Incoming Ryzen with Efficiency Cores

[Kenrou]

Hybrid 12-core config of AMD Ryzen 8000 "Strix Point" APU seemingly confirmed by leaked screenshots - VideoCardz.com

Possibly the first screenshot featuring Ryzen 8000 in action A website called Performancedatabases claims to have the first screenshot featuring the AMD Strix Point APU in action. The website has shared two screenshots showcasing AMD’s alleged Ryzen 8000 mobile CPU with the codename Strix Point...

videocardz.com

01:10 - AMD's Efficiency Cores - ZEN 4c
01:30 - AMD Zen 4c vs Intel Gracemont (E-Cores)
04:13 - Ryzen 8000 APUs with Zen 5c cores
04:48 - SMT/HT & Core Density
07:37 - Conclusion

 

[EarthDog]

Now....... I didn't watch the vid, but, e-cores on an APU, I like.

 

[Kenrou]

Yep, TechJesus said the ECores are much more efficient and you can pack more of them on a die than current Intel, although their relative worth for day to day is still fairly debatable IMO, I guess it will depend on price

 

[Woomack]

E-cores are fine as long as you can use them right. I'd rather have 2-4 more P-cores than 16 E-cores and this is probably what AMD is going to do next to beat Intel in amount of threads. Right now we have Intel mobile chips with 2 P-cores and 12 E-cores or similar configurations. It's still slow and good for maybe standard home usage like browsing the web. Gamers are still turning off E-cores as scheduling is not the best and cripples the performance.
I see it as the first gen will go to mobile devices and next they will push it to higher desktop series as they already have problems with CPU wattage while the competition pushes the number of threads. The same is doing Intel, they can't compete with P-cores, so put as many E-cores as they can, keeping a similar wattage. I highly doubt that Intel would play with E-cores if they could match AMD in multithreading using P-cores at the same wattage.

 

[mackerel]

Didn't watch video but saw this "news" elsewhere before. To keep things clear, I'll call Intel's P and E cores, and AMD's C and c cores. With Intel the exchange rate is 4 E cores (without HT) equivalent size to a P core. With AMD you get two c cores trading a C core, SMT intact. Half L3/core compared to chiplets, but the APUs already have half L3 vs chiplet, so the saving might not be as good in APUs. c cores will clock lower than C cores due to design. Probably more efficient threaded in throughput too.

Given this I think AMD APUs could go 4C+4c as a 6 core successor, and either 4C+8c or 6C+4c as an 8 core successor. It would be an interesting test if someone could arrange it, just how many P/C cores do you need before filling out with E/c cores? You could make the argument only so much needs to be done at high performance, and beyond that point you're looking at thread scaling anyway. The trick that is needed to making sure those threads that do depending on performance end up on the C cores. How well does Thread Director work, and does AMD need an equivalent?

With chiplet based CPUs I'm not so sure what AMD might do there. They could go up to 8C+16c for example, but as we saw with the 2 CCD X3D chips, scheduling could get messy. At least with the APUs they'll be unified so you wont suffer the penalty from moving between CCDs.

 

[Kenrou]

The one in the screenshot is a 4C/8c + SMT, so 4C/8t/8c/16t... It's a lot of threads for 45w TDP, again, why would you want this for a budget low power build, I mean, they are budget for a reason, it means you don't really need the horsepower, right? Suppose you could just run with SMT off?

Scheduling has never really worked properly for Intel even on W11, so not expecting to work well for AMD either

 

[mackerel]

I feel a budget build can be for two reasons: 1, you really don't need that much performance, and 2, you want as much performance as you can afford, but you can't afford higher offerings. Moving up the low end is always welcome. I'd even speculate what AMD could do in the former quad core space. 2C+4c for example? Personally I've never been a big fan of HT/SMT, and have wondered as we get to higher core counts will we see it going away. While it does increase MT throughput, it comes at the cost of individual threads going slower as well as increasing scheduling complexity. Still, it is generally a net performance positive even if in most cases it is pretty small. There will be some cases where it helps significantly, and in those, users will welcome its bonus.

Since I still haven't owned a hybrid CPU, I don't have personal experience on its effectiveness. There is the rare report of a game that might perform better from turning off E cores, but it seems in most cases it works well enough. There was even testing using Win10 which is not hybrid aware, and there wasn't much perf difference compared to Win11.

