AMD what you doing to fight off Alderlake?

[gpuerrilla]

Was this one of the many forum Dave's that either said the 3090FE was a rip off mug's edition, or slated the miners (yet went on to mine themselves and pay for everything)?

Makes reverberating reading for the classis "if AMD bring something competitive to the table then we maybe wont buy intel" brigade in the old CPU discussions.

 

[Dave2150]

Ok you purchased one, now show us how good it is

It was bought for gaming, as I play a few older games that just prefer Intel (bad engine design etc). It performed well with the 3080 and 3090 at 4k, though even a 10th gen i9, Ryzen 5000 series would have performed just as well on the whole. Was upgrading from a 6700k, so anything would have been quite the jump.

Can't easily show you the gaming performance, though included some pics showing my single thread performance (real beast at 5.4Ghz) and my SP90 silicon quality in UEFI, complete with factory set VID levels, showing the low voltage needed for stock 5.3Ghz.

If your next query is asking for my date of birth, ID and passport number, I'll have to pass, as busy waiting for my 12900k box to arrive today, to be worshipped at my intel pedastal, while I conjure my out of stock motherboard into existence

Spoiler

 

[subbytna]

Exactly! Dave has the experience now to advise others

Apply cold compress to burned area

 

[jigger]

Six Huge cores + four Zen 3 sized cores is more accurate, but yeah, didn't think of it like that, brutal in a single sentence.

10 cores 6000mhz memory and 125 watts
6 cores with 3200mhz at 100 watts.

 

[cliffy]

AMD is already doing research on it:
https://www.freepatentsonline.com/10698472.pdf

So you're taking an article that is a year old in which Macri never actually said "we're not doing it" as evidence that AMD won't do hybrid architectures, despite actual filed patents for such hybrid architectures existing?

And rumour/leaker he may be, but MLID's track record the past 18 months or so has been excellent, very few things he's discussed have proven to be incorrect.

Yep AMD have been researching it for years, i never denied that and yes they fled a patent for it in 2017. Plus i said the Machi had said it is not in AMD's current roadmap. There roadmap at the time went up to and included AM5

 

[Steampunk]

I really doubt we'll see AMD do big/little on the desktop. Sure, they may have done research and patents, but every company does just in case, or to deny it to their competitors. AMD still have so much potential in their chiplets. They are already working on making them on smaller processes (which TSMC are able to do, unlike Intel), improve Infinity Fabric (which doesn't have the limitations of Intel's ring bus), install V-cache (which Intel gave up on in the form of 3D cache), keep adding chiplets to increase core (which Intel don't have), and continue to make IPC improvements.

Intel have to go big/little because they have none of the potential of chiplets. All they have is big monolithic cores with very limited room for advancement. Intel can't even get off the 10+infinity process node, no matter how dense they say they make it. Sure, Intel can pump a lot more power to get higher clocks, shave the IHS and chip to try and deal with the heat that produces, but the gas tank is empty on that approach. The cupboard is bare. So Intel stuck some small cores on the side to try and say they have efficiency, and sold it as something new. Intel marketing again leads the way for their engineering.

Intel went with big/little because they had very little else. AMD don't have to do that as they already solved those problems with their chiplet approach, and have much more potential reward in continuing to build on that technology.

Maybe we shouldn't be be predicting that AMD will follow Intel (who in turn copied Samsung/ARM/Qualcomm et al from their mobile phone SOCs), we should be predicting that Intel will have to follow AMD in chiplets, Infinity Fabric, and an actual 7/6/5 nm process node.

 

[Murphy]

Yep AMD have been researching it for years, i never denied that and yes they fled a patent for it in 2017. Plus i said the Machi had said it is not in AMD's current roadmap. There roadmap at the time went up to and included AM5

Rumours, so take them or leave them, point towards Zen 5 in 2023 with what they've apparently codenamed Strix Point, apparently it will be the 3rd chip on AM5 (1st 5nm, 2nd 4nm, 3rd 3nm (not that node sizes matter much these days)).

