Discussion Zen 7 speculation thread

[LightningZ71]

Meme SKU, just like the 5G HFT parts from eons past.
I mean something you can just go and buy.

I mean, I personally know someone with that very part. I could quite literally walk up to him and offer to buy it tomorrow. I won't because the 9900K I have sitting on a shelf is a better chip, but still, it existed at retail and was actually purchase-able in the US (maybe via lottery for some, memory fails me), unlike Intel's first token, half broken 10nm parts.

 

[LightningZ71]

(Ignore, duplicate)

 

[Win2012R2]

Epic can fix the latter, and are in the process of doing so in their UE6 roadmap.

Epic really need to fix the issue of getting engine upgraded for existing games - games start dev at v5.X and years later when they release they don't use current version.

 

[soresu]

Epic really need to fix the issue of getting engine upgraded for existing games - games start dev at v5.X and years later when they release they don't use current version.

This is very much a non trivial endeavor if the devs deviate to any significant degree from the "off the shelf" engine code.

For example, the Star Wars MMO "The Old Republic" started development on a pre release (<1.0) version of the Hero Engine, and people have constantly asked why the devs don't just port it to the later v2 of Hero Engine.

The response is that the engine variant for SWTOR has been drastically customised and a simple port is not at all a viable move (even as far back as the games initial release).

That being said, Epic have purposefully made it as easy as possible to port between late UE4 and UE5, and likely major versions of UE5 too.

But some dev effort still has to be made, whether small or massive.

I hear about Stalker 2 making a jump from a very early version of UE5 to 5.5, and possibly 5.6 after that.

It stands to reason that if the devs keep non standard stuff to a minimum after the 5.5 upgrade then it will be much easier to make the jump to 5.6 later.

 

[Win2012R2]

the Hero Engine

What's that?

UE5 is the industry standard now with annual releases, it should be upgradeable by design, not being starting point and then dead end in game dev.

I hear about Stalker 2 making a jump from a very early version of UE5 to 5.5, and possibly 5.6 after that.

That team is doing the right thing.

 

[soresu]

What's that?

Excellent question.

Outside of SWTOR I'd never heard of it before, and never since 🤣😆😂

Clearly Bioware was firing on all cylinders when they made that choice. /s

 

[soresu]

UE5 is the industry standard now with annual releases, it should be upgradeable by design, not being starting point and then dead end in game dev.

Again, I don't think you appreciate how much many game projects deviate from the off the shelf state of the game engine.

Whether its individual dev team choices for optimisation, or extending/adding features by plugging into engine code that is subject to change in future versions, the main point is that the only way they can ever be somewhat certain that a project will translate cleanly from one version of UE5 to the next is by only using the totally virgin engine.

That is all on the game devs, not Epic.

 

[Win2012R2]

Again, I don't think you appreciate how much many game projects deviate from the off the shelf state of the game engine.

Totally appreciate that.

The issue here is that industry leading dominant engine designers should design it in a way that allows at least minor upgrades to benefit from annual improvements.

 

[MadRat]

Just keep everything scaling incrementally so the devs can keep up.

Once you jump to fiber for data transfers, how many pins can you eliminate?

 

[marees]

Assuming zen 7 comes out in 2028 or 2029 with RDNA 5 or 6 iGPUs & there is no dedicated NPU & the MS tax remains at 60 TOPS then how many WGPs are we looking at a minimum for the iGPUs 🤔🤔

 

[511]

Assuming zen 7 comes out in 2028 or 2029 with RDNA 5 or 6 iGPUs & there is no dedicated NPU & the MS tax remains at 60 TOPS then how many WGPs are we looking at a minimum for the iGPUs 🤔🤔

MS Tax is going to be 70 btw

 

[igor_kavinski]

the main point is that the only way they can ever be somewhat certain that a project will translate cleanly from one version of UE5 to the next is by only using the totally virgin engine.

They should actually do what they can with the vanilla engine instead of inflicting pain on themselves by helping Nvidia get higher framerates and losing the ability to simply switch to a newer engine version as soon as it is released.

 

[igor_kavinski]

The issue here is that industry leading dominant engine designers should design it in a way that allows at least minor upgrades to benefit from annual improvements.

It's totally Nvidia's meddling, just like IT managers often want only Intel hardware, project leads want to have Nvidia hardware optimized games which requires Nvidia plugins and other non-portable crap creeping into the game executable.

It used to be so bad before that certain games had to be manually patched to work on non-Nvidia hardware even though it had perfect DirectX feature compatibility.

 

[soresu]

They should actually do what they can with the vanilla engine instead of inflicting pain on themselves by helping Nvidia get higher framerates and losing the ability to simply switch to a newer engine version as soon as it is released.

It's definitely the ideal, but sometimes the base engine simply doesn't have everything they need to make their 'artistic vision' a reality on screen.

As engines gain more and more versatility you would think that this would be less of an issue, but at the same time the engine makers don't want to add bloat that will make it less attractive to use, so it's definitely a tightrope to navigate.

 

[igor_kavinski]

they need to make their 'artistic vision' a reality on screen.

Their artistic vision is just smoke and mirrors to incorporate Nvidia technical demo features into their games.

Nvidia is a cancerous company. That much became evident with the RTX generation. Rather than let people enjoy gaming, they burdened them with additional debt for questionable improvement in quality which doesn't look all that better compared to good old baked lighting done right.

 

[511]

Their artistic vision is just smoke and mirrors to incorporate Nvidia technical demo features into their games.

Nvidia is a cancerous company. That much became evident with the RTX generation. Rather than let people enjoy gaming, they burdened them with additional debt for questionable improvement in quality which doesn't look all that better compared to good old baked lighting done right.

