News AMD's Robert Hallock Dispels Zen 4 PCIe Gen4 Rumors, Talks Future of 'Zen Philosophy'

ottonis

Prominent
Jun 10, 2020
37
27
560
0
With the overwhelming success of big.Little architectures such as in the current crop of Snapdragon SoCs or M1 Apple silicon and some intriguing leaked AL benchmarks, I wonder whether AMD's approach of uniform small(er) efficient faster cores is outsmarting the competition or if they are just lagging behind w/r to microarchitecture and trying to sell it as the next best thing since sliced bread ("it's not a bug, it's a feature").
I really hope that AMD have some very smart and effective advancements up their sleeves for their upcoming Zen4 beside the stacked L3 cash.

What they really need to do is to integrate some simple hardwired ASICS into their APUs that would serve as ultra low energy pipelines for image signal processing / video/audio/HDR/ encoding/decoding tasks.
This will make all the content creators very happy.
 
Reactions: artk2219

TJ Hooker

Champion
Ambassador
What they really need to do is to integrate some simple hardwired ASICS into their APUs that would serve as ultra low energy pipelines for image signal processing / video/audio/HDR/ encoding/decoding tasks.
This will make all the content creators very happy.
You're in luck, AMD has had dedicate hardware codecs in place for years (as do Intel and Nvidia GPUs).

At one point they also had a DSP for audio processing (TrueAudio), not sure if that's still a thing though.
 
Reactions: artk2219

ottonis

Prominent
Jun 10, 2020
37
27
560
0
You're in luck, AMD has had dedicate hardware codecs in place for years (as do Intel and Nvidia GPUs).

At one point they also had a DSP for audio processing (TrueAudio), not sure if that's still a thing though.
I know. However, AMD's video encoding hardware (VCE/VCN) is not exactly regarded as the best on the market. Video encoding quality seems to lag behind nVidia NVENC in most use cases. And this was tested with full-sized Radeon cards, not APUs.
Apple's M1 systems have become the new darling of low budget YouTube productions because they allow for seamless and fluid playback of multiple 4k streams within NLEs such as PP or Resolve (or even better: FCP).

That's what I would love to see from future AMD's APUs: optimized and highly efficient, smooth operation of video editing software.
 
Oct 8, 2021
20
3
15
0
With the overwhelming success of big.Little architectures such as in the current crop of Snapdragon SoCs or M1 Apple silicon and some intriguing leaked AL benchmarks, I wonder whether AMD's approach of uniform small(er) efficient faster cores is outsmarting the competition or if they are just lagging behind w/r to microarchitecture and trying to sell it as the next best thing since sliced bread ("it's not a bug, it's a feature").
I really hope that AMD have some very smart and effective advancements up their sleeves for their upcoming Zen4 beside the stacked L3 cash.

What they really need to do is to integrate some simple hardwired ASICS into their APUs that would serve as ultra low energy pipelines for image signal processing / video/audio/HDR/ encoding/decoding tasks.
This will make all the content creators very happy.
I dont see overwhelming success for big little outside the mobile PC industry. Actually on desktops it is a waste of silicon.
 
Reactions: artk2219
Apr 1, 2020
278
124
860
0
3D V-Cache is a new technology AMD announced several months back that will give Ryzen up to a 15% performance improvement in gaming workloads.
With an asterisk the size of a small island nation no doubt. Also with that much L3 cache there's going to be a very tidy price premium attached for what is likely going to be a very situational based performance gain.

I want a price war again so we can win instead of AMD and Intel's bank accounts.
 
Reactions: artk2219

ottonis

Prominent
Jun 10, 2020
37
27
560
0
I dont see overwhelming success for big little outside the mobile PC industry. Actually on desktops it is a waste of silicon.
True, but isn't the world moving towards mobile?
The most recent numbers show a 54%, market share for mobile computers vs 43% for Desktop computers (see https://gs.statcounter.com/platform-market-share/desktop-mobile-tablet).

Many people do audio and video editing on their laptops. That was unheard of 10-15 years ago.
Of course, VFX artists and professionals will still prefer a full blown workstation, but more and more work is being done on notebooks nowadays.
 
Reactions: artk2219

spongiemaster

Respectable
Dec 12, 2019
1,409
670
2,060
0
I dont see overwhelming success for big little outside the mobile PC industry. Actually on desktops it is a waste of silicon.
Disagree. The overwhelming percentage of mainstream software is not highly threaded and that isn't going to change any time soon. Big little allows a higher percentage of CPU die space to be utilized more of the time thus making it a more performance optimized design for mainstream desktop use. On the flip side big little can be a power optimized design for mobile markets by disabling the big cores when not needed or to extend battery life. One design that can be optimized either way.
 
