Discussion Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)

[Vattila]

Except for the details about the improvements in the microarchitecture, we now know pretty well what to expect with Zen 3.

The leaked presentation by AMD Senior Manager Martin Hilgeman shows that EPYC 3 "Milan" will, as promised and expected, reuse the current platform (SP3), and the system architecture and packaging looks to be the same, with the same 9-die chiplet design and the same maximum core and thread-count (no SMT-4, contrary to rumour). The biggest change revealed so far is the enlargement of the compute complex from 4 cores to 8 cores, all sharing a larger L3 cache ("32+ MB", likely to double to 64 MB, I think).

Hilgeman's slides did also show that EPYC 4 "Genoa" is in the definition phase (or was at the time of the presentation in September, at least), and will come with a new platform (SP5), with new memory support (likely DDR5).

What else do you think we will see with Zen 4? PCI-Express 5 support? Increased core-count? 4-way SMT? New packaging (interposer, 2.5D, 3D)? Integrated memory on package (HBM)?

Vote in the poll and share your thoughts!

 

[BorisTheBlade82]

Most people do not seem to realize just how energy efficient Alder Lake is, if - and only if - it is locked at 125w TDP or less. This is presumably the range it was designed for before some marketing people decided to push it beyond sanity in order to win some benchmarks against the 5950X.

In CB23 MT the comparison is as follows:

• The 12900K uses 4.469 Joules per run, while the 5950X uses 4.149 Joules and the 5900X uses 5.274 Joules. So the 12900K is almost as energy efficient as the 5950X.
• In terms of performance it is also not that bad - only about 10% slower than the 5950X. But Intel seemingly could not live with that.
• In ST Alder Lake is the clear winner in terms of both performance and efficiency - hands down. This is due to the large energy overhead of the IFOP interconnect of Zen.
• To be fair the 5900X and 5950X have a PPT of 143w. Locked at 125w they would be a bit more efficient in MT as well.

In terms of RTL vs. Zen4 IMHO this is going to be a close call. If both are locked at 125w then RTL might win everything - efficiency and performance, ST and MT.

i9-12900K @ 125w

R9 5900X @ Stock


R9 5950X @ Stock (only one run, not 10 minutes)


P.S.
If you want to read more about the measurements above and a lot more numbers, please have a look at this thread: http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...y-of-x86-cpu-architectures.2597905/?view=date

 

[nicalandia]

lol @ non-descript AMD 32T model. Trying to leave the V-cache variant as the ace up your sleeve, eh?

Not at all.. I was talking about the Vanila 7950X

The 7950XD3D have been said that it will have a 30% lead(in gaming and memory sensitive apps) according to leaks.

AMD Ryzen 9 7950X3D, Ryzen 9 7900X3D, and Ryzen 7 7800X3D gaming performance could eclipse the vanilla Zen 4 parts by up to 30%

Ryzen 7000 CPUs with 3D cache are on their way. Rumors have it the processors will launch early next year with gaming performance that eclipses the Ryzen 7000 non-3D chips. So how much difference will the Ryzen 7000X3D SKUs make? Per RedGamingTech, the Zen 4 X3D SKUs can be up to 30% quicker...

www.notebookcheck.net

 

[Saylick]

Most people do not seem to realize just how energy efficient Alder Lake is in MT, if - and only if - it is locked at 125w TDP or less. This is presumably the range it was designed for before some marketing people decided to push it beyond sanity in order to win some benchmarks against the 5950X.

I think we're aware, and the Chips and Cheese article really brought it to light, which begs the question: if you wanted to compare Zen 4 with Raptor Lake, what TDP do you choose? It seems like there is an optimal working range that suits each CPU and this sweet spot may not be the same between both CPUs.

It's almost like comparing perf/W on a smaller, newer node GPU that is clocked higher than it's optimum range vs. a larger, older node GPU that you downclock so that its TDP matches the former part. At some point, even a large GPU on an older node looks power efficient when you downclock it because it is generally better to have a wider design run slow. E-core spam is literally that wider design.

 

[igor_kavinski]

In CB23 MT the comparison is as follows:

The 12900K uses 4.469 Joules per run, while the 5950X uses 4.149 Joules and the 5900X uses 5.274 Joules. So the 12900K is almost as energy efficient as the 5950X.

