News AMD Allegedly Testing Hybrid Processor with Zen 4 and 4c Cores

[bit_user]

No it doesn't.
First off 105W is 145W of actual power used.

That's a graph of peak power, which is of limited relevance to a discussion of energy-efficiency. What we care about is average power. Or, better yet, Joules per workload.

Just because intels node is more mature and can withstand the punishment of overclocking much easier,

It's not necessarily a matter of the process node. In fact, that both CPUs max out near the same temperature suggests it's not that, at all. Ryzen 7000's problem is firstly down to their lousy heatspreader solution.

And secondly, it seems they have a micro-architecture that doesn't need to be clocked as high, in order to deliver near maximum throughput. Of course, we don't really know if that's a micro-architecture limitation or a bottleneck elsewhere (e.g. DDR5).

 

[TerryLaze]

That's a graph of peak power, which is of limited relevance to a discussion of energy-efficiency. What we care about is average power. Or, better yet, Joules per workload.

The average and the max are practically the same when talking about a benchmark that will run at close to 100% from start to finish...

It's not necessarily a matter of the process node. In fact, that both CPUs max out near the same temperature suggests it's not that, at all.

Look closer at the review you constantly like to show people.
Intel hits 86 degrees at 330W usage while ryzen hits 94 degrees at 215W
That's not the same at all.
https://www.anandtech.com/show/17641/lighter-touch-cpu-power-scaling-13900k-7950x/3

Following on from the temperatures, despite pulling a figure of 330.3 W under full load, the peak core temperature of the i9-1300K was 8°C lower than the Ryzen 9 7950X, which hit 94°C under full load. Given that the power figures given aligned more with the settings on the 13900K than they did on the 7950X, the drop in temperatures on the Intel processor was much better received, with 53°C at 125 W and just 39°C at 65 W.

Ryzen 7000's problem is firstly down to their lousy heatspreader solution.

Which is the way it is because they couldn't fit some SMDs in their attempt to keep the sizes the same as previous gen.
Long term support has its downsides as well.

And secondly, it seems they have a micro-architecture that doesn't need to be clocked as high, in order to deliver near maximum throughput. Of course, we don't really know if that's a micro-architecture limitation or a bottleneck elsewhere (e.g. DDR5).

AMD has a great arch for server workloads, it's a compute workhorse, it's just a shame they released it as a desktop CPU.

 

[jeremyj_83]

AMD has a great arch for server workloads, it's a compute workhorse, it's just a shame they released it as a desktop CPU.

Zen 4 is a great uArch in general and not just for servers (Golden/Willow Cove is great as well). It is able to out perform the Intel CPUs at lower TDPs. Even if you use PPT as the metric, more times than not Intel needs 105-125W TDP to be faster than AMD at 88W. That means Zen is MORE efficient but it doesn't scale as well with additional power input. This is also why Ryzen laptops tend to have longer battery life when everything else is the same.

 

[DaveLTX]

Source?

Chips and cheese.

 

[bit_user]

Chips and cheese.

It would really help if you can find the specific article with that claim, so we can review it. It's a big claim that I'd either like to know for sure, or else maybe it could be that you misinterpreted what you saw. That's why it's best if you can reply to source requests with a link (or more).

 

[bit_user]

Look closer at the review you constantly like to show people.
Intel hits 86 degrees at 330W usage while ryzen hits 94 degrees at 215W
That's not the same at all.

The part you're missing is that it doesn't necessarily say anything about the manufacturing node. It could be explained by their poor packaging solution & heatspreader.

AMD has a great arch for server workloads, it's a compute workhorse, it's just a shame they released it as a desktop CPU.

That's easy to say until the laptop reviews start coming out. Then, Zen 4's efficiency will be clear to see.

 

[TerryLaze]

Zen 4 is a great uArch in general and not just for servers (Golden/Willow Cove is great as well). It is able to out perform the Intel CPUs at lower TDPs. Even if you use PPT as the metric, more times than not Intel needs 105-125W TDP to be faster than AMD at 88W. That means Zen is MORE efficient but it doesn't scale as well with additional power input. This is also why Ryzen laptops tend to have longer battery life when everything else is the same.

