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!

 

[ryanjagtap]

1
getting the most performance per watt out the box

I have not seen any Intel T-series processor in a retail box (DIY) or on any online site without it being a used SI/OEM part. It's not like we could buy that processor brand new without buying a Dell/HP/Lenovo system.

 

[Just Benching]

I have not seen any Intel T-series processor in a retail box (DIY) or on any online site without it being a used SI/OEM part. It's not like we could buy that processor brand new without buying a Dell/HP/Lenovo system.

You are not in europe i assume? Cause there are aplenty here

 

[szrpx]

1
getting the most performance per watt out the box

What does it matter if it's "out of the box"? It's basically Intel setting the power limit for you before you boot the system. Power efficiency is performance at a given power consumption, and I am not convinced the 7950x will be bested there at those power ranges.

Also, Intel's "T" skus still boost to much higher power than the TDP for a set duration, before falling to the actual TDP.

See the 12900T for instance:

 

[Just Benching]

What does it matter if it's "out of the box"? It's basically Intel setting the power limit for you before you boot the system. Power efficiency is performance at a given power consumption, and I am not convinced the 7950x will be bested there at those power ranges.

Also, Intel's "T" skus still boost to much higher power than the TDP for a set duration, before falling to the actual TDP.

See the 12900T for instance:View attachment 67997

Thats my point, out of the box doesnt matter. That's why efficiency comparisons should be done at iso wattage. The guy in the previous page disagreed, thats why i brought the 13900T as an example.

Therefore you agree with me, the comparisons posted with a 7950x at 240w vs a 13900k at 350w are irrelevant when it comes to efficiency.

Also the pl2 numbers are irrelevant as well. When talking about efficiency, we are interested in hourlong workloads. Nobody really cares about consumption for 50 seconds, but for 50 minutes or 50 hours, and in those workloads the 13900t will for sure beat everything zen 4 out of the box

 

[PJVol]

That will make it the most efficient cpu on planet Earth

thats why i brought the 13900T as an example

And your example is actually correct, 13900T may very well be the most efficient СPU.
If you want to compare energy efficiency for two chips it makes sense to do just that, by estimating the amount of energy spent by each CPU to perform some reference task, e.g Blender 3.3 or Handbrake

 

[PJVol]

Power efficiency is performance at a given power consumption

So you consider power consumption to be constant during most of workloads?

 

[szrpx]

So you consider power consumption to be constant during most of workloads?

Probably not constant, but a MT workload that is steady state, say looping cinebench performance at 65 W locked, as an example. I know a lot of "MT" tasks change between single and nT threads from start to completion.

 

[PJVol]

Probably not constant, but a MT workload that is steady state, say looping cinebench performance at 65 W locked, as an example. I know a lot of "MT" tasks change between single and nT threads from start to completion.

It may or may not be steady, and Cinebench is not very suitable due to its medium workload activity.
And yet, I don't quite understand the use of power to evaluate energy efficiency (see my response to a guy insisting on unified power cap)

 

[szrpx]

It might be, though Cinebench is too short imo.
And yet, I don't quite understand the use of power to evaluate energy efficiency.

Well, the performance in the application (score, time to complete, etc.) versus the energy consumed (in joules). Doing this for a range of power limits lets you figure out the most efficient power range for that specific task.

Say, cinebench score per joule consumed. More efficient CPUs (for this task) should have a higher cinebench score for every joule they use. For tasks that measure time to completion it's not as straightforward, but it's the same idea.

Chips and Cheese has some great data on this when they did an article going over Alderlake's efficiency (Goldencove and Gracemont).

Alder Lake’s Power Efficiency – A Complicated Picture

Reviews across the internet show Alder Lake getting very competitive performance with very high power consumption. For example, Anandtech measured 272 W of package power during a POV-Ray run. Our o…

chipsandcheese.com

 

[PJVol]

Doing this for a range of power limits lets you figure out the most efficient power range for that specific task.

It does, but it goes beyond the scope of comparing the energy efficiency of two architecturally different cpus, which is evaluated at their fused limits first.

More efficient CPUs (for this task) should have a higher cinebench score for every joule they use

Yep, the tests you linked show when less power is not equal efficiency.

