MLiD "Leaks". :laugh:
No SMT4 this time?:laugh:

Anyone that follow the news/rumors and actual leakers on twitter, can come up with that list.
This guys is a BS artist.

z1n0xMLiD "Leaks". :laugh:
No SMT4 this time?:laugh:

Anyone that follow the news/rumors and actual leakers on twitter, can come up with that list.
This guys is a BS artist.

He tends to get quite a lot correct but a lot of it is based on logical thinking and knowing history really.
Idk, I dont want his channel, watched some videos but they are boring af and he claims he does not want ot be a "leak" channel but he knows that that is where the money lies so thats what he does.

AMD will maybe release Zen 4 in maybe Q3 2022 that will maybe have a 6 nm IO die and maybe 2, or maybe 3 chiplets with maybe 8 cores each that maybe offer 20% IPC gain. I love rumours.

Isn't this the same guy that basically covers anything from 5~50% "IPC leaks" with his BS videos, he's like the YT version of WTFtech :rolleyes:

ZoneDymoHe tends to get quite a lot correct but a lot of it is based on logical thinking and knowing history really.
Idk, I dont want his channel, watched some videos but they are boring af and he claims he does not want ot be a "leak" channel but he knows that that is where the money lies so thats what he does.

That's only because he covers pretty much the entire spectrum, like I said WTFtech on YT :toast:

AusWolfAMD will maybe release Zen 4 in maybe Q3 2022 that will maybe have a 6 nm IO die and maybe 2, or maybe 3 chiplets with maybe 8 cores each that maybe offer 20% IPC gain. I love rumours.

Or maybe all of them, if rumors are to be believed :D

R0H1TIsn't this the same guy that basically covers anything from 5~50% "IPC leaks" with his BS videos, he's like the YT version of WTFtech :rolleyes:
That's only because he covers pretty much the entire spectrum, like I said WTFtech on YT :toast:
Or maybe all of them, if rumors are to be believed :D

I like +50% more than +20%... but if this is not percentage of increase of prices and bills.

Until I can see actual performance data/test results & install one in my rig with my grubby little hands, it's all friggin VAPORWARE to me, hehehe :)

How many decades have we been on this puny, minischule "5-10-15-20% gains" carnival ride anyways ????

Reminds me of Intsmel's never-ending 14nm ++++++++++++++++++++++++++++ routine :cry:

Title needs the word "reportedly" or something to the same effect.

bonehead123Until I can see actual performance data/test results & install one in my rig with my grubby little hands, it's all friggin VAPORWARE to me, hehehe :)

How many decades have we been on this puny, minischule "5-10-15-20% gains" carnival ride anyways ????

Reminds me of Intsmel's never-ending 14nm ++++++++++++++++++++++++++++ routine :cry:

well the name "Moore's Law is Dead" did not come from nowhere ;)

also im still on the 2600k so by the time ill upgrade (fingers crossed Alderlake is worth a damn otherwise Zen 3/4 it is) the 20%'s over time will have added up to a nice jump.

bonehead123How many decades have we been on this puny, minischule "5-10-15-20% gains" carnival ride anyways ????

I wouldn't go decades. 10%-20% has been from like 4 years now since Ryzen 1000 was released. Before that was 2-5% tops.

ZoneDymoHe tends to get quite a lot correct but a lot of it is based on logical thinking and knowing history really.
Idk, I dont want his channel, watched some videos but they are boring af and he claims he does not want ot be a "leak" channel but he knows that that is where the money lies so thats what he does.

He gets a lot correct because he also gets a lot wrong. He basically puts it all out there and flip-flops on everything. I got fed up and unsubscribed his channel. wccftech and videocardz are by far more reliable than this guy.

Kudos to him for finding this gold mine, tho'.

bonehead123How many decades have we been on this puny, minischule "5-10-15-20% gains" carnival ride anyways ????