And that is what you want on a desktop part why?!
You are all acting as if windows is working on CB logic, or anything a normal user would run for that matter.
People that use datacenter/server workloads on desktop are a super small niche, it's semi-professionals cheaping out on full fletched systems.

The part you're missing is that it doesn't necessarily say anything about the manufacturing node. It could be explained by their poor packaging solution & heatspreader.

Oh, ok, sure.
That's still a part of platform maturity though.
Although we do see that ZEN4 doubled the power draw they can safely use even with that packaging, you couldn't make a previous gen use that much power on normal cooling.

That's easy to say until the laptop reviews start coming out. Then, Zen 4's efficiency will be clear to see.

The people that care about how efficient CB is on laptop is going to be even smaller than the amount of people that care about it on desktop.
Idle and light loads is much more efficient on intel on desktop and that's more likely to be important for people that actually rely on their batteries to get through the day,
Nobody is not going to plug in their laptop while running a big CB render...

 

[bit_user]

The people that care about how efficient CB is on laptop is going to be even smaller than the amount of people that care about it on desktop.

I'm not talking about just Cinebench. I mean across the range. On the desktop, you're looking at chiplet overhead of the current Ryzen 7000 models, along with other things. A laptop platform cuts a lot of the overheads and mainly what's consuming power is the actual CPU and GPU cores.

Anyway, I don't want to get into a pointless speculative battle, especially when Dragon Range reviews should be appearing any day now and Phoenix should be landing within a month.

Nobody is not going to plug in their laptop while running a big CB render...

But it's not just a question of having adequate power. Even if you can plugin a laptop, the amount of Watts it's built to dissipate determines things like the size and weight of its cooling and batteries, which in turn affect the size, weight, and cost of the laptop. And if you burn a ton of power on a plugged-in laptop, fan noise can be a real issue. So, there are lots of reasons to care about efficiency even of scalable workloads, on laptops.

 

[usertests]

Chips and cheese.

It would really help if you can find the specific article with that claim, so we can review it. It's a big claim that I'd either like to know for sure, or else maybe it could be that you misinterpreted what you saw. That's why it's best if you can reply to source requests with a link (or more).

I glanced at these but I didn't see the efficiency comparison between Zen 4, Golden Cove, and Gracemont:

AMD’s Zen 4 Part 1: Frontend and Execution Engine
AMD’s Zen 4, Part 2: Memory Subsystem and Conclusion
AMD’s Zen 4, Part 3: System Level Stuff, and iGPU

 

[bit_user]

I glanced at these but I didn't see the efficiency comparison between Zen 4, Golden Cove, and Gracemont:

AMD’s Zen 4 Part 1: Frontend and Execution Engine
AMD’s Zen 4, Part 2: Memory Subsystem and Conclusion
AMD’s Zen 4, Part 3: System Level Stuff, and iGPU

There's a related efficiency-comparison of Alder Lake P and E -cores vs. Zen 2 (chiplet and monolithic).

​

​

​

Source: https://chipsandcheese.com/2022/01/28/alder-lakes-power-efficiency-a-complicated-picture/​

In the 7-zip case, both Zen 2 incarnations are more efficient than both Alder Lake cores, except when comparing the 3950X against Gracemont in the sub-2 GHz range.

However, when it comes to x264 transcoding, only the 4800H is able to maintain an efficiency lead, across the frequency range.

We can presume Zen 4 improved efficiency enough for the 7950X to maintain a lead across the board, but I'm not comfortable doing that. I would like to see the data.

 

[TerryLaze]

But it's not just a question of having adequate power. Even if you can plugin a laptop, the amount of Watts it's built to dissipate determines things like the size and weight of its cooling and batteries, which in turn affect the size, weight, and cost of the laptop. And if you burn a ton of power on a plugged-in laptop, fan noise can be a real issue. So, there are lots of reasons to care about efficiency even of scalable workloads, on laptops.

