Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)

[DisEnchantment]

Speculate at will

 

[A///]

which leak? You post walls of text most days.

 

[Geddagod]

which leak? You post walls of text most days.

I was talking about this trolled 'leek' I posted, it's nothing actually real lol.
But I am genuinely curious about the uOP and dual decoders possibilities on Zen 5.

 

[A///]

I was talking about this trolled 'leek' I posted, it's nothing actually real lol.
But I am genuinely curious about the uOP and dual decoders possibilities on Zen 5.

did you link to the correct post? you're talking about intel laptops.
you won't hear much about zen 5 until 2 months before launch.amd has become a tight ship over time and intel became the town bicycle more than a decade ago with people knowing everything months to a year in advance.

 

[Geddagod]

did you link to the correct post? you're talking about intel laptops.
you won't hear much about zen 5 until 2 months before launch.amd has become a tight ship over time and intel became the town bicycle more than a decade ago with people knowing everything months to a year in advance.

Oop I meant this post

 

[A///]

Oop I meant this post

hmm don't quote me on this but i believe zen 5 cores are a cleansheet design. 15-25% ipc gain isn't going to be good enough. amd learned from their mistakes with zen 2 by launching a new dramatic chipset and processor design as you recall zen+ was still a mono design. am5 launched with a refreshed zen 3 as zen 4 establishing the chipet. I don't expect x770 to come out and if it does I don't see it offering a whole lot over x670, the latter of which is getting rollouts to allow higher capacity dram. the prevailing issue on some x670 mobos are memory related and down to the bios and nothing physically wrong. Whatever amd does needs to be enough to make intel's then arrow a nothingburger or even the future panther lake.

 

[A///]

oops the only upgrade x770 may have is pcie6. it'd be in the window of deployment from the spec finalization by pcisig. i do not expect it to matter to most since pcie5 is largely useless now unless you're running multiple storage devices to use the available bandwidth with prior gen hardware. if not then amd will reuse x670 for zen 5 and its 3d but will go with x770 for a new zen 6 using a new socket and pcie6 and possibly ddr6 if it does release and ramp up fast like ddr5 did. I knew I'd forgotten to mention something earlier.

For amd's next socket beginning with zen 6 because am5 will have had 4 generations of processors and I don't expect them to make it long life like am4 is that they will clean sheet the design learning from their mistakes on x670/am5 including eliminating the ability to use non designed for coolers. IMO amd's biggest sin this gen was allowing ancient coolers to be used with modified mounting kits, they didn't learn from Intel's past mistakes. While ultimately the ihs and associated z height doesn't make too much of a difference in zen 4 it could address the minor issues and going forward with silicon degradation that's normal under stock and oc'd enviros there may possibly be more longevity in a traditional lga ihs height design. That german youtuber bruno or whatever his name is showed some performance benefit by removing the ihs but there would be room to improve from amd's end to the silicon whereas delidding is one facet of it all.

 

[Exist50]

hmm don't quote me on this but i believe zen 5 cores are a cleansheet design. 15-25% ipc gain isn't going to be good enough. amd learned from their mistakes with zen 2 by launching a new dramatic chipset and processor design as you recall zen+ was still a mono design. am5 launched with a refreshed zen 3 as zen 4 establishing the chipet. I don't expect x770 to come out and if it does I don't see it offering a whole lot over x670, the latter of which is getting rollouts to allow higher capacity dram. the prevailing issue on some x670 mobos are memory related and down to the bios and nothing physically wrong. Whatever amd does needs to be enough to make intel's then arrow a nothingburger or even the future panther lake.

Zen 3 was also a "from scratch" architecture though, so 15-25% would be pretty in line. And it's not like we have any reason to expect substantially more than that from Intel in the short term either, so why wouldn't it be "good enough"?

 

[A///]

Zen 3 was also a "from scratch" architecture though, so 15-25% would be pretty in line. And it's not like we have any reason to expect substantially more than that from Intel in the short term either, so why wouldn't it be "good enough"?

im under the firm belief amd didn't expect raptor lake to be as good as it was, and neither did intel. High power use aside it's a really good chip series and anyone who says it isn't apart from high power is lying to themselves. I explained to gedda that Intel has a goal but they're unsure if they'll simply meat that goal or blow past it. Zen 3 was more than just a new core design. If I had to guess here AMD is aiming for at minimum a true 30% ipc increase or higher. It gets even tougher with their plans on implementing various ai and hardware accelerators allegedly from their xilinx acquisition just as Intel plans on implementing that same tech. Microsoft said a few months ago they plan on developing windows to make sure of ai and hardware accelerators on processors. It took apple to do their own thing to wake up the idiots at these companies and realize they're leaving a lot of potential performance on the table.

