Don't even need to make assumption about MT and power limits, Robert Hallock has already stated in an interview recently that the increased power limits for Zen4 at the high end models was done simply to allow for higher all core clocks during multithreaded workloads - basically Zen4 is pre overclocked from the factory
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
AMD to unveil Zen 4 CPUs at CES 2022
- Thread starter Cromulent
- Start date
- Status
Don't even need to make assumption about MT and power limits, Robert Hallock has already stated in an interview recently that the increased power limits for Zen4 at the high end models was done simply to allow for higher all core clocks during multithreaded workloads - basically Zen4 is pre overclocked from the factory
Permabanned
- Joined
- 28 Sep 2018
- Posts
- 0
You'll want to examine the various MT workloads due to the addition of certain avx512 instructions.
Associate
- Joined
- 7 Nov 2017
- Posts
- 1,950
Are you really that ignorant of how there can be an increase for multithreaded performance compared to single (what apps actually just use a single core these days anyway)?
Don’t know why you are going about amd increasing power when intel had to up theirs to 240 W to beat AMD in multithread apps but loose to 5800x3D in a load of games.
Associate
- Joined
- 27 Apr 2007
- Posts
- 966
AMD have formally stated that power efficiency has increased significantly, so it’s way too early to jump to conclusions.
They might have stated a 25% improvement in last week’s announcement.
Then there’s the AVX-512 addition, so when you add all the info together including the max TDP for the socket, it’s impossible to say how much a 16 core chip might consume at stock.
They might have stated a 25% improvement in last week’s announcement.
Then there’s the AVX-512 addition, so when you add all the info together including the max TDP for the socket, it’s impossible to say how much a 16 core chip might consume at stock.
Caporegime
According to Tim from Hardware Unboxed, games, games use a single core, Tim from Hardware Unboxed is an idiot. The problem is Youtube idiots is where a lot of people get their info from.
The 7950X has a 170 watt TDP.
Associate
- Joined
- 19 Nov 2015
- Posts
- 108
Paraphrased a little there... They've acknowledged multicore use in games for some time. What they have stated several times over is that most games respond well to fast cores as opposed to many slower cores. Single thread performance is usually indicative of this.
Caporegime
I'm not the only one whose noticed this, MLID is constantly being told by his subscribers that single core is all important, to the point where he pulled them up for it and cited people like Tim from Hardware Unboxed as "should know better".
When you watch him you can see why, in his last video pretty much said he thought Rocket lake will beat Zen 4 in games because Rocket Lake has higher single threaded performance, he should know better so i don't know why he does this, its like he has some sort of cognitive dissonance when it comes to Intel.
The thing is it doesn't matter how much that is disproved they just keep doing it....
One thing I rarely ever see mentioned in the reviews is the subjective feel of systems. I find more cores feels "smoother", the number don't always give the full picture, especially when the average user has all sorts running in the background. Now if AMD can bring more cores and a higher clock speed on some of those cores even better. I'd say go for 8C16T and you won't go too far wrong.
Caporegime
I'd tend to agree with that, a 12 thread CPU will get you the same Frame Rates but a 16 thread can be smoother, depending on the game, i don't really know why, other than one has more resources than the other.
Just to continue of the ST in games thing, that hasn't really been true, for Nvidia at least, since Maxwell, since Nvidia introduced software thread scheduling that splits the main thread in to many threads, this is quite often why games actually run better on DX11 than DX12, because DX12 isn't doing anything new in that sense and DX11 is much more mature.
For AMD they introduced the same thing in the Vega architecture, but they do it on the GPU its self.
Caporegime
Also, Alder Lake clocks much higher, has a higher IPC and with DDR5, and yet is barley 10% faster than Zen 3 in games, you would think with all Alder Lake has going for it it would humiliate Zen 3, it doesn't.
The 5800X3D is clocked near a full Ghz lower, again DDR4 vs DDR5 running 2X as fast, the 5800X3D is at least as good in games as the 12900KS, in some heavy CPU bound games the 5800X3D crushes the 12900KS.
Innovation has pushed CPU architectures to way beyond from what we had even just 5 years ago.