Exactly, and a 65W laptop will be a 65W laptop, this only has any relevance if you want a certain performance level at that limit, like a certain amount of CB score...for whatever reason.
What's important is to get data for the things a laptop user actually will do on a laptop, like consume content and look at emails, or whatever actual use case it might be that people will do on battery only.

 

[bit_user]

Exactly, and a 65W laptop will be a 65W laptop,

Yeah, but with a more efficient CPU, maybe someone will decide they don't need to lug around a 65 W laptop and will instead go down to a 45 W model.

What's important is to get data for the things a laptop user actually will do on a laptop, like consume content and look at emails, or whatever actual use case it might be that people will do on battery only.

Compiling code should be on that list. A lot of developers use laptops, and compiling counts as a scalable workload.

 

[jeremyj_83]

And that is what you want on a desktop part why?!
You are all acting as if windows is working on CB logic, or anything a normal user would run for that matter.
People that use datacenter/server workloads on desktop are a super small niche, it's semi-professionals cheaping out on full fletched systems.

The people that care about how efficient CB is on laptop is going to be even smaller than the amount of people that care about it on desktop.
Idle and light loads is much more efficient on intel on desktop and that's more likely to be important for people that actually rely on their batteries to get through the day,
Nobody is not going to plug in their laptop while running a big CB render...

I am not understanding your argument here. You are literally saying you don't want an efficient uArch on desktop or laptop because they are plugged in and can use more power to do less work. By your logic the original Core (Conroe) design wasn't great for desktops because it didn't scale to the same power draw as Prescott.

We see that with Golden/Willow Cove that Intel needs to throw power draw out the window in order to keep up with Zen 4. I think AMD made a mistake in raising the TDP of Zen 4 because the performance scaling falls off the cliff after a 105W TDP. However, they did the Intel model and decided to go for absolute performance for bragging rights in benchmarks. Now don't forget that Intel needs the 250W draw in order to beat Zen 4 at 105W. Most of the time Intel needed 105-125 to beat Zen 4 at 65W. Remember that a lower TDP means you don't need a huge cooling solution. That means you system can be quieter and you SAVE energy.

 

[jeremyj_83]

Exactly, and a 65W laptop will be a 65W laptop, this only has any relevance if you want a certain performance level at that limit, like a certain amount of CB score...for whatever reason.
What's important is to get data for the things a laptop user actually will do on a laptop, like consume content and look at emails, or whatever actual use case it might be that people will do on battery only.

We already know that Zen and Core behave differently at different power levels. Zen at 65W tends to be faster than Core at 65W. There is a reason that on their top end SKUs the Intel laptop chips allow 157W boost. Intel's U series now has a 55W boost which is higher than AMD's. These higher boost leads to needing different cooling solutions, usually louder fans, and less battery life. Since Zen 2 was released in laptops, the AMD laptops have tended to have better battery life when all things are equal (like have a similar setup HP Spectre).

 

[RichardtST]

I hate seeing AMD getting dragged down to Intel level. E-cores are marketing crap. If I want more useless cores on my desktop I'll plug in my phone or a raspberry pi. I don't need slow cores. I don't want slow cores. I will not pay for slow cores. Just NO. If there is one ting that is guaranteed, it is that all the important time-sensitive threads will be scheduled on the slow cores. Programmers these days can't tell their *** from a hole in the wall. So just no. I will not buy a processor with e-cores. Forget it. E-cores count as negative value.

 

[TerryLaze]

Yeah, but with a more efficient CPU, maybe someone will decide they don't need to lug around a 65 W laptop and will instead go down to 45 W.

But that decision shouldn't be based on CB unless that's the reason you buy the laptop...

I am not understanding your argument here. You are literally saying you don't want an efficient uArch on desktop or laptop because they are plugged in and can use more power to do less work. By your logic the original Core (Conroe) design wasn't great for desktops because it didn't scale to the same power draw as Prescott.

No, look above, that's my argument here.
You have to base efficiency on the things you actually do and not on some imaginary thing that some people do but you have never even touched.