Maybe by then Windows 11 won't be a buggy mess and DDR5 will be a worthwhile replacement for 4.

 

[jigger]

AMD is already doing research on it:
https://www.freepatentsonline.com/10698472.pdf

AMD have been looking at all sorts since Sky bridge.

16c clusters and 3D stacking seems to be the strategy. 16+16+graphics would be more than enough for most workloads.
Big little might take off but it could also become irrelevant as the market could reject it.

I really doubt we'll see AMD do big/little on the desktop. Sure, they may have done research and patents, but every company does just in case, or to deny it to their competitors. AMD still have so much potential in their chiplets. They are already working on making them on smaller processes (which TSMC are able to do, unlike Intel), improve Infinity Fabric (which doesn't have the limitations of Intel's ring bus), install V-cache (which Intel gave up on in the form of 3D cache), keep adding chiplets to increase core (which Intel don't have), and continue to make IPC improvements.

Intel have to go big/little because they have none of the potential of chiplets. All they have is big monolithic cores with very limited room for advancement. Intel can't even get off the 10+infinity process node, no matter how dense they say they make it. Sure, Intel can pump a lot more power to get higher clocks, shave the IHS and chip to try and deal with the heat that produces, but the gas tank is empty on that approach. The cupboard is bare. So Intel stuck some small cores on the side to try and say they have efficiency, and sold it as something new. Intel marketing again leads the way for their engineering.

Intel went with big/little because they had very little else. AMD don't have to do that as they already solved those problems with their chiplet approach, and have much more potential reward in continuing to build on that technology.

Maybe we shouldn't be be predicting that AMD will follow Intel (who in turn copied Samsung/ARM/Qualcomm et al from their mobile phone SOCs), we should be predicting that Intel will have to follow AMD in chiplets, Infinity Fabric, and an actual 7/6/5 nm process node.

Pretty much how I see it. Big little is way for Intel to survive in the desktop for the next 5 years, but probably not Intel’s long term solution.

 

[Murphy]

Big little might take off but it could also become irrelevant as the market could reject it.

Indeed. IMO AMD are doing the right thing in waiting and letting Intel do the heavy lifting, big/little is a gamble outside of mobile computing and a lot of optimisation has to be done on the software front. (I think it's a gamble that will pay off as Intel will just throw money at any problems on the software front)

I suspect AMD will gain back the 5'ish percent performance advantage Intel now has when they launch 3DVcache in a few months and gain more when they switch to AM5 as by then DDR5 should perform better (currently latencies are a bit sucky)

 

[LePhuronn]

So let me ask you something, Is it Big - Little cores, Multi generation cores or just all the same core like they are now? He says it could be any of them.

You need to go back and rewatch it, because you've clearly misunderstood. His delivery isn't the best so it's easy to get wires crossed.

The "big little" design he's talking about marries Zen 5 chiplets as the "big" cores, and Zen 4 Dense chiplets as the "little" cores. So Zen 5 is both big little and multi generation cores. His information about "all the same core like they are now" was in reference to Zen 4 Dense designs like Bergamo. And since everything is still chiplet based, there's still scope for all "big" Zen 5.

But let's end this sidetrack here...

 

[Steampunk]

You need to go back and rewatch it, because you've clearly misunderstood. His delivery isn't the best so it's easy to get wires crossed.

The "big little" design he's talking about marries Zen 5 chiplets as the "big" cores, and Zen 4 Dense chiplets as the "little" cores. So Zen 5 is both big little and multi generation cores. His information about "all the same core like they are now" was in reference to Zen 4 Dense designs like Bergamo. And since everything is still chiplet based, there's still scope for all "big" Zen 5.

But let's end this sidetrack here...

Is that even big/little as we understand it now? That sounds more like big/big

 

[jigger]

Is that even big/little as we understand it now? That sounds more like big/big

Intel: We got big+little.
AMD: Cool, we got big+bigger.