Last edited:

JWNoctis

Upstanding
Jun 9, 2021
340
73
290
5
Disagree. The overwhelming percentage of mainstream software is not highly threaded and that isn't going to change any time soon. Big little allows a higher percentage of CPU die space to be utilized more of the time thus making it a more performance optimized design for mainstream desktop use. On the flip side big little can be a power optimized design for mobile markets by disabling the big cores when not needed or to extend battery life. One design that can be optimized either way.
They might be hedging on their newer architecture staying as comparatively efficient as they currently are. Why do 8C+8c when you could make 12C just as efficient and have better performance? For mobile processors, Alder Lake would need to have an almost 20% improvement in power efficiency over Tiger Lake which is on the same node, just to match the efficiency of current Zen 3 processors on the market, in my experience with the latter two.
 
Last edited:
Reactions: artk2219

JayNor

Commendable
May 31, 2019
299
42
1,710
0
"This is done by 3D stacking multiple layers of L3 cache on top of each other and connecting the layers to the CPU with a high-quality interconnect"

I read somewhere it is just one extra layer of cache on top of the current L3. TSM has talked about the ability to do multiple layers, but what is your source that says more than one layer is being stacked on top in this case?
 
Reactions: artk2219

escksu

Respectable
Aug 8, 2019
451
157
1,860
0
I dont see overwhelming success for big little outside the mobile PC industry. Actually on desktops it is a waste of silicon.
You need to note that laptops are outselling desktops 2:1 today. This means around 66% of the market share are laptops, only 33% are desktops.

And then, it matters for desktops too. Power consumption is not an issue for home users but its still an issue for corporates. When you have thousands of desktops, higher power conusmption means more money has to be spent on power and cooling. If you could reduce power by just 10W per PC, 1000PC equates to 10,000W (10KW), thats a big difference. ITs also a big difference for your air conditiong.
 

VforV

Respectable
Oct 9, 2019
289
135
1,870
1
lol the same intel that sat on skylake for 6+ years AMD will be fine.
This is not the same intel as of few years ago... you're underestimating the big sleeping (now awoken) giant with new leadership and new hardware designs. It's not gonna be the same.

You don't play with the giant as AMD does now, you keep punching him harder and harder. AMD slowing their punches is not good for them, at all.

They were much more agressive up to Zen3 (especially Zen2 and Zen3) and now they are getting softer with Zen 3D and slacking.
 

JWNoctis

Upstanding
Jun 9, 2021
340
73
290
5
This is not the same intel as of few years ago... you're underestimating the big sleeping (now awoken) giant with new leadership and new hardware designs. It's not gonna be the same.

You don't play with the giant as AMD does now, you keep punching him harder and harder. AMD slowing their punches is not good for them, at all.

They were much more agressive up to Zen3 (especially Zen2 and Zen3) and now they are getting softer with Zen 3D and slacking.
Slacking, or caught between a rock and a hard place by shortages, and difficulties with additional process, yield, and design? Just how long did it take for Zen 3 availabilities to improve?

But yes, unless AMD manages to pull a metaphorical rabbit out of their hat, it looked like open season for Alder Lake for the next half year at least at this point, if it's half as good as it's marketed to be.
 
Reactions: artk2219 and VforV

dalek1234

Commendable
Sep 27, 2019
53
18
1,535
0
Nice to finally get some information about Zen3 refresh with with 3D cache, straight from AMD. It puts all the misinformation out there, about there not being a Zen3 refresh, to rest.

Zen3+, or whatever they end up calling it, is going to be my next build. Early 2022? I can wait.
 
Reactions: artk2219

Gillerer

Distinguished
Sep 23, 2013
323
44
18,890
30
Zen 3D, not even a paper launch in December... tsk tsk tsk.
I can't believe any consumer would want a paper launch.

Companies would also have come to know that after all the disappointments with availability that we've had for the past 2 years, if they did a paper launch, the tech media would rip them a new one.

They might be hedging on their newer architecture staying as comparatively efficient as they currently are. Why do 8C+8c when you could make 12C just as efficient and have better performance? For mobile processors, Alder Lake would need to have an almost 20% improvement in power efficiency over Tiger Lake which is on the same node, just to match the efficiency of current Zen 3 processors on the market, in my experience with the latter two.
You can fit 4 small cores in the space of a single big core, so your math doesn't hold up in terms of die area and therefore cost. It would be 8C+8c versus 10C.

Efficiency isn't a single number, but a range that will differ based on the workload's intensity. Getting good efficiency out of big cores requires higher utilization. At low utilization, most of the big core is going to waste and it's consuming more power.

Little cores aren't about absolute performance, but low power for background tasks/idle - this is important for laptops and for companies that deploy thousands of computers in their offices sitting at almost idle most of the time, just displaying Excel etc. The benefit doesn't come only from less energy consumed by the computers or office A/C, but also corporate environmental requirements. It helps with greenwashing PR as well. See companies advertising that their operations are carbon-neutral - or will be, by year 202X?

Small cores an also be leveraged as additional computing power for heavily multi-threaded applications; sure they're not as good as big cores but they are 1/4 the size and consume less power at 100%, so the perf/power and perf/area efficiencies are similar enough even in that worse case scenario.

It's only if you require a few highly performant cores (e.g. gaming or Photoshop) that the little cores are mostly useless, but even then they're only useless for that specific task - they can still take care of background tasks so that more of the big cores is left for the main task.

