But, will the overall upgrade cost be worth the jump in performance.
For those already running a 7 series probably but maybe not for those on AM4 (5800X3D)
I'm extremely skeptical about this hybrid design. These dual CCD CPUs have one problem already, which is increased inter-chip latency between different CCDs and scheduler has to take this into account. And now we are adding another level of decisions on top of that, depending on some arbitrary cache vs frequency decisions, which imho will never work 100%. Scheduler will never be able to predict whether some process will benefit from more cache at some point. Even like 1kb executable can benefit from 50MBs of cache and vice versa. It's similar to Intel's e-cores. There will always be situations, where the system will run suboptimal.
Because of this I am looking more to the 7800X3D rather than the above models, which in theory could run better assuming there isn't latency spides, do wonder if the cache on one side is going to make this worse.
Though going form a 8700k to 7800X3D I am guessing will be such a huge jump looking at a few more % increase in a few cases just won't matter
Glad power looks to be more managed this time around as well as the 7000 series seemed a little power hungry (eco mode seems to be a game changer i've heard)
Is this cost saving on the 79xx3d ? On one hand you get a ton of more cache and a lower clock speed on one cc'd and on the other you can boost buisness as usual .... So in theory single core oriented applications that just need speed will not loose from the lower clocks and giving more weight on the non 3d ccd and when an application benefits from the increased cache will be using the 3d ccd , in theory that can be best of both worlds witha huge .... Enormous if ! IF the windows scheduler does good scheduling nwitch is a big if.
If I would have to take a guess is that this is cost saving measure first .... And trying to make the best out of it later.
There's only one thing i don't understand and thats the 7800X3D is a single ccx compared to the 7900 and 7950 with 2 ccx's so how can they all be 120w.
This makes no sense to me. I would of thought the 7800X3D would be a 95w or less because of the 1 ccx.
There's only one thing i don't understand and thats the 7800X3D is a single ccx compared to the 7900 and 7950 with 2 ccx's so how can they all be 120w.
This makes no sense to me. I would of thought the 7800X3D would be a 95w or less because of the 1 ccx.
a few info from someone who has both a 5950x and 7950x and a 13900k
- I repeat this often I don't know what this TDP is supposed to be but it's nothing real, I had a 3960x too and it was the ONLY cpu ever were tdp with stock settings (no motherboard oc) corresponded AMD sold it at 280W and it did 280.0W irl it never happened again my 7950x full stock goes up to 220W and the 13900k goes to 350W stock and 380W oc let's be clear ONLY the cpu not the whole pc we are at 2x or more of advertised TDP do not think those are 120-150w cpus they aren't (5950x was 188w with clock tuner ryzen tweaking for a 29'112 cbr23)
- a few things people don't know or realize with AMD CCD intel behaves like a single cpu with smaller cores inside, amd is really like a dual cpu
- the upside of AMD ? using cpu affinity or process lasso you can dedicate CCX1 to games or whatnot and CCX2 to obs or whatever and they will almost not interfere with each other, intel e-cores are much more linked to your main 8 cores, 13th gen improved a lot on that aspect tough I can now dedicate background apps to them, it didn't really work on 12th gen it gave me weird stuttering in games, do this daily on 13900k didn't notice fps drops and problems
- each CCX can and should have different clocks and voltage
- the second chiplet is always slower and worse quality than the 1st one I just read someone "I want to run max clocks on all ccx" nope not going to work it usually needs to run 100-150Mhz less if you can't do that it'll limit your entire cpu assembly (why clock tuner ryzen was so great, don't think there's a 7000x version yet) I'm not saying this...ryzen master is, it gives your core ratings and you'll notice CCX2 are all lower
you can kind of see it here, this is a full stock amd settings 7950x
I logged the top clocking cores here (all on CCX1) and the "worst" the 5505MHz one was on CCX2 they all had clocks around that roughly 200Mhz less than CCX1 if you tried to old school "all core overclock" you would be limited to CCX2 max clocks to be stable, for me you should really treat them as two different cpus
@kakiharaFRS
I also noticed stuttering on 12th gen, I have a 13700kf now and no problems as of yet. I would like to add this as well about intel CPUs with 9th-12th gen I would make a "gaming profile" in the bios and I would disable HT or E core or both and would see an improvement in gaming performance. Now with 13th gen if I disable E cores my gaming performance goes down a bit. If they went 16 P cores it would bet pulling almost 500watts at 5.3ghz all core.
