Speculation: Ryzen 4000 series/Zen 3

[mopardude87]

I just HAD a feeling i had to go x570, cause i do wanna try some 4000 series action so yup i did ok this time. I nearly bought a B450 over x570 with hopes of 4000 support.

 

[eek2121]

Got confused with the MEMCLK:FCLK.

1:1 is actually 2:1 (2:1 statement is referring to 1600 MHz FCLK being tied to 3200 MHz MCLK)
2:1 is actually 4:1.

DDR4 3200 runs at 1600MHz FYI.

 

[Valantar]

DDR4 3200 runs at 1600MHz FYI.

Yeah, that's a distinction that a lot of people forget, that DDR speeds are MT/s (megatransfers/second) which is 2x the clock speed thanks to the double part of double data rate RAM.

 

[NostaSeronx]

DDR4 3200 runs at 1600MHz FYI.

Actually, it runs at 400 MHz, FYI. The I/O bus runs at 1600 MHz. With the transfer rate being 3200 MHz.

For purposes of simplification only the last number is used.

So, if AMD's fabric is 1600 MHz and it transfers fabric data once, while the mem I/O is 1600 MHz and it transfers mem data twice. Is that not 2:1?

Overall doesn't matter in the long run asynchronous data movement is more efficient and faster than synchronous data movement.
=> Specfically, if it is MCLK=3200 MHz and FCLK=2400 MHz or such. Renoir's decoupling is the step in the right direction. Plus, future memory technologies implementing DVFS for MCLK/Mvoltage. (Ex: Samsung's upcoming LPDDR5-6400 DRAM supports 1.05V@6.4GHz and 0.9V@3.2GHz+(Lower speed and lower voltage when full speed isn't needed))

 

[Valantar]

Actually, it runs at 400 MHz, FYI. The I/O bus runs at 1600 MHz. With the transfer rate being 3200 MHz.

For purposes of simplification only the last number is used.

So, if AMD's fabric is 1600 MHz and it transfers fabric data once, while the mem I/O is 1600 MHz and it transfers mem data twice. Is that not 2:1?

That definitely sounds like 1:1 to me - same clock speed, whether one bus is double pumped and the other isn't doesn't affect this.

Overall doesn't matter in the long run asynchronous data movement is more efficient and faster than synchronous data movement.
=> Specfically, if it is MCLK=3200 MHz and FCLK=2400 MHz or such. Renoir's decoupling is the step in the right direction. Plus, future memory technologies implementing DVFS for MCLK/Mvoltage. (Ex: Samsung's upcoming LPDDR5-6400 DRAM supports 1.05V@6.4GHz and 0.9V@3.2GHz+(Lower speed and lower voltage when full speed isn't needed))

AFAIK all LPDDR standards do things like this, including current LPDDR4X. That is a big part of how they are so much more power efficient than regular old DDR standards. Though it might be that upcoming standards have more steps for a more dynamic clock regulation scheme.

 

[DrMrLordX]

Anyone here own a Renoir laptop yet? And does Ryzen Master work on Renoir?

 

[tamz_msc]

Anyone here own a Renoir laptop yet? And does Ryzen Master work on Renoir?

I'm pretty sure that Ryzen Master doesn't work on Renoir, or Picasso and Raven Ridge for that matter.

 

[DrMrLordX]

I'm pretty sure that Ryzen Master doesn't work on Renoir, or Picasso and Raven Ridge for that matter.

I think it only works in the AM4 configurations, now that I've read up on it.

 

[Panino Manino]

With this upgrade situation, I'm thinking, there'll be no "Zen 2+"?
I expected that people would still be able to buy a new gen of Zen 2 and APU with Navi, or this will not happen after all?

 

[uzzi38]

With this upgrade situation, I'm thinking, there'll be no "Zen 2+"?
I expected that people would still be able to buy a new gen of Zen 2 and APU with Navi, or this will not happen after all?

No Zen 2+ for most of us. Straight to Zen 3 we go.

Any Zen2+RDNA2 uArch chips will likely be very specific to very few customers, and definitely not be for desktop.