Only problem with your method of deducing energy usage is employing CB23 exclusively. Almost no one does that in real life. What you need is a comprehensive set of popular and everyday-use scriptable applications/games and then do complete side by side comparison of the energy usage.

Other than that objection, I think you've done a great job.

 

[Saylick]

Not at all.. I was talking about the Vanila 7950X

The 7950XD3D have been said that it will have a 30% lead(in gaming and memory sensitive apps) according to leaks.

AMD Ryzen 9 7950X3D, Ryzen 9 7900X3D, and Ryzen 7 7800X3D gaming performance could eclipse the vanilla Zen 4 parts by up to 30%

Ryzen 7000 CPUs with 3D cache are on their way. Rumors have it the processors will launch early next year with gaming performance that eclipses the Ryzen 7000 non-3D chips. So how much difference will the Ryzen 7000X3D SKUs make? Per RedGamingTech, the Zen 4 X3D SKUs can be up to 30% quicker...

www.notebookcheck.net

lol, then you simply could have said so.

 

[BorisTheBlade82]

@Saylick
I totally agree. I am in for comparing the SKUs just as the vendor supposed them to be.
But in this case we wanted to get an idea of how Raptor Lake and Zen 4 will compare. Both will have more or less the same consumption limits - 230w vs. 250w. RPL will have 8 more efficiency cores which will in turn run with a lower clock in a more efficient range.
I am placing my bets on RPL being more efficient in every regard - in spite of AMD having a node advantage. Of course I am not so sure about performance.

 

[Abwx]

In terms of RTL vs. Zen4 IMHO this is going to be a close call. If both are locked at 125w then RTL might win everything - efficiency and performance, ST and MT.

Both locked at 125W the 13900K stand no chance in perf and perf/watt, you seem to forgot the node discrepancy, expect a massive perf/watt advantage for the 7950X.

Besides your graphs are not accurate, the average power with the 5950X is taken during a single run while the 12900K is for a lot of runs between wich the CPU power is lower, and this artificialy lower the 12900K whole average.

 

[nicalandia]

@Saylick
I am placing my bets on RPL being more efficient in every regard - in spite of AMD having a node advantage. Of course I am not so sure about performance.

Since you are betting against the odds. What do yo want to bet?

 

[BorisTheBlade82]

@Abwx
Please see my post above - I am fully aware of the node advantage. I just do not think that this will weigh up the core-number-advantage and the interconnect disadvantage.

Other than that we will have to wait and see - just a couple more days
I am just hoping that some of the Reviewers I was in contact with will give PES a try when holding RPL and Zen4 in their hands.

 

[Abwx]

@Abwx
Please see my post above - I am fully aware of the node advantage. I just do not think that this will weigh up the core-number-advantage and the interconnect disadvantage.

Other than that we will have to wait and see - just a couple more days
I am just hoping that some of the Reviewers I was in contact with will give PES a try when holding RPL and Zen4 in their hands.

The core count advantage would hold if the small core had either SMT or a way higher throughput/Hz.

Throughput/Hz wise a Zen 4 core + SMT is equivalent to 2 e cores, so there s no advantage for the 13900K even if the e cores were to be clocked at same frequency than the Zen 4 cores.

Anyway thoses reviews cant come soon enough, dunno when the NDA end, should be in a week or two.

 

[deasd]

Please see my post above - I am fully aware of the node advantage. I just do not think that this will weigh up the core-number-advantage and the interconnect disadvantage.

Sometimes adding cores doesn't mean the efiiciency would be added up. Especially when the high-end SKUs are already top-tier golden silicon samples.

here's the 10900K(10C) tested by TPU, compare it to 9900KS(8C) you can see the Cinebench task energy consumption is roughly the same:




I think that using experience from old times when CPUs have a lot of OC/undervolt headroom, to speculate effciency of new CPUs which is already reach their limit and have no OC/undervolt headroom might get things wrong.

 

[desrever]

Most people do not seem to realize just how energy efficient Alder Lake is, if - and only if - it is locked at 125w TDP or less. This is presumably the range it was designed for before some marketing people decided to push it beyond sanity in order to win some benchmarks against the 5950X.