We see that with Golden/Willow Cove that Intel needs to throw power draw out the window in order to keep up with Zen 4. I think AMD made a mistake in raising the TDP of Zen 4 because the performance scaling falls off the cliff after a 105W TDP. However, they did the Intel model and decided to go for absolute performance for bragging rights in benchmarks. Now don't forget that Intel needs the 250W draw in order to beat Zen 4 at 105W. Most of the time Intel needed 105-125 to beat Zen 4 at 65W.

Only if you do radical cherry picking...
On a medium cherry picking where you still only test server type workloads but at least a few more than just one you see that the 7950x is 13% more efficient at the same power draw.
Now if everything else would be the same then 10-13% would be enough to tip the scale but on lighter loads and idle AMD loses hard, and desktop users don't do server type workloads 24/7.
https://www.computerbase.de/2022-10...bschnitt_leistung_in_apps_bei_reduzierter_tdp

Remember, this is already a decent amount of cherry picking in favor of AMD.
Scroll a bit up and you will see the single and idle comparisons, where in idle the 7950x uses 255% the power the unlimited 13900k uses and in single it uses 122% .

Edit:

Remember that a lower TDP means you don't need a huge cooling solution. That means you system can be quieter and you SAVE energy.

I already showed this before, at the same TDP, heck at much higher TDP intel is is still cooler than AMD and needs much less cooling.
Intel hits 86 degrees at 330W usage while ryzen hits 94 degrees at 215W
https://www.anandtech.com/show/17641/lighter-touch-cpu-power-scaling-13900k-7950x/3

Following on from the temperatures, despite pulling a figure of 330.3 W under full load, the peak core temperature of the i9-1300K was 8°C lower than the Ryzen 9 7950X, which hit 94°C under full load.

 

[jeremyj_83]

But that decision shouldn't be based on CB unless that's the reason you buy the laptop...

No, look above, that's my argument here.
You have to base efficiency on the things you actually do and not on some imaginary thing that some people do but you have never even touched.

Only if you do radical cherry picking...
On a medium cherry picking where you still only test server type workloads but at least a few more than just one you see that the 7950x is 13% more efficient at the same power draw.
Now if everything else would be the same then 10-13% would be enough to tip the scale but on lighter loads and idle AMD loses hard, and desktop users don't do server type workloads 24/7.
https://www.computerbase.de/2022-10...bschnitt_leistung_in_apps_bei_reduzierter_tdp

Remember, this is already a decent amount of cherry picking in favor of AMD.
Scroll a bit up and you will see the single and idle comparisons, where in idle the 7950x uses 255% the power the unlimited 13900k uses and in single it uses 122% .

You do realize you have PROVEN my point that Zen 4 is MORE EFFICIENT than Golden/Willow Cove. You have shown that while doing WORK the AMD CPU is using LESS POWER to get MORE performance. That is EFFICIENCY!!!!!!! This is not cherry picking.

"AMD has cranked up its power consumption to keep Ryzen 7000 competitive, but Raptor Lake still sucks more power. Overall, the Zen 4 architecture paired with the TSMC 5nm process is more efficient than Intel's fifth revision of the 'Intel 7' process and the x86 hybrid architecture.

Raptor Lake consumes more power than Alder Lake, but it is also much faster than its predecessor, earning it some leeway. As we see in our renders-per-day measurements, the 13900K still takes a step forward with better power efficiency than the previous-gen 12900K. However, AMD still holds the advantage in all key power criteria.

Both chips are designed to run at their maximum rated thermal specification (100C for Intel, 95C for AMD) during heavy workloads, so you shouldn't be surprised to see them run at these elevated temperatures even when they are paired with the recommended 280mm AIO (or air equivalent) for Intel, or 240mm AIO (or air equivalent) for AMD. However, Intel's system consumes more power, resulting in higher thermal output.

Winner: AMD

Intel has made plenty of progress, but AMD still holds the crown of the most power-efficient chips. Not only do they suck less peak power, but they also accomplish more work per unit of power consumed. That results in an overall win in power consumption, efficiency, and thermal output, so you'll end up with a cooler and quieter system.