 

[Geddagod]

Just for fun
Disclaimer: I have not watched the video yet, just posting for the memes

 

[Exist50]

im under the firm belief amd didn't expect raptor lake to be as good as it was, and neither did intel. High power use aside it's a really good chip series and anyone who says it isn't apart from high power is lying to themselves. I explained to gedda that Intel has a goal but they're unsure if they'll simply meat that goal or blow past it. Zen 3 was more than just a new core design. If I had to guess here AMD is aiming for at minimum a true 30% ipc increase or higher. It gets even tougher with their plans on implementing various ai and hardware accelerators allegedly from their xilinx acquisition just as Intel plans on implementing that same tech. Microsoft said a few months ago they plan on developing windows to make sure of ai and hardware accelerators on processors. It took apple to do their own thing to wake up the idiots at these companies and realize they're leaving a lot of potential performance on the table.

AMD and/or Intel might not have expected quite such a gain from Raptor Lake, but I can't say I see the relevance. It's not like they could just decide to add +10% more IPC because of it. And we have very little to go on regarding Lion Cove. Imo, it just boils down to the fact that both AMD's new uarchs and incremental changes have been relatively consistent in their gains, and there's little to indicate that they even can provide +30% in Zen 5's expected timeline.

As for AI, that will definitely be a focus for Microsoft, but again has almost nothing to do with CPU uarch. MS demands everyone add dedicated inference accelerators. For Intel, that's Movidius IP, and for AMD, Xilinx.

 

[A///]

AMD and/or Intel might not have expected quite such a gain from Raptor Lake, but I can't say I see the relevance.

Were you born dumb or became dumb through repeated blows to your head? You didn't consider that AMD couldn't have seen Intel's roadmap and extrapolated performance output based on Intel's generation over generation performance increases including but not limited to slapping on more cores and changing microarchitecture?

. It's not like they could just decide to add +10% more IPC because of it.

You don't say, Sherlock? Did you miss the fact that roadmaps exist and they would have planning accordingly developing Zen 4 several years ago and underestimated Intel's future processor design capability? Are you under some impression that neither company knows of each other's roadmaps before most of the general public does through leaks?

And we have very little to go on regarding Lion Cove. Imo, it just boils down to the fact that both AMD's new uarchs and incremental changes have been relatively consistent in their gains, and there's little to indicate that they even can provide +30% in Zen 5's expected timeline.

There was little to indicate Intel could get their house in order and there was little to indicate Zen 2 and Zen 3 were anything great as they turned out to be. There was also little indication that AMD's Zen 4 could have performed as well as it did despite all the naysers, but we're here in the current and we know they did well. AMD beginning Zen 4's design several years ago had estimated that Intel would have improved but not at the level they had with Raptor minus the power use. Going back a few sentences here there was little to indicate AMD could have pulled off the IPC increases they did with Zen 2 and 3.

As for AI, that will definitely be a focus for Microsoft, but again has almost nothing to do with CPU uarch.

Except there is. Unless M$ is consistent on doing all AI server side then they'll be relying on localised efforts. If the necessary hardware doesn't exist it's slow.

MS demands everyone add dedicated inference accelerators. For Intel, that's Movidius IP, and for AMD, Xilinx.

MS hasn't demanded anything. I'd love to see what source you have unless you pulled that out of your behind. MS incorporating AI in the future relies on hardware support. You either have it or don't. MS isn't in the position to axe more historic processor lines than they did with Windows 11.

I'm beginning to see why some of you are subject to private ridicule on here. I've no children of my own but I didn't think I had to treat a twenty something as a child explaining basic concepts to them on a Thursday evening.






Insulting members is not allowed.


esquared
Anandtech Forum Director

 

[Exist50]

You didn't consider that AMD couldn't have seen Intel's roadmap and extrapolated performance output based on Intel's generation over generation performance increases including but not limited to slapping on more cores and changing microarchitecture?

You still seem to be under the impression that AMD can arbitrarily set IPC targets. Sorry, but it doesn't work like that.

MS hasn't demanded anything.

MS requires you have a dedicated inference accelerator for "Windows Studio" effects.

How to use Windows Studio effects on Windows 11

Windows 11 can enhance your video calling experience with AI-powered Windows Studio effects like background blur. Here's how to use them.

www.xda-developers.com

Ironic to whine about explaining basic concepts when you're grossly unfamiliar with the topic at hand.

Except there is. Unless M$ is consistent on doing all AI server side then they'll be relying on localised efforts. If the necessary hardware doesn't exist it's slow.

In case you weren't aware, none of these inference accelerators are CPUs. And if I have to spell it out even more blatantly, CPU uarch only applies to CPUs...

I've no children of my own but I didn't think I had to treat a twenty something as a child explaining basic concepts to them on a Thursday evening.

You seem more interested in bending over backwards to insult people than discussing anything related to computers.

 

[A///]

You still seem to be under the impression that AMD can arbitrarily set IPC targets. Sorry, but it doesn't work like that.