The 5950X would be way more efficient if you drop the power down to 65W, just limiting power at any given power isn't going to give you proper efficiency numbers. Zen 3 cores are probably most efficient at like 2.5W per core, even a 16 core CPU in that case would become most efficient at less than 50W. The 5950X is already running way above it's efficiency curve too.

Look at how far intel is on mobile, AMD is almost 2x the performance per watt:

 

[Saylick]

Since you are betting against the odds. What do yo want to bet?

Tradition 'round these parts says that whoever is wrong normally has to eat a whole can of Fancy Feast.

 

[Markfw]

Only problem with your method of deducing energy usage is employing CB23 exclusively. Almost no one does that in real life. What you need is a comprehensive set of popular and everyday-use scriptable applications/games and then do complete side by side comparison of the energy usage.

Other than that objection, I think you've done a great job.

One more problem. You do NOT test the 12900k at stock, but you do the 5950x. Try eco mode on the 5950x and then run it again.

 

[biostud]

You willing to bet money on it? 100$ says 13900t will wipe the 7950x in efficiency out of the box

At same wattage? Or both @ default settings, because if I remember correctly you prefer to compare efficiency when the power limit is equal?

 

[APU_Fusion]

I love the smell of astroturfing in the morning

 

[igor_kavinski]

One more problem. You do NOT test the 12900k at stock, but you do the 5950x. Try eco mode on the 5950x and then run it again.

Really good point. I wish my office would get a Dell Optiplex with i7-12700 or i9-12900. Really curious to see what PL1/PL2 settings these OEMs are using.

 

[BorisTheBlade82]

@desrever
Please don't get me wrong. Zen2 and upwards are the unquestioned efficiency leaders in the x86 space - especially the monolithic mobile versions. It is just against the Desktop versions were Intel does not fare as bad as everyone seems to think.

@Markfw
In the thread I linked there are also stock numbers for the 12900K. I just wanted to point out that the ADL architecture and Intel7 are much better than the insane 241w limit Intel applied make it seem. From a technical PoV this is highly interesting to me.

 

[nicalandia]

For the very first time in the last decade AMD CPUs will have higher rated max boost than Intel?

 

[Markfw]

@desrever
Please don't get me wrong. Zen2 and upwards are the unquestioned efficiency leaders in the x86 space - especially the monolithic mobile versions. It is just against the Desktop versions were Intel does not fare as bad as everyone seems to think.

@Markfw
In the thread I linked there are also stock numbers for the 12900K. I just wanted to point out that the ADL architecture and Intel7 are much better than the insane 241w limit Intel applied make it seem. From a technical PoV this is highly interesting to me.

But.... You need to do them both@stock or both are power savings, ECO mode for the 5950x is just that

 

[eek2121]

Most people do not seem to realize just how energy efficient Alder Lake is, if - and only if - it is locked at 125w TDP or less. This is presumably the range it was designed for before some marketing people decided to push it beyond sanity in order to win some benchmarks against the 5950X.

In CB23 MT the comparison is as follows:

• The 12900K uses 4.469 Joules per run, while the 5950X uses 4.149 Joules and the 5900X uses 5.274 Joules. So the 12900K is almost as energy efficient as the 5950X.
• In terms of performance it is also not that bad - only about 10% slower than the 5950X. But Intel seemingly could not live with that.
• In ST Alder Lake is the clear winner in terms of both performance and efficiency - hands down. This is due to the large energy overhead of the IFOP interconnect of Zen.
• To be fair the 5900X and 5950X have a PPT of 143w. Locked at 125w they would be a bit more efficient in MT as well.

In terms of RTL vs. Zen4 IMHO this is going to be a close call. If both are locked at 125w then RTL might win everything - efficiency and performance, ST and MT.

i9-12900K @ 125w

R9 5900X @ Stock
View attachment 67433

R9 5950X @ Stock (only one run, not 10 minutes)
View attachment 67434

P.S.
If you want to read more about the measurements above and a lot more numbers, please have a look at this thread: http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...y-of-x86-cpu-architectures.2597905/?view=date

Oh uh, great, remind me again, when was the 5950x released again? What? November 5th, 2020? Surely Alder Lake was released around the same time, right? A year later in 2021? Oh, but surely AMD launched a new product by then because they were clearly outmatched. What? No? 🤣 I can't.