A 20% IPC uplift generation-to-generation is pretty good in recent history, especially seeing how IPC gains get harder to come by as more revisions are done to an architecture. Intel Core from ~2010 to ~2016 was ~5-10% per generation, before stagnating entirely for several years. Percentage gains also compound over time: If you have +10% each time for three successive generations the fourth gen has 133.1% the IPC of the first. If you have +20% each time you suddenly have 172.8% in the same time - that's a pretty major difference, despite the percentages being close. Are you expecting 50-100% improvements generation-over-generation? 'Cause if so, you need to lower your expectations. That isn't going to happen again unless someone completely revolutionizes how computers work. Remember, thousands of very smart people have been working on making their current(-ish) way of working better for several decades. It stands to reason that it's harder to come by major improvements by now.

Slightly disappointing IPC lift, just when AMD could have struck the hammer blow to Intel, still 20% is far better than most Intel IPC uplifts (combined, for the most part!) over the last 10 years.

Also disappointing is DDR5 performance, the numbers released so far look awful, with DDR4 being the better option for at least another 2 years. It would have been good to see AMD drop one last AM4 CPU refresh, with a nice 10% IPC uplift.

But the real question will be if AMD has the ability or balls to up the core count on AM5.

stimpy88Slightly disappointing IPC lift, just when AMD could have struck the hammer blow

Just want to point out something, haven't got beef with anybody;
The hardest tape out is the one just leap frogging the previous. These chips used to cost a lot. Now, AMD can keep the interconnects the same and just upgrade execution units year after year. You do that with a monolithic die and you have just blown your investment payoff projections with huge tapeout costs.

stimpy88Slightly disappointing IPC lift, just when AMD could have struck the hammer blow to Intel, still 20% is far better than most Intel IPC uplifts (combined, for the most part!) over the last 10 years.

Also disappointing is DDR5 performance, the numbers released so far look awful, with DDR4 being the better option for at least another 2 years. It would have been good to see AMD drop one last AM4 CPU refresh, with a nice 10% IPC uplift.

But the real question will be if AMD has the ability or balls to up the core count on AM5.

Why would they increase the core count? 16 cores on a consumer platform is already bonkers. Anything more would require a third CCD or increasing the core count per CCD, which, respectively, is impossible or highly unlikely. They've got TR and EPYC for those rather few use cases where 16c32t still isn't more than enough. Given that current-gen consoles are 8c16t and current-gen games barely make use of those cores/threads, there's not going to be a benefit for gaming for a long, long time. So why invest significant engineering effort for the very, very niche use case of enabling 20 or 24 cores in an MSDT platform? Also, imagine trying to cool 3 8-core 5nm CCDs in an AM4-sized package at any kind of useful all-core clock speed ... no thanks.

DDR5 brings the bandwidth, so it'll be good for bandwidth-limited use cases (iGPU gaming is getting one hell of a boost at least), but beyond that consumer use cases are more latency bound. And DDR5 is designed primarily for servers, which are mostly bandwidt-bound, hence it staying at the same or even increasing latencies in order to increase bandwidth. DDR4 and DDR5 is likely to perform equivalently for most non-datacenter uses for quite some time. DDR5 might not ever be tangibly faster in terms of latency - but then DDR RAM latencies haven't budged much at all since DDR(1).

PS: don't forget that even Intel designed its fastest cpu cores using the same MCM layout. Clarkdale cpus with their seperate memory controller wasn't the fastest, but in clock speed it was as fast, if not the fastest, there was.

btarunraccording to a report by Moore's Law is Dead

Could someone give a hint of the track record of the said source?

medi01Could someone give a hint of the track record of the said source?

All over the place. Seems to have some decent sources at times, but also well known to blurt out anything and everything as if it were fact and never address it again when half of it turns out to be wrong.

It would be great if AMD continued to put competitive pressure on Intel. Consumers win in the end.

Divide OverflowIt would be great if AMD continued to put competitive pressure on Intel.

I would not refer to this:



as competitive pressure.

(Largely corrupt) OEM ties is the only reason Intel is not feeling major pain from it, notebook/server market in particular.

