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

[DisEnchantment]

Speculate at will

 

[Goop_reformed]

Did you see some leaks or did you just write random nonsense?

Well there are 2 camps. 15%-20% gang and the sky high ipc gang. I've heard some from the 1st but have I'm yet to hear from the second

 

[Glo.]

LPCAMM2 allows for 256 bit bus with just two memory modules.

 

[adroc_thurston]

LPCAMM2 allows for 256 bit bus with just two memory modules.

Wires and shoreline costs are no bueno.

 

[Glo.]

Wires and shoreline costs are no bueno.

Well then, build a port on that shoreline to help with loading and unloading.

 

[adroc_thurston]

Well then, build a port on that shoreline to help with loading and unloading.

OMI is dead dawg

 

[Rigg]

Switching sockets for Zen 6 would go over about as well with consumers/reviewers as a 43% MSRP increase for 9950X. If AMD want's a sure fire way to burn up the good will they've built up since Zen launched stuff like this is the formula.

 

[adroc_thurston]

If AMD want's a sure fire way to burn up the good will they've built up since Zen launched stuff like this is the formula.

They really don't need that anymore.

 

[igor_kavinski]

43% MSRP increase for 9950X.

A percent increase in price for every percent increase in performance

 

[Tigerick]

They really don't need that anymore.

So adroc, do you still think Zen5 16-core will provide 40% IPC improvement with $999 price tag ?

 

[adroc_thurston]

So adroc, do you still think Zen5 16-core will provide 40% IPC improvement with $999 price tag ?

I don't need to think.

 

[Tigerick]

I don't need to think.

Indeed

 

[SpudLobby]

something something turin-dense

That’s just core power though and for servers. The QC figure is the SoC + DRAM + power delivery/VRMs, the platform, and it’s pretty impressive how much performance they can eek out at 2W with that while AMD and Intel need 5-6W.

Maybe Strix will be nearly X Elite-caliber on ST perf/W curves but I’m not hopeful.

 

[adroc_thurston]

That’s just core power though and for servers

Did you really forget which other products have Dense piles?

The QC figure is the SoC + DRAM + power delivery/VRMs, the platform

Don't have to tell me that, Andrei is annoying as is with that metric.

and it’s pretty impressive how much performance they can eek out at 2W with that while AMD and Intel need 5-6W.

It's cute but franky worthless given you need moar perf for foreground tasks than 2w allows and for background this is miles and eons above what Apple low power ghetto does.

Maybe Strix will be nearly X Elite-caliber on ST perf/W curves but I’m not hopeful.

Didn't ask your opinion.

 

[Rigg]

A percent increase in price for every percent increase in performance

Nvidia math!

 

[SpudLobby]

Did you really forget which other products have Dense piles?

Ok, but marred by AMD fabric? Zen 5 with Strix sounds great on paper. Likely to be a big improvement.

Don't have to tell me that, Andrei is annoying as is with that metric.

Well he’s right.

It's cute but franky worthless given you need moar perf for foreground tasks when 2w power allows and for background this is miles and eons above what Apple low power ghetto does.

Agreed except the marginal cost of another core at the same performance in that case isn’t going to be another full 2W. Anyway, they’ll scale down much better than other P cores (or likely AMD/Intel e cores). Yeah long term they’ll need real e cores like Apple’s if they want to stay competitive.

Didn't ask your opinion.

Expected better from you, not that creative. Yours are certainly ubiquitous, cheap, hostile, and ofc often unsolicited.

 

[adroc_thurston]

but marred by AMD fabric?

s'fine.

Well he’s right.

It's a gameable metric by the ways of power delivery spam, which is something completely alien to PC lands.

Anyway, they’ll scale down much better than other P cores (or likely AMD/Intel e cores).

Again, those don't go low enough to make a dent versus Apple in the background and dear god foreground power isn't anything too spectacular.
15W cinememe machine materiel which makes sense given the server core background. But again, Z5dense exists. has SMT even, undisputed cinememe@mm^2 king.