Neither of the chips in our AMD Ryzen 9 7950X vs Intel Core i9-13900K faceoff comes with a bundled cooler, but Raptor Lake's higher power consumption means you'll need a beefier cooler to contain the Core i9-13900K than you will for the Ryzen 9 7950X."
https://www.tomshardware.com/news/a...cooling-intel-core-i9-13900k-vs-ryzen-9-7950x

Yes the Intel chip uses less power at idle, most likely because it is a monolithic design. We already know that the I/O die for Zen consumes a fair amount of power by itself and this is exacerbated at idle when looking at power consumption numbers. No matter what the Intel chip needs more power to do the same amount of work as the AMD chip.

 

[TerryLaze]

You do realize you have PROVEN my point that Zen 4 is MORE EFFICIENT than Golden/Willow Cove. You have shown that while doing WORK the AMD CPU is using LESS POWER to get MORE performance. That is EFFICIENCY!!!!!!! This is not cherry picking.

Yes, by 13% if you only run server workloads, and if you run those 24/7 or you only start up your system to only run server workloads.

 

[jeremyj_83]

Yes, by 13% if you only run server workloads, and if you run those 24/7 or you only start up your system to only run server workloads.

It isn't just server workloads that are more efficient on Zen 4. Blender is not something that is usually run on a server. Sure you might be running x264 on a server but it is also very common on workstations or PCs. I can go on and on with the stuff that was show and they were all MORE EFFICIENT on Zen 4 and they were not just server workloads either. Unless you think that anything but video games and Office are server workloads.

 

[TerryLaze]

It isn't just server workloads that are more efficient on Zen 4. Blender is not something that is usually run on a server. Sure you might be running x264 on a server but it is also very common on workstations or PCs. I can go on and on with the stuff that was show and they were all MORE EFFICIENT on Zen 4 and they were not just server workloads either. Unless you think that anything but video games and Office are server workloads.

Do you do them for more than 10% of your time on the PC??
Because I can bet you anything you want that the average person will be using them less than that, if at all.
The only people that will be using them for more time would be semi-pros or outright server/workstations.

 

[bit_user]

I think AMD made a mistake in raising the TDP of Zen 4 because the performance scaling falls off the cliff after a 105W TDP. However, they did the Intel model and decided to go for absolute performance for bragging rights in benchmarks.

Don't forget that these were the X models. Look at the non-X models, and you'll see their lower limits result in much better efficiency numbers. We should further expect to see the APUs doing even better on efficiency, as has historically been the case.

Sadly, none of the reviews I found measured actual power consumption. However, ComputerBase has a nice performance comparison tool to let you see how much performance you lose with the lower limit. At the end, they summarize it in the following table:

Model	TDP/PPT	Multi-core performance	Single core performance	gaming performance
Ryzen 9 Non-X vs X
Ryzen 9 7900X	170/230 watts	100%	100%	100%
Ryzen 9 7900	65/88 watts	89%	97%	98%
Ryzen 7 Non-X vs X
Ryzen 7 7700X	105/142 watts	100%	100%	100%
Ryzen 7 7700	65/88 watts	93%	95%	97%
Ryzen 5 Non-X vs X
Ryzen 5 7600X	105/142 watts	100%	100%	100%
Ryzen 5 7600	65/88 watts	93%	95%	97%

Source: https://www-computerbase-de.transla..._sl=de&_x_tr_tl=en&_x_tr_hl=en&_x_tr_pto=wapp
​

In contrast, their article covering non-K Raptor Lake models lacks a nice summary table, but it does have both an interactive table of Power Consumption data and the same interactive performance comparison tables as the above review. This enables direct comparisons, on both fronts, with both K-series Intel processors and both X and non-X AMD CPUs.

https://www-computerbase-de.translate.goog/2023-02/intel-core-i5-13500-test/?_x_tr_sl=de&_x_tr_tl=en&_x_tr_hl=en&_x_tr_pto=wapp

It seems quite limited in the number of test cases involving artificial power limits, but you can at least use it to compare power consumption and performance at stock settings.

 

[bit_user]

But that decision shouldn't be based on CB unless that's the reason you buy the laptop...

I didn't say it should. You're the one injecting Cinebench into this discussion of laptops.