EA said F to Nvidia buy saying no RT in BF6

 

[Thunder 57]

From Agner Fog:

View attachment 128813

Assuming that a context switch happens in less than 10 clock cycles, that could allow 50% smoother gaming performance on average.

You just have to be the smartest guy in the room, son't you? Calling CEO's idiots and saying how dumb CPU architects are. Or how the P4 was really awesome but left to die because politics. You used to be fun and sometimes still are but have also been crossing over into this realm of annoying.

Branch prediction is very accurate and it would be monumentally stupid to run both at once and pick the correct one.

 

[igor_kavinski]

WARNING: adroc mimicry ahead.

You just have to be the smartest guy in the room, son't you?

I am.

Calling CEO's idiots

They mostly are and get away with it. I have first hand experience of it in my own company.

saying how dumb CPU architects are.

They are only human after all. They are allowed to be dumb.

Or how the P4 was really awesome but left to die because politics.

Because it was and is awesome.

You used to be fun and sometimes still are but have also been crossing over into this realm of annoying.

Here's a coupon code XXYYZZ. Feel free to buy anything else you want.

Branch prediction is very accurate and it would be monumentally stupid to run both at once and pick the correct one.

With reduced core power consumption on upcoming advanced processes, it may not be that prohibitive to burn some extra power for the sake of some extra fps.

 

[adroc_thurston]

adroc mimicry ahead.

Looks pretty bad.
It only works when you're trying to crush nerd dreams, and you're a dreaming nerd yourself.

 

[DavidC1]

With reduced core power consumption on upcoming advanced processes, it may not be that prohibitive to burn some extra power for the sake of some extra fps.

Opposite actually. Process gains are starting to stall, while costs are increasing at a rapid rate. "New process gen" in 2026+ is like 0.5x of 2003-2024. That's on top of where the gains have been reducing for past 20 years. After 0.18u, they needed increasingly more complex process, different materials, and layers to advance. Starting with copper interconnects.

@Thunder 57 There are still areas where Branch prediction accuracy can be much improved. Those will require lot more work though. Essentially some will basically require predicting the future and will never get there.

Calling CEO's idiots and saying how dumb CPU architects are.

Experts can often get myopia and sometimes need an outsider's point of view. Whereas they can do details better, they might still miss the big picture. Just like I don't need to know every little detail an engineer designing engines to know basics of how they work. You can have very egregious examples like that guy at Intel quoting his daughter wanting more laptops so that's how the rest of the world might think, even though in "first world" countries like US have record number of people on food stamps, buying on extreme debt. Today at Costco(Canada) there was a flyer saying donate so Canadian kids can have breakfast. Really stretching the definition of first world and developed nowadays for average people.

 

[soresu]

EA said F to Nvidia buy saying no RT in BF6

They didn't explicitly say no RT, they just said it isn't a launch or near term patch priority.

So don't even expect it a few months out like the DTDA path tracing patch, though probably a wee bit sooner than the Witcher 3 RT patch 😂

 

[soresu]

Experts can often get myopia and sometimes need an outsider's point of view

This is (at least IMHO) precisely why Jim Keller never stuck around, and why many µArch engineers tend to roam.

Stick around too long and you will not only get stuck in your ways, you will encourage others to think the same way too.

 

[Thunder 57]

Opposite actually. Process gains are starting to stall, while costs are increasing at a rapid rate. "New process gen" in 2026+ is like 0.5x of 2003-2024. That's on top of where the gains have been reducing for past 20 years. After 0.18u, they needed increasingly more complex process, different materials, and layers to advance. Starting with copper interconnects.

@Thunder 57 There are still areas where Branch prediction accuracy can be much improved. Those will require lot more work though. Essentially some will basically require predicting the future and will never get there.

Experts can often get myopia and sometimes need an outsider's point of view. Whereas they can do details better, they might still miss the big picture. Just like I don't need to know every little detail an engineer designing engines to know basics of how they work. You can have very egregious examples like that guy at Intel quoting his daughter wanting more laptops so that's how the rest of the world might think, even though in "first world" countries like US have record number of people on food stamps, buying on extreme debt. Today at Costco(Canada) there was a flyer saying donate so Canadian kids can have breakfast. Really stretching the definition of first world and developed nowadays for average people.

Didin't you know you can buy food with Klarna now. Just 4 easy payments! And the rise of get your money before payday apps is a bad sign.

 

[DavidC1]

This is (at least IMHO) precisely why Jim Keller never stuck around, and why many µArch engineers tend to roam.

Stick around too long and you will not only get stuck in your ways, you will encourage others to think the same way too.

There was an article at Intel some time ago about an engineer talking about how you don't want to focus just on your field so much, because even in terms of inventions broadening makes your product better. Our brain is the ultimate general purpose processor.

My father was talking about how he went to a specialist in dentistry and couldn't figure out why he had pain. He did some searching and some articles suggested that it may be due to a weak heart, which he does have. Another example of myopia.

 

[igor_kavinski]

There was an article at Intel some time ago about an engineer talking about how you don't want to focus just on your field so much, because even in terms of inventions broadening makes your product better. Our brain is the ultimate general purpose processor.

My father was talking about how he went to a specialist in dentistry and couldn't figure out why he had pain. He did some searching and some articles suggested that it may be due to a weak heart, which he does have. Another example of myopia.

The Multidisciplinary Approach to Thinking

The Multidisciplinary Approach to Thinking by Peter Kaufman is transcribed and reproduced. The talk offers a simple formula for living the best life you could possibly live.

fs.blog