I'm extremely skeptical about this hybrid design. These dual CCD CPUs have one problem already, which is increased inter-chip latency between different CCDs and scheduler has to take this into account. And now we are adding another level of decisions on top of that, depending on some arbitrary cache vs frequency decisions, which imho will never work 100%. Scheduler will never be able to predict whether some process will benefit from more cache at some point. Even like 1kb executable can benefit from 50MBs of cache and vice versa. It's similar to Intel's e-cores. There will always be situations, where the system will run suboptimal.
back on the pre 5000 chips, with the 4 core complexes, there were problems with windows bouncing threads across caches, The CPPC2 prefered cores feature, allows the os to rank cores by performance level, normally this would be used for primarily power saving, or peak performance with few threads, by default windows bounces threads around for thermal distribution reasons , so in this case it will bounce a single thread process between the best 2 cores with cppc2 on, this presents a problem for amd, since the best cores may not be on the same ccd, so what they did instead is sort of abuse the CPPC2 feature by assigning values based on the silicon quality, locality and some other heuristics , rather than just the highest clocking/ most efficient cores, this is done at boot time done by the smu, With the point of prevent threads from bouncing between caches, ryzen 5000 does this as well with all of the best cores being assigned on one die even if there are better cores on the other one, so lightly threaded applications won't bounce them to the other die. this tends to work really well because basically no applications using 6-8 threads will ever have cache locality problems, and applications that can use more threads are not likely to be adversely affected by dual caches in a way where the tradeoff of chiplets is not worth it.
The same thing is done with ryzen 7000 where all of the best cores are assigned to one die, I imagine the same thing will be done for the x3d chips, the question being how will they deal with the clockspeed vs cache performance differences. IMO its likely that they will just "eat" the difference assigning all of the best cores on the vcache die, this would make the most sense, The gut feeling I get is that the loss of clockspeed will be sufficiently offset by the cache, that there will be a negligible difference in performance, regression being rare and <5%. if you look at how the ryzen 7000 chips behave in thermally limited situations , it doesn't tank your clockspeeds, so its hard to say how much frequency will be lost for the lower thermal headroom, could be as low as 100-200mhz in light work loads.
It is also possible amd could change the cppc assignment dynamically, achieving something similar to what intel does with the thread director, but that might be a bit tricky.
But, will the overall upgrade cost be worth the jump in performance.
For those already running a 7 series probably but maybe not for those on AM4 (5800X3D)
I think anyone with a 5800X3D right now is good until the next generation of AMD/Intel cpus.
They are still very competitive, and for the price/performace hard to beat.
I think anyone with a 5800X3D right now is good until the next generation of AMD/Intel cpus.
They are still very competitive, and for the price/performace hard to beat.
Agreed and im glad i replaced my 3800X with one, i suspect the 7***X3D is the next evolution rather than the revolution.
Right now, as it is, for me personally, the only AMD CPUs I'd buy are either the 7800X3D, or the 5800X3D.
Singe CCD, for gaming use only, no BIOS issues, no cross cash latency issues as far as I've understood, no Windows schedular sorcery needed to run.
Right now, as it is, for me personally, the only AMD CPUs I'd buy are either the 7800X3D, or the 5800X3D.
Singe CCD, for gaming use only, no BIOS issues, no cross cash latency issues as far as I've understood, no Windows schedular sorcery needed to run.
Check out the 4k gaming results for the 7900 non X compared to the cheaper 5800X3D 😉
I'm extremely skeptical about this hybrid design. These dual CCD CPUs have one problem already, which is increased inter-chip latency between different CCDs and scheduler has to take this into account. And now we are adding another level of decisions on top of that, depending on some arbitrary cache vs frequency decisions, which imho will never work 100%. Scheduler will never be able to predict whether some process will benefit from more cache at some point. Even like 1kb executable can benefit from 50MBs of cache and vice versa. It's similar to Intel's e-cores. There will always be situations, where the system will run suboptimal.
Well, I've seen several people questioning the purpose of non-X3D models, and aside from the fact those models can clock higher, they ought to have better latency too.
There's no such thing as a one-size-fits-all chip, especially in the x86 world, where the architecture carries a tremendous amount of baggage for the sake of compatibility.
This is why I say over and over again: buy what you need, not what looks good on paper, and not what reaches some arbitrarily high number that you can't make use of.
Reddoguk:
There's only one thing i don't understand and thats the 7800X3D is a single ccx compared to the 7900 and 7950 with 2 ccx's so how can they all be 120w.
This makes no sense to me. I would of thought the 7800X3D would be a 95w or less because of the 1 ccx.
Picolete:
I find it weird that the 7800X3D has a 120W TDP while the 7700X has a 105W TDP, while supposedly running at higher Hz
Alessio1989:
higher clocks on more cores.
Not just higher clocks on more cores, but you have to account for transistors too. The V-cache accounts for a lot of transistors, those transistors are likely to be under heavier load, and they're stacked (which means they'll run hotter - as temperature goes up, typically so does power draw). Honestly, I'm a little surprised the X3D models don't draw more power.
I logged the top clocking cores here (all on CCX1) and the "worst" the 5505MHz one was on CCX2 they all had clocks around that roughly 200Mhz less than CCX1 if you tried to old school "all core overclock" you would be limited to CCX2 max clocks to be stable, for me you should really treat them as two different cpus