Expected better from you, not that creative. Yours are certainly ubiquitous, cheap, hostile, and ofc often unsolicited.

Don't need to be upset.
move the goalies to Hamoa-next please, might have better chances in the 2026 market (or at least I hope so).

 

[SpudLobby]

s'fine.

I hope!

It's a gameable metric by the ways of power delivery spam, which is something completely alien to PC lands.

It’s awesome.

Again, those don't go low enough to make a dent versus Apple in the background and dear god foreground power isn't anything too spectacular.

My point was just that lighting up another one of them at that same speed won’t require another 2W of active platform power. It’ll be lower than that. Apple e-core-tier? No. Agreed. But do I think their cores are more efficient than what AMD/Intel have for background tasks? Seems likely.

Foreground power is pretty good. For Qualcomm at 9W — AMD needs about 16-17W to match that performance, and with Intel, even more. Alternatively, @ 9W Qualcomm have about 25-35% more perf than AMD and 51% more than Intel.

I suspect Strix will improve this for AMD though. Andrei will tell us how it does.

15W cinememe machine materiel which makes sense given the server core background. But again, Z5dense exists. has SMT even, undisputed cinememe@mm^2 king

I’m looking forward to Turin D in servers yeah.

Don't need to be upset.

Don’t start this, haha.

move the goalies to Hamoa-next please, might have better chances in the 2026 market (or at least I hope so).

I think the X Elite will be pretty good. Longhorn even mentioned the pricing isn’t over the top for what it is which makes sense to me but we’ll see.

Correct it is late. And yes absence of real littles hurts their case(s).

 

[Mahboi]

Switching sockets for Zen 6 would go over about as well with consumers/reviewers as a 43% MSRP increase for 9950X. If AMD want's a sure fire way to burn up the good will they've built up since Zen launched stuff like this is the formula.

I agree, but I have to say that Z5 really suplexes the table here.
- it offers a massive growth path from Z4
- the necessity to move to a faster RAM is now earlier than previously thought
- Z6 is supposed to jump the I/O die and uncore from Z2 era tech by 4 whole gens

There's now a way to satisfy all the people who bought Z4 and thought "I'll be able to upgrade to something much better in a few years". Z5, if it is the last of its socket, will not be phased out soon. Look at Z3 if you want proof. It's also 40% supposed perf bump, so while its not Z1->Z3, it is a large upgrade, and probably available all the way to the end of 2026 at least.

I always assumed that Z6 would be on AM5 because there's no way they'll get a new socket going unless they're going to change the RAM, they're not making another socket on DDR5 and changing it 2-3 years after at most. But now, with LPCAMM2 and possibly others into the mix, and the massive bump in CPU perf, it may be more of a tough decision. I'm not knowledgeable enough about tech to know how will changing the I/O die impact changing memory type, but they could very well decide to just jump everything to a new socket, new RAM, new I/O die all at once. (new Northbridge? New memory on package?)

The growth of perf is already going to satisfy a ton of the Z4 clients, and AM5 is still going to last through 2025, like AMD promised. And if AM6 brings worthwhile improvements, people will move on quick. Those who don't will buy Z5 and still have a banger CPU.

Also, there's something I've been thinking for a fair bit, it's that the "upgradability" of AM4 was an extremely Youtuber thing to shill.
If you somehow got data on the amount of people that picked a 3000 series or 2000 series and upgraded to 5000, you'd find it to be much smaller than one might think, despite the red carpet unrolled for it.

Most people will either wait for years, and decide that rather than bumping their CPU up, they may be better off just buying a new PC entirely, or they'll bump up quickly, and at a push, would have bought a whole new mobo + RAM + CPU because they wanted to upgrade anyway. The people that fall into that valley of "bump every 2-3 years, but do not bump every 1-2 years, or buy a new system every time" may be a smaller pool than presented.
Besides the Youtubers seeking the clout of "Speaking for the People", I wonder if there is an actual client mass that's seeking this actively.

 

[adroc_thurston]

I hope!