Only if you do radical cherry picking...
On a medium cherry picking where you still only test server type workloads but at least a few more than just one you see that the 7950x is 13% more efficient at the same power draw.
Now if everything else would be the same then 10-13% would be enough to tip the scale but on lighter loads and idle AMD loses hard, and desktop users don't do server type workloads 24/7.
https://www.computerbase.de/2022-10...bschnitt_leistung_in_apps_bei_reduzierter_tdp

You talk about cherry-picking, but then you choose to focus on K vs. X, which are both being boosted above their efficiency window.

If you want to have a more serious discussion of efficiency, then we should look at something like i5-13500 vs. 7700. In that match-up, the AMD processor beats it by 1% in multi-core and 5% in single-core Leistungsrating (Performance Rating). Both CPUs have a TDP of 65 W, but the Intel model has a PL2 of 154 W and the AMD model has a PPT of 88 W.

When you look only at X- and K-series, then it's basically a contest of who can boost the best/highest. However, that's not relevant for laptops. In that case, what we care about is who scales down better. And, from the look of it, it's Zen 4. Again, we'll find out soon enough.

Scroll a bit up and you will see the single and idle comparisons, where in idle the 7950x uses 255% the power the unlimited 13900k uses and in single it uses 122% .

Since this is an article on laptop processors, we should consider how much better AMD's G-series APU do on idle power. Expand the graph and you should see the 5700G idles more efficiently than any Alder Lake CPU in the chart. That suggests that AMD should have no problem getting idle power under control, when it comes to laptops.

I already showed this before, at the same TDP, heck at much higher TDP intel is is still cooler than AMD and needs much less cooling.

That's not an argument about computational efficiency or process node. Again, you're just highlighting the thermal bottleneck in AMD's packaging solution.

 

[bit_user]

It isn't just server workloads that are more efficient on Zen 4. Blender is not something that is usually run on a server.

Don't be drawn into a pitched battle between K vs. X models. It's a classic misdirection tactic. He's lured you onto the terrain he feels is most defensible.

This is an article about laptop CPUs and the relevant point is how well these cores scale down, not up. Never forget the context.

 

[TerryLaze]

I didn't say it should. You're the one injecting Cinebench into this discussion of laptops.

Nobody, including you, brought up any efficiency benches that weren't about CB or 3d render, video transcode and so on.

You talk about cherry-picking, but then you choose to focus on K vs. X, which are both being boosted above their efficiency window.

If you want to have a more serious discussion of efficiency, then we should look at something like i5-13500 vs. 7700. In that match-up, the AMD processor beats it by 1% in multi-core and 5% in single-core Leistungsrating (Performance Rating). Both CPUs have a TDP of 65 W, but the Intel model has a PL2 of 154 W and the AMD model has a PPT of 88 W.

In your previous post you mentioned actual draw versus artificial power limits but now you use the artificial power limits as an argument...
They didn't have the 13500 but the 13400 has an PL2 of 148 just 6W below the 13500 and it barely goes above 70W
(And no, the 7700 doesn't hit its max power either)
https://www.techpowerup.com/review/intel-core-i5-13400f/21.html

That's not an argument about computational efficiency or process node. Again, you're just highlighting the thermal bottleneck in AMD's packaging solution.

That's your opinion...
And even if it is due to the packaging it still is what it is.

Also this is what I was replying to, it had nothing to do with computational efficiency or process node at all.

Remember that a lower TDP means you don't need a huge cooling solution. That means you system can be quieter and you SAVE energy.

 

[TerryLaze]

Don't be drawn into a pitched battle between K vs. X models. It's a classic misdirection tactic. He's lured you onto the terrain he feels is most defensible.

This is an article about laptop CPUs and the relevant point is how well these cores scale down, not up. Never forget the context.

You know what's important for laptops?!
Having a super efficient iGPU that can do hardware acceleration on your browsing/ media consumption and face time or can stream video to the net, one that also has basic AI features to sort pictures and apply filters on them without having to use the cpu cores or the power hungry dGPU.
Where are the benches on all that?