ValantarWhy would they increase the core count? 16 cores on a consumer platform is already bonkers. Anything more would require a third CCD or increasing the core count per CCD, which, respectively, is impossible or highly unlikely. They've got TR and EPYC for those rather few use cases where 16c32t still isn't more than enough. Given that current-gen consoles are 8c16t and current-gen games barely make use of those cores/threads, there's not going to be a benefit for gaming for a long, long time. So why invest significant engineering effort for the very, very niche use case of enabling 20 or 24 cores in an MSDT platform? Also, imagine trying to cool 3 8-core 5nm CCDs in an AM4-sized package at any kind of useful all-core clock speed ... no thanks.

DDR5 brings the bandwidth, so it'll be good for bandwidth-limited use cases (iGPU gaming is getting one hell of a boost at least), but beyond that consumer use cases are more latency bound. And DDR5 is designed primarily for servers, which are mostly bandwidt-bound, hence it staying at the same or even increasing latencies in order to increase bandwidth. DDR4 and DDR5 is likely to perform equivalently for most non-datacenter uses for quite some time. DDR5 might not ever be tangibly faster in terms of latency - but then DDR RAM latencies haven't budged much at all since DDR(1).

Because that is the rumour going round, that they may go up to 24 cores on the desktop this time.

DDR5 is a bust, at least for now, and certainly does not "bring the bandwidth", at least according to a press release on this site a couple of weeks ago. DDR4 has higher bandwidth, and MUCH better latency right now, and probably for the next couple of years, until DDR5 ramps up its speed and lower it's timings. - Just like every single memory standard we have ever had on the PC - The new standard is slower, has crappy timings, and is expensive... We have seen this many times before.

I think a 20% IPC gain, but a loss of memory bandwidth and higher memory latency, is not exactly stunning this time round - especially if we keep the thought of Intel not sitting on it's ass much longer taken in to account.

MLiD make many claim but in his video, he color code it based on how accurate he think they are. I watch his video, take it as entertainement but i dont take them as cash. Nothing beat real benchmark. But to be honest he was pretty accurate on many things. I tend to respect him a little bit more than some other channel that have title like X CPU or GPU CRUSHES or things like that. RedTechGaming is well know for that. Click bait title about future product but you watch his video and you didn't learn anything.

at least MLiD talk about the actual stuff so. But like usual, rumors tend to become more real when multiple reliable source start to talk about it.

stimpy88Because that is the rumour going round, that they may go up to 24 cores on the desktop this time.

DDR5 is a bust, at least for now, and certainly does not "bring the bandwidth", at least according to a press release on this site a couple of weeks ago. DDR4 has higher bandwidth, and MUCH better latency right now, and probably for the next couple of years, until DDR5 ramps up its speed and lower it's timings. - Just like every single memory standard we have ever had on the PC - The new standard is slower, has crappy timings, and is expensive... We have seen this many times before.

I think a 20% IPC gain, but a loss of memory bandwidth and higher memory latency, is not exactly stunning this time round - especially if we keep the thought of Intel not sitting on it's ass much longer taken in to account.

DDR5 is bringing more bandwidth. DDR5-6400 is double the bandwidth of DDR4-3200. The actual real life latency (And not the memory timming) is in the same ballpark as all previous generation of DDR memory (about 14ns), The memory have new function that are game changer like same bank refresh (Only a bank is refreshed and the rest of memory can continue to be accessed unlike right now when the memory ops pause while the memory is being refresh). Also Onboard voltage controller, on die ECC, dual channel per dimm. etc.

And, same thing as all previous generation. JEDEC SPEC ram will suck versus previous DOCP/XMP 'retail memory'. There will be the same XMP/DOCP retail memory kit around that have better timing than JEDEC.

The only thing there is at first it is going to be cost prohibitive versus DDR4. But DDR5 is better and will perform better than DDR4 in same scenario (JEDEC VS JEDEC or XMP high end kit vs XMP high end kit)

20% uplift is spicy!

No fabs/supply

Delayed 2 years ;)

z1n0xMLiD "Leaks". :laugh:
No SMT4 this time?:laugh:

Anyone that follow the news/rumors and actual leakers on twitter, can come up with that list.
This guys is a BS artist.

