Yep. I bet both consoles are on 7nm.Every bit of info leaked or released so far suggests that the PS5 APU is built with 7nm non-EUV process. Including this one:
Baseless Next Generation Rumors with no Technical Merits [post E3 2019, pre GDC 2020] [XBSX, PS5]
- Thread starter BRiT
- Start date
- Status
Fan noise cancelling headphones would be mandatory and add 30 dollars to the cost of the console at launch
Do you redefine the specified meaning of teraflops just to keep right, or is this seriously how you perceive the unit after years of relating it to other non specific factors like a 'real world performance'?It's not nitpicking but making important clarifications with regards the exactness of what you stated. You posted a pure, logical assertion, "1==1, period." Anyone looking up how GPUs work could think that's how it works, but it isn't. Now if you are saying that in real terms, it makes no difference because of the flops actually being used RDNA will see no advantage, you have an argument, but that's different to saying "1==1" and it's also along the lines of people saying 1 GCN TF != 1 RDNA TF because in real world terms, utilisation is better where you called it bullshit.
If you run a real world application benchmark, do you get a teraflops result measuring the performance?
Where on earth it is an 'engineering fact' that 1 != 1?
If doing MADDs is all the work that is done, like here, then how can 1 be larger than 1, depending on the worker?
Obviously this makes no sense to me, and bringing in geeky and informed sounding but unrelated terms does not make me feel less right.
But i give up. HW Raytracing is best thing ever, better HW is the solution to all problems, and one is a bit larger than one.
Lessons learned by somebody who looks up how GPUs work :|
What TF is MADD?
Mothers against drunk driving.
In light of my recent post and lapse of memory it must be
Mark (Cerny) Against Duck Demos !
The spoilsport
Question Context:
TF = The Fuck
Answer Context:
TF = TeraFlops
Flops = Floating-Point Operations
MADD = Multiply, Add
GPU = Graphics Processor Unit
They use MADDs to measure TFs because of how GPUs are optimized to do both at the same time.
Obligatory Horrible Plane analogy to AMD GCN and RDNA:
TFs indicate how much can be transported from point A to point B (passenger and luggage units combined).
GCN plane is assumed every trip is done with exactly 2 passengers with 3 additional pieces of luggage to fit in exactly 2 Passenger Units and 3 Luggage Units. It can not convert Luggage Units to Passenger Units. If a passenger travels without cargo, the unused luggage units can not be used and goes unfilled. The plane leaves every 4 hours.
RDNA plane is assumed every trip is done with different passengers with different amount of luggage. It is able to convert Luggage Units into Passenger Units or Passenger Units into Luggage Units so it's more flexible and able to better fill it's internal capacity and make scheduling easier and more efficient. The plane leaves every hour.
TFs indicate how much can be transported from point A to point B (passenger and luggage units combined).
GCN plane is assumed every trip is done with exactly 2 passengers with 3 additional pieces of luggage to fit in exactly 2 Passenger Units and 3 Luggage Units. It can not convert Luggage Units to Passenger Units. If a passenger travels without cargo, the unused luggage units can not be used and goes unfilled. The plane leaves every 4 hours.
RDNA plane is assumed every trip is done with different passengers with different amount of luggage. It is able to convert Luggage Units into Passenger Units or Passenger Units into Luggage Units so it's more flexible and able to better fill it's internal capacity and make scheduling easier and more efficient. The plane leaves every hour.
Janne Kylliö
Newcomer
Possibly, though I'm not really seeing how XSX could have 56 CUs without smaller node?
AbsoluteBeginner
Regular
Its interesting that 2.0GHz makes no sense with 36CUs, but 1.8GHz with 60% more CUs to match ~13TF does. How does that work out?
AbsoluteBeginner
Regular
I would not be surprised by N7P. Its same process, only 2nd gen and does not require any redesign.Yep. I bet both consoles are on 7nm.
I have no idea which posts you are refering to. Please quote.
Are you seriously wondering how 12TF or 13TF is possible? As it's always been done between midrange and higher end gpus of the same series. More CUs and more memory controllers.
The skepticism against 2ghz is that higher clocks are hitting thermal density limits and have an efficiency drop. There's a point where it's not possible to conduct the heat to the heat sink no matter how big it is or how fast the fan turns.
Last edited:
Nice
But on both GCN and RDNA the planes can be used also from passengers from a parallel universe to occupy unused seats.
So the RDNA advantage only matters after the airport opens its doors once a day, to fill the seats faster.
On the other hand, GCN has the advantage a flight takes only 4 hours, while on RDNA it takes 5, IIRC, this helps with reaching distant destinations faster.
In the GCN and RDNA universe both transfer 1 billion passengers per weak, but due to quantum tunneling effects and fluctuous appearance of matter, RDNA universe spawns 0.3 billions of additional passengers out of nothing.
(still puzzling what's passengers vs. luggage exactly, admitted
)Given the tests were for both 1.8GHz and 2.0GHz it's possible that it could end up anywhere between that. No one is speculating about it but there is a possibility that it's lower than 9.2Tflops if that 36CU is correct.
I'm guessing if you test 1.8 then move up to 2.0 then the final version is at least above 1.8GHz...
I'm guessing if you test 1.8 then move up to 2.0 then the final version is at least above 1.8GHz...
Panino Manino
Newcomer
I'm thinking, isn't the PS5 STRONGER than the SEX?
Remember a good while ago rumors about an "Arcturus" GPU? That was meant for Microsoft and was based on Vega? More specifically,not Navi, because that architecture was being made for Sony?
Now we have all this talk about the SEX having 3 more TFlops than the PS5, and with way more SP. The rumored number, 3584, is the same number as the Vega 56 and a number way higher than it appears we'll see in a RDNA 1.0 GPU. The higher clocks is because the SOC will be made on 7nm instead of 14nm. At first glance this may make it sound like the SEX is stronger than the PS5 that will only have 2304 SP, however we all know that Navi performs generally much better than Vega, right? Didn't we see that the RX 5700 performs even better than Vega 64? So even with a few less TFlops wouldn't the PS5 still be a bit stronger in practice if all the other specs are more or less the same?
With the necessity to add hardware to RT would Microsoft still pay AMD to design a different and bigger GPU based on RDNA 1.0?
Remember a good while ago rumors about an "Arcturus" GPU? That was meant for Microsoft and was based on Vega? More specifically,not Navi, because that architecture was being made for Sony?
Now we have all this talk about the SEX having 3 more TFlops than the PS5, and with way more SP. The rumored number, 3584, is the same number as the Vega 56 and a number way higher than it appears we'll see in a RDNA 1.0 GPU. The higher clocks is because the SOC will be made on 7nm instead of 14nm. At first glance this may make it sound like the SEX is stronger than the PS5 that will only have 2304 SP, however we all know that Navi performs generally much better than Vega, right? Didn't we see that the RX 5700 performs even better than Vega 64? So even with a few less TFlops wouldn't the PS5 still be a bit stronger in practice if all the other specs are more or less the same?
With the necessity to add hardware to RT would Microsoft still pay AMD to design a different and bigger GPU based on RDNA 1.0?
Last edited:
- Status
Similar threads
- Locked