 

[Woomack]

Some (mainly older) games lose 20-30% performance as without proper scheduling, they randomly pick E cores. I assume it doesn't matter for most gamers, but those who aim for the highest FPS in online/multiplayer games may want to disable efficient cores. The same if you do virtualization-oriented work, then E-cores sometimes cause problems. I had big problems with that using Hyper-V and some other things at work. I guess the biggest problems were already solved as I can run Hyper-V VMs on i5-1240P CPU, but I still can't assign all 16 threads or I see a black screen after OS installation. Before, I couldn't even install anything when E-cores were enabled.

Unless I talk about a specific architecture, then I will still use P/E cores naming as no matter if I talk about AMD or Intel, or whatever is their own naming, in general I mean performance or efficient cores.

It's clear that games have the highest FPS on the highest frequency or the best performing cores (if we look at X3D CPUs). Most games can't use all CPU threads, so gamers disable HT/SMT, second CCX or E-cores to achieve better results. Most games still use 2-4 threads to the max, some can use 6-8. Barely any game can use the full potential of 8+ cores with SMT/HT. You can see that looking at the average load of new games, and it's usually around 30-50% when the GPU runs at 100% (of course there are exceptions).
If scheduling was working perfectly then I assume that we could have so weird CPUs as 4 P cores and 12 E cores. Some of the current Intel CPUs are ridiculous. Like who needs 4 E cores when already have 4 P cores with HT. Better would be to have 6 P cores and nothing else. There were also only E-core CPUs, but I can't find them now. I guess it could be a good option for low power business solutions.

Personally I see 4 P cores as a minimum and then could be 8 E cores and it would be just right for most users. This is what I have in my new NUC and I feel it's just right, but core frequencies could be higher. At least on that NUC, I see that all cores are loaded about the same during mixed load tests. I would expect a higher load on P cores in some tasks.

I just feel that if AMD is introducing their C/c architecture in mobile computers, then it will appear in regular desktops and gaming series. Intel somehow successfully introduced their efficient cores, so AMD will probably follow. It feels like Intel is pumping top CPUs with E-cores only to beat AMD in multitasking and popular benchmarks. E-cores wouldn't be required in top i7/i9, but without it, Intel wouldn't fit into TDP and couldn't compete with AMD. Even now it looks pretty bad but results in tests are good enough to convince potential users.

 

[Kenrou]

One thing to note though, TechJesus said that in Intel the E cores are all Celeron/Pentium, so besides lower speed/cache, they also have lower IPC than the P cores, but AMD will make all cores pure Ryzen, just lower speed/cache. That should be a huge boost to horsepower (albeit somewhat lacking in the gaming department), especially if you can overclock them, either by bios or Ryzen master app?

 

[EarthDog]

TechJesus said that in Intel the E cores are all Celeron/Pentium, so besides lower speed/cache, they also have lower IPC than the P cores, but AMD will make all cores pure Ryzen, just lower speed/cache. That should be a huge boost to horsepower

Maybe. IIRC, the little cores on Intel are ~= to skylake which isn't a slouch.

 

[Woomack]

I heard that in AMD, efficient cores will be one Zen generation behind, so as long as they will be faster than Intel, then still lower IPC. Something like current APUs. I can be wrong as I saw mixed news about the same things in the last weeks.

I bet we will be able to overclock them, but the same as now, a proper motherboard and BIOS will be required. Most Intel mobile CPUs are locked and there is not even any option in BIOS to adjust settings. AMD APUs are usually unlocked, but BIOS usually doesn't have options to overclock the CPU.

 

[mackerel]

Maybe. IIRC, the little cores on Intel are ~= to skylake which isn't a slouch.

I had to look it up recently, since I too also remembered the comparison to Skylake. However looking at the pre-Alder Lake release slides the comparison they did was 2 Skylake cores (with HT) to 4 E cores (without HT). I don't know of a wide task IPC test of E cores vs Skylake which would be interesting.

I heard that in AMD, efficient cores will be one Zen generation behind, so as long as they will be faster than Intel, then still lower IPC. Something like current APUs. I can be wrong as I saw mixed news about the same things in the last weeks.

I haven't seen that claim. C and c cores are reported to be same generation, specifically same microarchitecture. So no problems that Intel has with P/E cores being different e.g. the loss of AVX-512. The differences from a user perspective is smaller L3/core (if comparing to chiplet) and lower clocks.