With a vectorized load, Golden Cove finishes the task so much faster that it ends up using less total energy than Gracemont, even though Gracemont draws less power

I'd suggest taking a look at THG's approach in their Zen3 and ADL reviews. IMHO it better covers the arch-specific nuances of power management for CPUs from both vendors. They plot 2D chart for the energy against time, where the closest to the bottom-left corner cpu is more efficient.

PS: funny thing is they referred to an article claiming RAPL measurements sufferes accuracy, whereas in the same article authors acknowledged that Zen2 internal energy telemetry is pretty accurate.

 

[lobz]

Dolphin Emulator as well, but not for the same reasons, check this out:

https://twitter.com/x/status/1567129743077638145

Low? It's a (up to) 4,6GHz P core (GB ST lines up very well to 12400's low 1600s) and 3,3GHz E cores (lower than any 12th gen). With a PL2 of around 117W.

You estimated a PL2=117W (or around that) Alder Lake to be within 2% of a PL2=150W Alder Lake? Really? The clocks on their own should be a dead giveaway that 13400 is not = to 12600K, let alone the much more modest PL2. 7600X will eat it for breakfast. Especially in lightly-threaded tasks and gaming.


I see no one is looking at the most obvious culprit of all from that GB run. 13400 seems to be running in single channel mode (RAM reported as 8GB). But GB isn't the most trustworthy in reporting system specs, so who knows.
Either way, I don't think this SKU will crack 10K MT points, even under a more optimal setup. People expecting it to cost the same as 12400 should prepare for disappointment, in about 3 months or so.

Some developer this dude must be..... it was Dolphin's own Twitter account that wrote about themselves: 'such an IPC focused load' and not AMD..... 🤦

 

[Henry swagger]

People arguing in this thread about efficiency.. wait for reviews from hardware unboxed and gamers nexus for the trye numbers from both companies

 

[Markfw]

People arguing in this thread about efficiency.. wait for reviews from hardware unboxed and gamers nexus for the trye numbers from both companies

ahhh. NOT my favorite review sites. How about from 5-10 different review sites that are respected. and NOT videos !!! I want slides that I can link to and things I can read.

and what are "trye numbers" ??

 

[Kaluan]

12900K@50W gets demoloshed in performance by similarly 'power hungry' 5600X system @ stock...

This whole conversation about 13900T@35W being some efficiency one-hit-wonder by Intel really bores me and it's really scraping the barrel of straw grasping...

That SKU will be absolutely STARVING at 35W, and even when it's when using PL2 boosting (106W+), which most higher end mobos can sustain indefinitely with one click in the BIOS, it'll still be severely underfed.

Low power Zen4 SKUs or regular ones in ECO mode will crush them in efficiency. Hell even stock X parts probably will as well.

 

[Henry swagger]

ahhh. NOT my favorite review sites. How about from 5-10 different review sites that are respected. and NOT videos !!! I want slides that I can link to and things I can read.

and what are "trye numbers" ??

Lol.. i was meaning " true numbers from reviewers "

 

[tamz_msc]

This one is very close to the 2,000 Integer Score at stock with DDR6000 RAM
View attachment 67993


It's within 5% of the best integer score that the 13900K has posted so far

12900KS can do ~1900 integer score with 6000-6400 MT/s memory.

ASUS System Product Name - Geekbench

Benchmark results for an ASUS System Product Name with an Intel Core i9-12900KS processor.

browser.geekbench.com

 

[Just Benching]

12900K@50W gets demoloshed in performance by similarly 'power hungry' 5600X system @ stock...

This whole conversation about 13900T@35W being some efficiency one-hit-wonder by Intel really bores me and it's really scraping the barrel of straw grasping...

That SKU will be absolutely STARVING at 35W, and even when it's when using PL2 boosting (106W+), which most higher end mobos can sustain indefinitely with one click in the BIOS, it'll still be severely underfed.

Low power Zen4 SKUs or regular ones in ECO mode will crush them in efficiency. Hell even stock X parts probably will as well.View attachment 68007View attachment 68008

This review is obviously flawed, it is easy to tell, you cant have the 12600k being more efficient than the 12900k at same wattage, lol. Read the comments on the review for more information

No wonder people think amd is more efficient with reviews like these

 

[DrMrLordX]

This review is obviously flawed, it is easy to tell, you cant have the 12600k being more efficient than the 12900k at same wattage, lol.