Again, I realize it's hard to read and why people like Mark need to correct you often because you cannot read, like here. I never said they set an arbitrary number. Please don't put words in my mouth, child. I said they set a goal. They can get close, meet or pass that goal. Does that make more sense or shall I dumb it down further so it can go through your eyes, into your cortex and you can then comprehend what I've said?

MS requires you have a dedicated inference accelerator for "Windows Studio" effects.

And this is how I know you're as clueless as you are. This isn't what I was referring to. What about future Windows do you not understand? Microsoft has goals beyond the next 2-3 years. We're talking about the next 15-20 years of Windows development. Shall I explain how long a year is now?

Ironic to whine about explaining basic concepts when you're grossly unfamiliar with the topic at hand.

Again, no. You cannot read. Is English not your first language? It's understandable if it isn't and I apologize if it isn't. I do understand that some English phrases may be difficult to non native speakers.

You seem more interested in bending over backwards to insult people than discussing anything related to computers.

If I wanted to insult you I'd simply tell you what I thought of you.

 

[Exist50]

Again, I realize it's hard to read and why people like Mark need to correct you often because you cannot read, like here. I never said they set an arbitrary number. Please don't put words in my mouth, child. I said they set a goal. They can get close, meet or pass that goal. Does that make more sense or shall I dumb it down further so it can go through your eyes, into your cortex and you can then comprehend what I've said?


And this is how I know you're as clueless as you are. This isn't what I was referring to. What about future Windows do you not understand? Microsoft has goals beyond the next 2-3 years. We're talking about the next 15-20 years of Windows development. Shall I explain how long a year is now?

Again, no. You cannot read. Is English not your first language? It's understandable if it isn't and I apologize if it isn't. I do understand that some English phrases may be difficult to non native speakers.


If I wanted to insult you I'd simply tell you what I thought of you.

If you're just here to troll, then find another forum. It was clearly my mistake to think you were even the slightest bit interested in computing.

 

[A///]

If you're just here to troll, then find another forum. It was clearly my mistake to think you were even the slightest bit interested in computing.

If anyone is a troll it's you. Don't play victim when you're clueless and try to weasel your way out of everything.

 

[Tarkin77]

Just for fun
Disclaimer: I have not watched the video yet, just posting for the memes

this video is a complete waste of time

he pulled the 15-26% IPC and 2-9% clock increase out of his behind

thats like saying ST performance is 17-37% higher. No shit, Sherlock!

"my sources say" ... yeah, sure, whatever.

 

[bakyt115]

this video is a complete waste of time

he pulled the 15-26% IPC and 2-9% clock increase out of his behind

thats like saying ST performance is 17-37% higher. No shit, Sherlock!

"my sources say" ... yeah, sure, whatever.

If you fan of AMD and your are feeling bad, catch cold or something else.
There are good therapy. Your can watch MLIDs new video about zen5 and your hormones of happiness breaks new high.

 

[Exist50]

If you fan of AMD and your are feeling bad, catch cold or something else.
There are good therapy. Your can watch MLIDs new video about zen5 and your hormones of happiness breaks new high.

Honestly, I'm surprised his IPC numbers are as reasonable as they are, given the insane hype around Zen 5 coupled with his particular history. Of course, the range is still so wide he's clearly BSing as usual, but no sense stating the obvious.

That said, I don't think anything is enough to slow down the hype train at this point. Need to just let it run its usual course.

 

[DisEnchantment]

Please post clickbait rumors elsewhere including the mudslinging.
Lots of folks really hate it to see an update on the thread only to find something very remote from CPU Architecture or related at all to the thread.
This thread can be locked for the next 6 months if it helps.

 

[esquared]

All right everyone.

Stop with the insults or this thread gets locked.


esquared
Anandtech Forum Director

 

[A///]

What's sad about mlid apart from being a dud is he has potential to be a legitimate tech avenue but instead follows the guise of other similar channels like gamer's meld, another useless unit of bandwidth forever rotting the internet. Hype, money and infamy reign supreme on these channels. Imagine being well to do engineer at GM and then turning to this rubbish for life's fulfillment. What an odd fellow.

He's got being a decent orator and a chin like jay leno going for him at least.

 

[soresu]

I don't expect x770 to come out and if it does I don't see it offering a whole lot over x670

It may simply offer a lower power chipset for the same IO.

Which is nothing to sniff at if you are already upgrading from AM4.

 

[soresu]

High power use aside it's a really good chip series and anyone who says it isn't apart from high power is lying to themselves

The server / workstation / datacenter market is no small thing, and those guys are pretty power / thermals conscious.

 

[A///]

The server / workstation / datacenter market is no small thing, and those guys are pretty power / thermals conscious.

Of course, but what does that have to do with raptor lake for desktop which is what I was talking about?

 

[soresu]

Of course, but what does that have to do with raptor lake for desktop which is what I was talking about?

Ah fair, I'm one of those pesky power (mostly sound) conscious desktop users though.

The current woes over energy bills may cause some formerly unbothered desktop/enthusiast users to rethink power in their considerations too.