I bet we will be able to overclock them, but the same as now, a proper motherboard and BIOS will be required. Most Intel mobile CPUs are locked and there is not even any option in BIOS to adjust settings. AMD APUs are usually unlocked, but BIOS usually doesn't have options to overclock the CPU.

Even then, don't get your hopes up. They're understood to have used a different layout compared to "regular" cores which contribute to the density, at the cost of maximum clock. It would still be interesting to see where that limit lies.

 

[Woomack]

Even then, don't get your hopes up. They're understood to have used a different layout compared to "regular" cores which contribute to the density, at the cost of maximum clock. It would still be interesting to see where that limit lies.

It can be like with Intel E cores or can be locked/limited because of problems similar to X3D CPUs. We have to wait to find out. Intel E-cores are barely overclocking but can force them to run at the max clock, which helps in overall performance.

Tbh., I care more about good enough performance for my needs and low temps/power usage nowadays. My daily CPU is Ryzen 7600/80W and it still boosts up to 5.3GHz, which is more than enough for all new games. I have two of them, but the second one is more like a side toy. I delided it a couple of days ago and will test some stuff soon. Hopefully today Noctua cooler+mounting kit for delided AMD arrives.

 

[mackerel]

The c cores are the same logical design, but made on a different fab design rule set optimised for density. Higher density usually results in lower clock. We have seen this in the other direction with Intel 14nm. Skylake through Kaby Lake to Coffee Lake was all on nominally 14nm process, but the "+" of the time was in part reducing density to improve clocks and efficiency.

I'm still using the 7920X I bought as replacement to the 7800X. I look at current CPUs and think, due to having to drop to a consumer platform I'm not really getting an upgrade. I'd like more low thread count performance which is available, but other areas are more of a side-grade or even downgrade.

 

[EarthDog]

are more of a side-grade or even downgrade.

I think a lot of users (not named Mackerel, lol) get caught up in the numbers and marketing game. I mean, will it really break most user's hearts that they lost some PCIe lanes? Do you need more than 192GB of memory or the bandwidth from quad core+ architecture? I'd think a 7950X or 13900K would run circles around the 7920X across MOST scenarios.

 

[unsafesteagle]

That was socket 2066?
I did think highly of quad channel ddr.
System was smooth.
Surprised 12600k easily out performed the 7900x from what I could see.

 

[mackerel]

I think a lot of users (not named Mackerel, lol) get caught up in the numbers and marketing game. I mean, will it really break most user's hearts that they lost some PCIe lanes? Do you need more than 192GB of memory or the bandwidth from quad core+ architecture? I'd think a 7950X or 13900K would run circles around the 7920X across MOST scenarios.

I'm not sure I'm reading that right, either it is targeted at me or not. It sounds like it is trying to do both.

Again, what I want is more low thread count performance. Any current gen CPU will do that. It doesn't even have to be top end. Skylake-X is still Skylake at heart, and between IPC increases and more clock, current CPUs might be ball park 50% faster in that measure. Great?

Ram bandwidth has been a pain point for me since at least Haswell. Current DDR5 systems are a sidegrade to quad channel DDR4. Loss of PCIe lane count I could tolerate providing the whatever mobo I pick has sufficient functionality built in, so it isn't a major deal. Newer gen PCIe is a nice plus. But at the end of the day I ask, at what cost? To go new CPU + mobo + DDR5 ram to replace what I have is no small chunk of change. We're taking high end GPU money there. I ask myself, is there enough advantage for the cost? I'm not sold on it. I'm thinking maybe I'll hold out another two Intel generations and see what we have then.

 

[EarthDog]

I'm not sure I'm reading that right, either it is targeted at me or not. It sounds like it is trying to do both.

Wat? You're looking into it way too deep. You're one of the few who can use some of that stuff. Others see bigger number and think 'must have, bigger is better' but the reality is that it doesn't matter for an overwhelming majority of home users...

.... though it can be argued with covid and more WFH today, there are more power users where something like the HEDT platform and what it offers makes sense to more users.

 

[DaveB]

So, the AMD fan boys were very critical of Intel's efficiency cores, what will they do when AMD starts selling them?

 

[Kenrou]

TBF I still don't see the point of efficiency cores at all, Intel or AMD, maybe if they used them to replace HT/SMT? Anyone care to explain?