Even Hawk is fine actually, LPDDR issue is gone.

It’s awesome.

Naa, not in the market where the likes of Dell shipped insufficient choke and VRM cooling counts on 2kilobuck laptops.
it's pennies easily saved, hard sell in PC OEM space.

My point was just that lighting up another one of them at that same speed won’t require another 2W of active platform power.

Yeah but that's frankly cinememe-adjacent stuff more than any real PC usecases, where background stuff will sit pegged at Vmin and foreground will be blasting 4.3 for all that JS perf.

For Qualcomm at 9W — AMD needs about 16-17W to match that performance, and with Intel, even more

yeah but that doesn't work against Z5 across the entire curve.
in 2023 would've been awesome, now, not so much.
More on June 3rd.

Andrei will tell us how it does.

oh no i'll get to be annoying first this time.

I’m looking forward to Turin D in servers yeah.

frankly Strix is more interesting given that CWF seems to be more of a 2h'25 part realistically now.

Don’t start this, haha.

I gotta.

I think the X Elite will be pretty good.

Competent part but nothing market-breaking for 2024 and market-breaker is what QC needs.

Longhorn even mentioned the pricing isn’t over the top

oh Hamoa pricing is bad.
Maybe the 10c chop is reasonable but the full part is too cashmoney.

 

[gdansk]

Also, there's something I've been thinking for a fair bit, it's that the "upgradability" of AM4 was an extremely Youtuber thing to shill.

You were doing pretty well but lost me here. How do you explain the 5000 series selling so well in 2024 - even $300 parts? They are seldom doing new builds with a 5800X3D - a 7600X is cheaper and typically better. It's not some YouTuber shilling conspiracy, upgradability is simply a useful thing and the AM4 people are doing it even if they don't need to.

But this time AMD worded things in a way that Zen 6 is a question mark. And probably because they weren't sure about the memory.

 

[SpudLobby]

Even Hawk is fine actually, LPDDR issue is gone.

Nice.

Naa, not in the market where the likes of Dell shipped insufficient choke and VRM cooling counts on 2kilobuck laptops.
it's pennies easily saved, hard sell in PC OEM space.

Yep. But as a consumer I don’t care and am willing to pay the difference and then a bit. Whether or not they will wise up just depends I guess on how important getting that last bit of power saving is for mobile in the future.

Yeah but that's frankly cinememe-adjacent stuff more than any real PC usecases, where background stuff will sit pegged at Vmin and foreground will be blasting 4.3 for all that JS perf.

Ya. Well we’ll see what the behavior looks like. With Windows power plans + multiple SKUs not all will hit 4.3, there will be a bit more optionality than with like Apple.

yeah but that doesn't work against Z5 across the entire curve.
in 2023 would've been awesome, now, not so much.

Yeah, I expect Z5 to narrow that deficit a good bit in the 9+W range. It’ll be a good.
Also agree the 1-1.5Y delay has dampened the wow factor to this but I suspect the battery life and responsiveness will still be pretty great. If Strix can make that a meaningless point without costing much more (I think it will cost more though) then yeah there’s not a point to the X Elite.

More on June 3rd.

Looking forward to it!

oh no i'll get to be annoying first this time.

Looking forward to this too.

frankly Strix is more interesting given that CWF seems to be more of a 2h'25 part realistically now.

Strix is what I am most excited to see at this point after full X Elite reviews.
Is that soon for CWF? Or late. I haven’t kept up. 144-288c (e-core) on 18A right? That’s about all I know.

oh Hamoa pricing is bad.
Maybe the 10c chop is reasonable but the full part is too cashmoney.

I believe that at the peak version yeah. Though they have lower clocked and cut down SKU variation they’ll milk for less it seems and for me that’s worthwhile, I don’t need all 4.3GHz and 12c. 4GHz & 8-10C is superb, I’ll take the savings.

 

[adroc_thurston]

But as a consumer I don’t care and am willing to pay the difference and then a bit.

Yea but it's always been the free space for OEM penny pinching.

