 3)
AVX512 support
2021 launch, but possibly late

That's... pretty tame.

 

[tamz_msc]

That's... pretty tame.

Yeah, I still expect 20%+ IPC for Zen 3 over Zen 2 based on what Forrest Norrod said. 10-15% would be a major disappointment, at least for me. The rest of the rumor seems plausible, except for the 5 month gap between A0 and B0. I don't suppose it should take 5 months to fix SMT.