Zen 4 is on a superior process. There is no getting around this for Intel. You can't beat the laws of physics/thermodynamics. Raptor Lake will be competitive, for sure, but it won't "stomp" anything. It will run hot and fast while consuming far more power than a Zen 4 chip. When you turn the power limits down, Zen 4 will win in terms of efficiency.

Also, no, you can't compare the 13900T or any other lower power part to a mid-range Zen 4. They are different chips targeting different segments of the market. Some of you folks just don't understand how companies like AMD, Intel, Qualcomm, and Apple work. Comparing a 13900t to a 13900k is apples to oranges, to say nothing of comparing the 13900t to a7600x. The only proper comparison of the 13900t would be to a similar AMD part, which, like the 13900t, won't be launched until next year.

EDIT:

To put it in blood:

1. The 7600x and 7700x are cost optimized. They sacrifice efficiency and performance for cost.
2. The 13900t and 13900k are performance optimized. They sacrifice cost for improved performance and efficiency.

 

[Zucker2k]

Oh uh, great, remind me again, when was the 5950x released again? What? November 5th, 2020? Surely Alder Lake was released around the same time, right? A year later in 2021? Oh, but surely AMD launched a new product by then because they were clearly outmatched. What? No? 🤣 I can't.

Zen 4 is on a superior process. There is no getting around this for Intel. You can't beat the laws of physics/thermodynamics. Raptor Lake will be competitive, for sure, but it won't "stomp" anything. It will run hot and fast while consuming far more power than a Zen 4 chip. When you turn the power limits down, Zen 4 will win in terms of efficiency.

Also, no, you can't compare the 13900T or any other lower power part to a mid-range Zen 4. They are different chips targeting different segments of the market. Some of you folks just don't understand how companies like AMD, Intel, Qualcomm, and Apple work. Comparing a 13900t to a 13900k is apples to oranges, to say nothing of comparing the 13900t to a7600x. The only proper comparison of the 13900t would be to a similar AMD part, which, like the 13900t, won't be launched until next year.

EDIT:

To put it in blood:

1. The 7600x and 7700x are cost optimized. They sacrifice efficiency and performance for cost.
2. The 13900t and 13900k are performance optimized. They sacrifice cost for improved performance and efficiency.

This post is all over the place. Comparing the 13900t to 13900k is apples to oranges? But comparing it to a future competitor's product on a different arch is an apple to apple comparison? Also, the 13900t shouldn't be compared to a mid-range Zen 4 but rather the 13900k because they're both performance optimized? What a head scratcher.

 

[yuri69]

Lock both processors to 125W (as the reasonable highend TDP) and then run the test suite. Running just the CB is quick&easy but having a comprehensive suite would be much better. For instance, The Stilt managed to run 36 benchmarks in order to evaluate Zen 3 IPC.

 

[Zucker2k]

Lock both processors to 125W (as the reasonable highend TDP) and then run the test suite. Running just the CB is quick&easy but having a comprehensive suite would be much better. For instance, The Stilt managed to run 36 benchmarks in order to evaluate Zen 3 IPC.

And efficiency shouldn't be confined to multithreaded apps. OG Zen, up to Zen 3 are not that efficient, thanks to the IOD, in single and few cores scenarios. In my opinion, typical desktop use is not embarrassingly parallel. Aside from area efficiency at the design/performance level, this is perhaps the best side effect of the hybrid system. In the case of Intel, you have 8 cores for doing all your foreground tasks that are not heavily multithreaded until you run a heavily multithreaded program and then the efficiency cores also come into play. It's a well-balanced concept and Intel is doing great with it, looking at the success of ADL on the desktop so far.

 

[yuri69]

And efficiency shouldn't be confined to multithreaded apps. OG Zen, up to Zen 3 are not that efficient, thanks to the IOD, in single and few cores scenarios.

Sure, the broader set the better - as always. Ranging from purely single threaded, including lightly threaded to HPC-like parallel tasks.

When one designs a perfect evaluation of a general purpose processor from purely engineering standpoint it would include *every* single workload (very single compatible binary/program). This can't be reached but still, the closer the better.