Even if gamers don't need small cores, and would gladly give them all away for just 1 more big core, the rest of the world disagrees.

The extra cache is jist a temp. measure. It cannotnbe a solution.
Well, more cache has been AMD's solution for the Ryzen 5000 series. It's mostly due to the 32MB L3 cache present on all models that even the Ryzen 5 5600X compares so well against the i9's in gaming.

As long as Intel doesn't pull ahead massively on frequency and/or IPC or add more cache to their midrange parts, AMD gets to reap the benefit.

I'd argue that large caches are here to stay. As computing gets faster and faster, the need to feed the cores (or GPU) with data increases as well. Even if memory or memory bus technologies provide the required bandwidth, latency will still be a problem, and cache is the solution.
 
Last edited:
Reactions: JWNoctis

Makaveli

Distinguished
Jan 15, 2001
1,141
211
19,670
4
This is not the same intel as of few years ago... you're underestimating the big sleeping (now awoken) giant with new leadership and new hardware designs. It's not gonna be the same.
I will believe all of this once I see it from intel. One successful launch of ADL isn't going to make up for 6 years of basically not doing much. It will require multiple launches will have to see what raptor lake is doing after ADL etc.

Talk is cheap as they say :)
 

Soaptrail

Honorable
Jan 12, 2015
232
57
10,670
1
I dont see overwhelming success for big little outside the mobile PC industry. Actually on desktops it is a waste of silicon.
I also disagree. How many threads does windows have that do not need to run at 4GHz or more? If you have dedicated little cores where the heat generated is not as high you could get more output from big cores so it could be a win win but windows has to put the right threads to the right cores in order to get that benefit. AMD and Intel can stick little cores between big cores to help with heat distribution as well.
 
I also disagree. How many threads does windows have that do not need to run at 4GHz or more? If you have dedicated little cores where the heat generated is not as high you could get more output from big cores so it could be a win win but windows has to put the right threads to the right cores in order to get that benefit. AMD and Intel can stick little cores between big cores to help with heat distribution as well.
Intel has speed step and C states and whathaveyou for a decade if not more by now, nothing backgroundy is running at 4Ghz unless you have a locked 4Ghz all core overclock, modern cores run this stuff at just a few hundred Mhz, differences should be very small here.
The little cores are just there because they are cheaper than full cores and still give a pretty big performance boost in productivity without breaking the bank in power draw.
Anything else is PR and should only have very little benefit, like maybe in idle power state the smaller cores might use a bit less energy than the bigger cores but it should only be a very minor amount.
 

hannibal

Distinguished
Interesting to see if Zen4 have v-cache and normal versions from release or do they release v-cache versions later just they will do with zen3...
Also interesting so see how much more v-cache versions cost compared to normal version. 50% more? It has to be substantial, because only two cpu chip versions aka 5900 and 5950 are getting v-cache versions for zen3. Most likely the same thing with zen4.
 

D1v1n3D

Honorable
May 8, 2015
7
1
10,515
0
Did you guys miss that he hinted at not only a 3d cache improvement for the 50xx refresh He briefly talked about higher frequency as well. I am willing to bet we are getting 5 plus Ghz on these refreshes would also be the reason why it was pushed back to 2022 instead of end of month next month launch like leaks suggested to bin the new b2 revisions for faster turbo boosts. They have to do more than just 15% to keep up with or beat Intel's next gen that is showing to be pretty damn decent in most facets also as code gets mature for these big little chips it's only going to get better and AMD has to count on that. 5900x and 5950x already boost over 5ghz with PBO with adequate cooling, with better bins we could see 5.1 5.2 5.3ghz i mean shoot the 5950x with LN hit 6362Mhz while others were able to get 5.2ghz on 5900x without LN how stable who knows but AMD can do this with a 6nm refresh and 3d vcache.
 
Oct 8, 2021
20
3
15
0
You need to note that laptops are outselling desktops 2:1 today. This means around 66% of the market share are laptops, only 33% are desktops.

And then, it matters for desktops too. Power consumption is not an issue for home users but its still an issue for corporates. When you have thousands of desktops, higher power conusmption means more money has to be spent on power and cooling. If you could reduce power by just 10W per PC, 1000PC equates to 10,000W (10KW), thats a big difference. ITs also a big difference for your air conditiong.
Intel already has low voltage CPU for laptop . they can big little them as tey wish and keep the desktop CPU big only ...

as for corporates , they can use mobile CPU desktops to save energy with big little if intel wishes . no need to bring big little into the powerful desktops.
 
Oct 8, 2021
20
3
15
0
True, but isn't the world moving towards mobile?
The most recent numbers show a 54%, market share for mobile computers vs 43% for Desktop computers (see https://gs.statcounter.com/platform-market-share/desktop-mobile-tablet).

Many people do audio and video editing on their laptops. That was unheard of 10-15 years ago.
Of course, VFX artists and professionals will still prefer a full blown workstation, but more and more work is being done on notebooks nowadays.
I agree , but the big little then sould be in the mobile CPU only not high end desktops. that is the lower Voltage variants made for mobile devices
 

ASK THE COMMUNITY

TRENDING THREADS