If Strix can make that a meaningless point without costing much more (I think it will cost more though) then yeah there’s not a point to the X Elite.

That's kinda the issue, stx1/krk1 BL bump is let's just say higher than the 5-7% I expected.

Is that soon for CWF? Or late

Late, the expectations were mid-late H1 as a fast followup to the DOA duo.

Though they have lower clocked and cut down SKU variation they’ll milk for less it seems and for me that’s worthwhile, I don’t need all 4.3GHz and 12c. 4GHz & 8-10C is superb, I’ll take the savings.

Kinda the point, other vendors also have the cheaper chops.
QC just prices their parts like they own the market since 8cx g1 and that hasn't worked.

 

[Ghostsonplanets]

MLID with a Zen 5 video:

He also shared another Z5 slide

 

[Mahboi]

You were doing pretty well but lost me here. How do you explain the 5000 series selling so well in 2024 - even $300 parts? They are seldom doing new builds with a 5800X3D - a 7600X is cheaper and typically better. It's not some YouTuber shilling conspiracy, upgradability is simply a useful thing and the AM4 people are doing it even if they don't need to.

But this time AMD worded things in a way that Zen 6 is a question mark. And probably because they weren't sure about the memory.

These two things are not related. A socket will sell out a lot of its last gen CPU, and a larger pool of socket owners will reinforce that.
The clout seeking Youtubers aren't about upgradability, they're about regular upgradability. Because they mistakenly think that a lot of people buy every gen of CPU and don't want to re-buy a mobo and RAM with it.
Most people actually buy a system, run it, and consider changing its parts not when something new is out but when the system is failing at its tasks.

Their idea is:
- Buy Z4
- Z5 comes out, you can easily replace Z4
- Z6 comes out, you can easily replace Z5
It cost you nothing more than the CPU and you saved money.

That model is only applicable to a small minority of nerds that upgrade as soon as they can. The common buyer upgrades as soon as he needs.
The vicious circle here is that the nerds that have this buying pattern are the people who watch their every video, comment, like, interact, and populate their discords and social media interactions (and forums like the one we're in right now). So their community, the people they interact with the most, who share this buying pattern because they are part of the Nerd Elite, encourage this belief that regular upgrades is in demand.

But for the large majority of customers, the buying pattern is:
- Buy Z4
- Z5 comes out, and it's massively better, but expensive and your CPU still works great
- You leave it in until it grows too weak or prices go down
Then:
A)
- Z6 is on AM5, Z5 prices come down
- You decide, based on perf/price, to take Z5 (cheaper) or Z6 (expensive)
- Take Z5, and wait for Z6 to come down in price too, or see if jumping to AM6 is worth it then
- Take Z6, and you've skipped spending a dime on Z5
B)
- Z6 is on AM6, Z5 prices come down anyway
- you realise that Z5 is your last seat on the train
- mald for a bit
- You decide, based on perf/price, to take Z5 (AM5, cheaper) or Z6 (AM6, very expensive)
- Take Z5 (now or after prices go down further), and you still have an upgraded PC that'll last you another 2-3 years or more, after which you'll build a new one after 4+ years with this one
- Take Z6 and spend a ton but get a whole new tier of PC

Ultimately, what's the difference? Your decision is entirely driven by perf/price conditions. If Z6 is excellent enough that it warrants the new mobo + RAM + CPU, then you go for it. If it's not, you stick to Z5. Lifetime of the socket hasn't changed your conditions at all, unless you have decided to buy every generation. Sure, longer sockets mean less general expenses. But it's not nearly as valuable as the Youtubers shill it. It makes you use an older mobo, likely older, slower RAM. A clean build has its own merits.

Key point though: Z6 has to bring in huge leaps if it is on AM6. Otherwise, everyone will look at the 20-25% perf increase and the massive price of a whole new system and go "nope, just gonna take Z5 and hold on to it for years". Exactly like what happened to the first 3Q of Z4, when AM4/Z3 was just a better deal.
Rather than a very vocal minority of turbo nerds and their Knight Youtuber of The Consumer's League heroes, this is really what I expect AMD to decide on: can Z6 be a large upgrade? If yes, AM6 it, because the performance will justify the jump. If not, it'll be on AM5.