 

[eek2121]

Zen 2+ is not likely to happen, though even if it did, it wouldn’t be a bad thing.

I expect AMD is about to throw it’s weight behind Zen 3. I am willing to bet we will get a Zen 3+. AMD likely won’t label it as such, of course, but I don’t see them rushing consumers to a 5nm node. You will note that AMD’s consumer roadmap has a gap in it.

 

[soresu]

Zen 2+ is not likely to happen, though even if it did, it wouldn’t be a bad thing.

Yeah, as an SKU refresh at best, likely near to Zen3/Vermeer release in order to clear stock.

 

[lightmanek]

Anyone here own a Renoir laptop yet? And does Ryzen Master work on Renoir?

No, it doesn't, but there are nice little P-State tools you can kill your laptop with while overclocking
I know as I've killed my ASUS TUF FX505 with Ryzen 3750H in Cinebench when I pushed power limits over 65W sustained For 80% of my R20 run I had all core clock of 3.9-4.0GHz, but then black screen happened ...

Still, it worked great for any reasonable setting which VRM could still sustain and cooling was keeping up well enough with fan in jet engine mode.

I have my replacement laptop back and it runs great, but I'm no longer pushing it that hard. Renoir should be better in that regard as I expect much beefier VRM's on H parts due to longer and higher boost settings.

When I can, I will upgrade my ASUS to Renoir based model.

 

[darkswordsman17]

Yeah there's about zero chance of Zen 2+. I'd actually be a little curious if there would be any point at all to keep producing Zen 2 chiplets, or if they could just put Zen 3 chiplet paired with the Zen 2 IOD. Perhaps the performance improvement would be enough that the 3950X might move to being 12 core, and the 8 core chips being 6 cores. Maybe they'd lose a bit of multi-threaded performance but it'd be made up for by the extra IPC (so they lose say 20% from 25% fewer cores, but with the 15% IPC its really only losing 5% in highly multi-threaded, while it'd maybe gain in lower threaded and other tasks like maybe gaming; or maybe ones where the full cache isn't working or something). But this way, they make use of the defective Zen 3 chips, while keeping the fully enabled ones as a premium. And it'd let them clear out IOD if they need to do that.

We'll see about Zen 3+, I think it makes some sense as I have a hunch that Zen 4 won't be til 2022, so that'd leave 2021 without new product. Which going chiplets means AMD can update just part of the overall CPU package. Seems that the rumor about Zen 3+ is it'll bring probably a new I/O that could offer DDR5 support. Which it might could also be that the big thing is APUs go chiplet, which would let them keep the AM4 socket and boards (whereas going DDR5 and/or PCIe 5 will likely require some updates to the boards).

 

[CHADBOGA]

Yeah there's about zero chance of Zen 2+. I'd actually be a little curious if there would be any point at all to keep producing Zen 2 chiplets, or if they could just put Zen 3 chiplet paired with the Zen 2 IOD. Perhaps the performance improvement would be enough that the 3950X might move to being 12 core, and the 8 core chips being 6 cores. Maybe they'd lose a bit of multi-threaded performance but it'd be made up for by the extra IPC (so they lose say 20% from 25% fewer cores, but with the 15% IPC its really only losing 5% in highly multi-threaded, while it'd maybe gain in lower threaded and other tasks like maybe gaming; or maybe ones where the full cache isn't working or something). But this way, they make use of the defective Zen 3 chips, while keeping the fully enabled ones as a premium. And it'd let them clear out IOD if they need to do that.

We'll see about Zen 3+, I think it makes some sense as I have a hunch that Zen 4 won't be til 2022, so that'd leave 2021 without new product. Which going chiplets means AMD can update just part of the overall CPU package. Seems that the rumor about Zen 3+ is it'll bring probably a new I/O that could offer DDR5 support. Which it might could also be that the big thing is APUs go chiplet, which would let them keep the AM4 socket and boards (whereas going DDR5 and/or PCIe 5 will likely require some updates to the boards).

For what it's worth, the YouTuber, Moore's Law is Dead, thinks that there will be a minor bump in Zen 2 speeds in July, with something like a 3620 and/or 3720 products released.