 

[LePhuronn]

Is that even big/little as we understand it now? That sounds more like big/big

I think it's a case that "big little" has just stuck around since the ARM comparison when Alder Lake's design was first leaked. It's more accurate to use Intel's terminology in having "performance" cores and "efficiency" cores. So the rumoured design for Zen 5 is to have the performance chiplet/s built on Zen 5 and the efficiency chiplet/s built on Zen 4 Dense. Zen 4 Dense is suggested to be a cut down Zen 4 design so more of the resulting smaller cores can be packed into a chiplet.

And given it's still chiplet based, you get the cost effective mix-and-match capability. You could build a pure parallel number cruncher with just the Zen 4D efficiency chiplets, you could go full-bore Threadripper type monster with purely Zen 4 performance chiplets, or you could do the hybrid style with a chiplet of each (although Zen 5 desktop is rumoured to support 3 chiplets). Plus, this approach has the benefit that if hybrid style desktop CPUs are a bust, AMD haven't invested time and money in hybrid dies like Intel, they literally just don't bother to mix chiplets.

 

[jigger]

I think it's a case that "big little" is the phrase going around because of the ARM comparison when Alder Lake's design was first leaked. It's more accurate to use Intel's terminology in having "performance cores" and "efficiency" cores. So the rumoured design for Zen 5 is to have the performance chiplet/s built on Zen 5 and the efficiency chiplet/s built on Zen 4 Dense. Zen 4 Dense is suggested to be a cut down Zen 4 design so more of the resulting smaller cores can be packed into a chiplet.

The big difference is AMD would use a coherent fabric and clusters of upto 16 cores.

 

[LePhuronn]

The big difference is AMD would use a coherent fabric and clusters of upto 16 cores.

Goes without saying.

 

[jigger]

Goes without saying.

Sometimes not on here. Intel’s big+little would look very different to AMD’s.

TBH I don’t think AMD need to consider it. Big cores ahoy.

 

[humbug]

Intel: We got big+little.
AMD: Cool, we got big+bigger.

The only problem with that idea is Zen 3 cores are very much smaller than Intel's P core size, and yet the IPC is identical.

I don't want this to be construed as hating on Intel i like the 12700K but i'm also going to say what i think is reality, Intel's CPU cores are just really really inefficient, they don't have this Big - Little design because its efficient, which in ARM's case it is, but for Intel its because they cannot scale their cores, they are enormous, very slow for their size and they draw a huge amount of power and they can't get many of them in to a single package without runaway out of control power levels.

AMD's cores are very small, very fast and very power efficient and they scale, a 5950X is 90% faster than a 5800X with 0% power gain, hows that for scaling?
They don't need a big little design and they are not going to get sucked into it, Dr Lisa Sue said as much, i'm paraphrasing but basically said "we don't need it and we are not going to get drawn in to that trap".

 

[jigger]

The only problem with that idea is Zen 3 cores are very much smaller than Intel's P core size, and yet the IPC is identical.

I don't want this to be construed as hating on Intel i like the 12700K but i'm also going to say what i think is reality, Intel's CPU cores are just really really inefficient, they don't have this Big - Little design because its efficient, which in ARM's case it is, but for Intel its because they cannot scale their cores, they are enormous, very slow for their size and they draw a huge amount of power and they can't get many of them in to a single package without runaway out of control power levels.

AMD's cores are very small, very fast and very power efficient and they scale, a 5950X is 90% faster than a 5800X with 0% power gain, hows that for scaling?
They don't need a big little design and they are not going to get sucked into it, Dr Lisa Sue said as much, i'm paraphrasing but basically said "we don't need it and we are not going to get drawn in to that trap".

I agree with you. AMD have been outstanding and Intel are in trouble. Alder lake is two of Intel’s past designs hashed together in a less than ideal chip. Intel’s issues remain.

 

[SeeNoWeevil]

Intel's CPU cores are just really really inefficient

 

[jigger]

The only voices saying this are the ones in your head.

Is that what the voices tell you, Intel cores are weak?

Maybe listen to those voices sometimes they might be right.