I'd personally still bet at random that Z6 will still be on AM5. I do not have any info, it's just wet finger wind guessing. But I don't see AMD having to redo the I/O die, move to LPCAMM2 or DDR6 or anything else, and have a large CPU perf increase all at once. It seems like a very large burger to swallow to me. And they're not the risky types. I'd bet that they'll safely get a new I/O die and tweak Z5 cores into a solid Z6, sell it on AM5, and then prepare the leap to a new socket and new ram, rather than do everything at once.

 

[Rigg]

I agree, but I have to say that Z5 really suplexes the table here.
- it offers a massive growth path from Z4
- the necessity to move to a faster RAM is now earlier than previously thought
- Z6 is supposed to jump the I/O die and uncore from Z2 era tech by 4 whole gens

There's now a way to satisfy all the people who bought Z4 and thought "I'll be able to upgrade to something much better in a few years". Z5, if it is the last of its socket, will not be phased out soon. Look at Z3 if you want proof. It's also 40% supposed perf bump, so while its not Z1->Z3, it is a large upgrade, and probably available all the way to the end of 2026 at least.

I always assumed that Z6 would be on AM5 because there's no way they'll get a new socket going unless they're going to change the RAM, they're not making another socket on DDR5 and changing it 2-3 years after at most. But now, with LPCAMM2 and possibly others into the mix, and the massive bump in CPU perf, it may be more of a tough decision. I'm not knowledgeable enough about tech to know how will changing the I/O die impact changing memory type, but they could very well decide to just jump everything to a new socket, new RAM, new I/O die all at once. (new Northbridge? New memory on package?)

The growth of perf is already going to satisfy a ton of the Z4 clients, and AM5 is still going to last through 2025, like AMD promised. And if AM6 brings worthwhile improvements, people will move on quick. Those who don't will buy Z5 and still have a banger CPU.

While I'm as excited and intrigued as the next guy for what Zen 5 will bring to the table, I think we need to separate rumor from reality before we can make these kinds of declarations. I've watched way too many AMD hype trains derail to not take the rumors with a grain of salt.

DDR6 will most likely be commercially available when zen 6 comes out. The chances that it's commercially viable on a consumer desktop platform in 2026 seem much less likely in my view. AM5 came over 2 years after DDR5 was released and still faced initial hesitation from consumers to adopt it because of increased motherboard and memory costs vs older platforms. More PCIE bandwidth and higher memory speeds increase motherboard cost.

LPCAMM2 seems irrelevant to a socketed desktop platform. I'm intrigued by its prospects for mobile but I don't see it being worth the latency compromise on a platform where the compact size and power savings are meaningless.

Also, there's something I've been thinking for a fair bit, it's that the "upgradability" of AM4 was an extremely Youtuber thing to shill.
If you somehow got data on the amount of people that picked a 3000 series or 2000 series and upgraded to 5000, you'd find it to be much smaller than one might think, despite the red carpet unrolled for it.

Most people will either wait for years, and decide that rather than bumping their CPU up, they may be better off just buying a new PC entirely, or they'll bump up quickly, and at a push, would have bought a whole new mobo + RAM + CPU because they wanted to upgrade anyway. The people that fall into that valley of "bump every 2-3 years, but do not bump every 1-2 years, or buy a new system every time" may be a smaller pool than presented.
Besides the Youtubers seeking the clout of "Speaking for the People", I wonder if there is an actual client mass that's seeking this actively.

Apologies in advance for getting a bit snarky but I have to be honest...... This section of your post comes off as a long winded way to call the Aussie Steve and Tech Jesus' of the world disingenuous shills. Yawn. AM4 's longevity/compatibility has been, and continues to be, good for consumers. Your unfalsifiability fallacy doesn't change that.