He also seems to think that Zen 3+ is a certainty.

 

[mopardude87]

Ryzen Master is a god of a application. These profiles still to good to be true almost. All that information and stuff dayum. Like i said some time ago AMD got more game then EA.

 

[eek2121]

For what it's worth, the YouTuber, Moore's Law is Dead, thinks that there will be a minor bump in Zen 2 speeds in July, with something like a 3620 and/or 3720 products released.

He also seems to think that Zen 3+ is a certainty.

I can tell you with absolute certainty that he is wrong.

 

[CHADBOGA]

I can tell you with absolute certainty that he is wrong.

About both things?

 

[Ajay]

I can tell you with absolute certainty that he is wrong.

So, yeah, this guy usually just throws a bunch of redacted up against the wall just to see what sticks.
But, how are you 100% sure that there will be no Zen3+? Got a link, proof, other info?

You should know better than to hide profanity with asterisks in the tech sub-forums.
We have a zero tolerance policy for profanity in the tech sub-forums.

Iron Woode
Super moderator

 

[Valantar]

Zen + was on AMD roadmaps from day 1. There has never at any point in time been a Zen 2+ on any AMD roadmap. As such, assuming there will be no such thing is very safe, especially now as we're around the (assumed) midpoint between generations. Zen + was likely a stopgap for AMD to both get into a rhythm of (more or less) annual updates while also allowing them some time to see how Zen worked in the real world and tweak it accordingly. Once they got into this rhythm with Zen 2 there's no reason to go back to minor "+" updates unless the next arch refresh becomes extremely difficult to realize or something else (like waiting for a node) is holding them back.

As for Moore's law is dead, that channel is utter garbage. I wouldn't put any faith into anything said there whatsoever. Period.

 

[DrMrLordX]

No, it doesn't, but there are nice little P-State tools you can kill your laptop with while overclocking

I was more interested in knowing more about switching Renoir between synched and desynched IF.

 

[lightmanek]

I was more interested in knowing more about switching Renoir between synched and desynched IF.

This will be down to laptop BIOS I'm afraid, without manufacturer support we can only hope for BIOS modding community help. I know that Chinese early review showed you could change memory speeds on the Renoir laptop, so there is a hope some models will allow it.

 

[Valantar]

I was more interested in knowing more about switching Renoir between synched and desynched IF.

As I said above, AMD's comment on that subject is that FClock is decoupled from MClock in Renoir. Talking about it in terms of ratios or being in sync is thus meaningless. You should as such be able to set FClock to whatever you want as long as the BIOS and/or any software exposes the option.

 

[amrnuke]

Re: Zen 3+, I do think it would be very reasonable for AMD to work on this IF they are doing Zen3 on N7P. I do not have the slightest clue if they are, but here's my rationale:

1) porting N7P to N6 w/o retooling (as I understand it) for fairly easy 5-10% improvements
2) extend the shelf life of B550/X570 and give further upgrade path
3) a nice gesture of good will, as they eschew AM4 and move on full speed ahead
4) will give them more time to work out kinks on Zen4 / AM5 (almost 2 year cycle between Zen3 release and a Zen4 release could be created - if they need it)

 

[Thunder 57]

Re: Zen 3+, I do think it would be very reasonable for AMD to work on this IF they are doing Zen3 on N7P. I do not have the slightest clue if they are, but here's my rationale:

1) porting N7P to N6 w/o retooling (as I understand it) for fairly easy 5-10% improvements
2) extend the shelf life of B550/X570 and give further upgrade path
3) a nice gesture of good will, as they eschew AM4 and move on full speed ahead
4) will give them more time to work out kinks on Zen4 / AM5 (almost 2 year cycle between Zen3 release and a Zen4 release could be created - if they need it)

It was my understanding that N6 was meant as a cheap-ish way for clients on N7 to get an an upgrade path, and that N7P was superior to N6.

TSMC: N7+ EUV Process Technology in High Volume, 6nm (N6) Coming Soon

www.anandtech.com

By contrast, N7+ uses different design rules, but also provides more benefits than N6 when compared to N7.

Of course, things may have changed.