Exactly, it's just speculation based on stuff found on Twitter, Reddit and various forums. The "business model" of these "leakers" (both on YouTube and various websites) is to jump on every trend and create click-bait headlines, wild predictions etc. Their so-called "leaks" are all over the place, and people tend to forget all the mispredictions.
The few times they are right, then pretty much anyone could have reached the same conclusion.

People need to use their common sense and see through this. Pretty much all these leakers claim to have numerous "sources" inside Intel, AMD and Nvidia. But think about that for a moment; a real source with access to useful information under NDA would risk losing their job and could be facing a large lawsuit for leaking this, and they are supposedly doing so regularly to nobodies on YouTube with nothing in return?
In reality though, they are pulling all of these "leaks" out of their tuchus.

R0H1TIsn't this the same guy that basically covers anything from 5~50% "IPC leaks" with his BS videos, he's like the YT version of WTFtech :rolleyes:
That's only because he covers pretty much the entire spectrum

Yeah, his "IPC" estimates are all over the place, even for Alder Lake he has been posting claims at ~10-30% at various times, claiming it to be accurate at the time. But in reality though, real IPC is determined by architecture, and is actually known fairly precisely at the design stage, it's one of those metrics that will not change (unless they have to disable features etc.).

His claims in this video about Zen 4 not being design complete, and "things" can change clearly displays he is clueless about how microarchitectures are developed. The features of an architecture is decided before the design is begun, no major feature can be added far into the design process. Things like AVX-512, SMT4 etc. are fundamental design features that affects the entire design of the pipeline. In the past, he has claimed Zen 3 exists in SMT2 and SMT4 variants, if this were true, then these would have been developed as separate designs from the start, yet we've seen no evidence of its existence.

So remember this; real sources who actually work on these projects would know these details, and they are not subject to change. Last minute design changes are just excuses used by "leakers", when in reality the features of Zen 4 were decided years ago.

But this guy usually takes the cake whenever he claims to know much more, but don't want to tell us yet for some reason.
Anyone can do this; claim to know next week's lottery numbers, but only reveal them after the fact…

stimpy88Slightly disappointing IPC lift, just when AMD could have struck the hammer blow to Intel, still 20% is far better than most Intel IPC uplifts (combined, for the most part!) over the last 10 years.

No one should complain if we can get ~20% real IPC gains ever two years for a while.
That will be great, and result in noticable improvements in responsiveness etc.
And remember, you don't have to buy every generation ;)

stimpy88It would have been good to see AMD drop one last AM4 CPU refresh, with a nice 10% IPC uplift.

"Refresh"?
IPC gains come from architectural improvements, so unless some feature have been disabled/tuned down due to a bug, you will not see any IPC gains from a "refresh".

stimpy88But the real question will be if AMD has the ability or balls to up the core count on AM5.

I actually think it would be a bad move for AMD to do this, in the long term;
The majority of desktops CPUs are sold to system integrators, and most of their customers look at specs like clock speed and core count (and is blissfully unaware that those high-core 65W CPUs will throttle like crazy), so good "specs" on paper sells. This is why Intel (and probably AMD) is moving to hybrid CPU designs for the desktop, even though it makes little practical sense. It's hard to "go back" and sell a faster 6 core once your customers already bought a 16 core of the previous generation. There are a lot of great CPU improvements coming in the next 5+ years, but they will require a lot of die size, so building >16 core low TDP CPUs with just high-performance cores is unrealistic for the foreseeable future.

I was listening to one of his earliest videos back when and he had a guest on. He was saying to the guest how he wanted to start a You Tube channel and felt that doing a "tech" site would be low hanging fruit that he felt he could tap into, not because he gave a sh!* about the subject but because the interest was growing so fast and it would be easier to cash in. To the dismay of the guest by the way. He proclaims this burning passion so much that He's even conned himself into believing it. He says I and me more times in one ten minute video than a politician during a two month road trip.
Shyster