The 5800X3D is clocked near a full Ghz lower, again DDR4 vs DDR5 running 2X as fast, the 5800X3D is at least as good in games as the 12900KS, in some heavy CPU bound games the 5800X3D crushes the 12900KS.
Innovation has pushed CPU architectures to way beyond from what we had even just 5 years ago.
Why are you getting so upset when I'm just pointing out that the main reason for AMDs large 35% MT boost is power limits?
They could literally package a zen 3 chiplet on the new AM5 with higher power limits and get 20%+ more MT performance so 35% on a new arch/node isn't that impressive considering zen 3 on the same 7nm as zen 2 with the same power limit achieved 15% MT and 23% ST.
The 5800X3D benefits from the large cache so will likely be better or as good as Zen 4 in a lot of games atleast till the Vcache versions arrive which if your a gamer then it seems a no brainier to hold out for over the vanilla zen 4 which looks mediocre at best for gaming.
Don't forget that even the vanilla zen 3 has an L3 cache advantage over ADL, 12600k 20mb / 12700k 25mb 12900k 30mb L3 vs 5600X / 5800X 32mb or 64mb for the 12 / 16 core parts and we've seen how much a difference L3 can make with CPUs like the 5800X3D and on the other side of the coin the poorly performing ryzen 5 5500 with just 16mb L3.
I can see there being a lot more use of increased cache sizes. Apparently there are issues with latency if you got too large though. I get the feeling that CPU's are going to far outstrip the software in the next 5 years.
Associate
- Joined
- 7 Nov 2017
- Posts
- 1,950
Zen 3 v Zen 4 V ADL running the same power would be ALD then Zen 3 then Zen 4 at the top. Zen 3 cannot run at 5.5 ghz on ambient cooling. It is not just a slight increase in TDP for the reason for the 35% increase in MT performance. Intel need 240 w to get the better of the 5950X. The 12900K can hardly be cooled properly.
Caporegime
It is intresting, the 6 cores Zen 3 has access to the same 32MB cache pool as the 8 core, and yet there is no difference gaming performance, its the same for the 12700K vs 12900K, clock for clock there is no difference despite 25MB vs 30MB of L3, normally its how much cache per core you have.
Perhaps its not quite as simple as just add cache.
Associate
- Joined
- 27 Apr 2007
- Posts
- 966
AMD have recently revealed that Zen 4 has the following improvements over Zen 3 in Cinebench:
Overall performance increase >35%
Performance per watt gains >25%
If we take those figures as being for the same workload, Cinebench, that gives an increased wattage of 8%.
That figure is calculated from the increased performance and efficiency numbers.
So if we use the 142W AM4 socket limit as an actual wattage for Zen 3, Zen 4 is consuming 153W.
An 11W increase hardly puts it in Intel territory.
Even though that is based on actual AMD numbers, it still has to make too many assumptions to tell us that much
Overall performance increase >35%
Performance per watt gains >25%
If we take those figures as being for the same workload, Cinebench, that gives an increased wattage of 8%.
That figure is calculated from the increased performance and efficiency numbers.
So if we use the 142W AM4 socket limit as an actual wattage for Zen 3, Zen 4 is consuming 153W.
An 11W increase hardly puts it in Intel territory.
Even though that is based on actual AMD numbers, it still has to make too many assumptions to tell us that much
Caporegime
Zen 4 will gain 25% performance per watt, according to AMD, Robert said the 16 core Zen 4 they demoed was more than 105 watts but less than 170, its probably around 140 watts, there is some headroom before they reach the 170 watt cited TDP.
Caporegime
120 watts at the 12v EPS for the 5950X, according to Steve Burk.
Associate
- Joined
- 27 Apr 2007
- Posts
- 966
Data varies a lot depending on the article as there are a lot of variables.
That's why I went with the PPT value as it's fixed and is an official worst case scenario, of one kind.
So using 120W and the 8% increase puts Zen 4 at just under 130W.
As that is likely a 16C part, it gives space for a 24C chip within the AM5 socket PPT at a decent clock speed.
Caporegime
You're right, either way its not high.
- Status