Why not? e-cores aren't really more power efficient than p-cores. They're just more area efficient.

 

[Just Benching]

Why not? e-cores aren't really more power efficient than p-cores. They're just more area efficient.

Because the 12900k has more p and e cores,it cannot be less efficient than the 12600k at any wattage. Thats like the 5600x being more efficient than the 5950x at iso wattage. Aint happening.

Lots of reviews show the 12900k doing 24k cbr23 at 125w. Tpup has it at 18k. Obviously flawed. You can ask anyone with a 12900k if you dont trust me to run at the tpus power limits. They will you scores that are up to 70-80% bigger than the ones TPU has.

Only the 5950x can beat the 12900k in mt efficiency, and that's only at high wattages. If you drop to like 70w or less the 12900k wins there as well.






Enough of your Intel in the Zen 4 thread.
When you get back, this is the thread you need to post in.

Page 86 - Question - Raptor Lake - Official Thread

Page 86 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

esquared
Anandtech Forum Director

 

[DrMrLordX]

Because the 12900k has more p and e cores,it cannot be less efficient than the 12600k at any wattage.

Unless the 12900k is binned for higher leakage.

Regardless it looks like Raphael will be more efficient than Alder Lake or Raptor Lake @ isopower, no matter how much you may protest. Hopefully someone will think to test for that, not that you're likely to accept the results.

 

[szrpx]

Because the 12900k has more p and e cores,it cannot be less efficient than the 12600k at any wattage. Thats like the 5600x being more efficient than the 5950x at iso wattage. Aint happening.

Lots of reviews show the 12900k doing 24k cbr23 at 125w. Tpup has it at 18k. Obviously flawed. You can ask anyone with a 12900k if you dont trust me to run at the tpus power limits. They will you scores that are up to 70-80% bigger than the ones TPU has.

Only the 5950x can beat the 12900k in mt efficiency, and that's only at high wattages. If you drop to like 70w or less the 12900k wins there as well.






Enough of your Intel in the Zen 4 thread.
When you get back, this is the thread you need to post in.

Page 10 - Discussion - Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads

Page 10 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

esquared
Anandtech Forum Director

I'll just post this here. If you can find a 12900K doing a similar score at same power, I'll happily admit you're right.

Page 340 - Discussion - Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)

Page 340 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[TESKATLIPOKA]

Videocardz: AMD Ryzen 9 7950X overclocked to 6.7 GHz scores 50K in Cinebench R23

Scores 48235 in CB R23 at 6.45GHz with power consumption of only 271W.
Power consumption is very low for that clockspeed.

or

Scores 50395 in CB R23 at 6.7GHz with unknown power consumption.

 

[Henry swagger]

Videocardz: AMD Ryzen 9 7950X overclocked to 6.7 GHz scores 50K in Cinebench R23

Scores 48235 in CB R23 at 6.45GHz with power consumption of only 271W.
Power consumption is very low for that clockspeed.

or

Scores 50395 in CB R23 at 6.7GHz with unknown power consumption.

Impressive zen 4 threadripper will be insane.. a 32 core zen 4 can do 110k easily

 

[Kaluan]

Videocardz: AMD Ryzen 9 7950X overclocked to 6.7 GHz scores 50K in Cinebench R23

Scores 48235 in CB R23 at 6.45GHz with power consumption of only 271W.
Power consumption is very low for that clockspeed.

or

Scores 50395 in CB R23 at 6.7GHz with unknown power consumption.

Lines up in CB points AND clocks (vs the 40,4K points AIO ~5,4GHz leak we had earlier) pretty much linearly, but that was expected. The max wattage is... let's just say... interesting. I wonder what's happening there. I know silicon reacts differently under exotic cooling but... what?

 

[Asterox]

Videocardz: AMD Ryzen 9 7950X overclocked to 6.7 GHz scores 50K in Cinebench R23

Scores 48235 in CB R23 at 6.45GHz with power consumption of only 271W.
Power consumption is very low for that clockspeed.

or

Scores 50395 in CB R23 at 6.7GHz with unknown power consumption.

Lol, only 270W with 6.45ghz all core for 48000 score in CB R23.

Hm if legit, that is far crazy than Klingon overclocking CPU.

i9 13900K needs 350W for 40000 in CB R23.