• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Next-Gen PS5 & XSX |OT| Console tEch threaD

Status
Not open for further replies.

mckmas8808

Mckmaster uses MasterCard to buy Slave drives
A heck of a lot more particle physics next gen. Dust, ash, embers, sparks, etc. Assuming the result isn't broken in some way.

My pet theory is that it's also a partial step into building out RDNA 2.0 for hybrid RT, the same logic that calculates particle physics really fast should also be applicable to ray tracing, if made even faster.

Interesting! If that's the case, then a 200,000 is far and away better than Nvidia's 2080's 80,000 score. Maybe AMD has something on their hands with their approach to particle physics and RT.
 

LordOfChaos

Member
you are presenting a VERY misleading view on Swedish (and most european taxes).
Yea they pay 30% income tax but they also pay and ADDITIONAL 30% in hidden taxes. Sweden has a personal tax rate of over 60%. So yea, “free” health care really isnt expensive.

No, I was specifically responding to someone saying a 30% income tax rate was nuts. That alone is pretty normal, that's all I said.

When accounting for other taxes like housing and gst, we spent about 43% of our income on taxes in Canada (vs 37% in the US)



Anywho


Next-Gen Taxes & Tariffs |OT| Countries/Healthcare/Leaks Thread
 
Last edited:

Darius87

Member
bit off topic, but can someone explain why nvidia's turing cards has such heavy perf penalty with RT on vs RT off? even if these cards has dedicated RT cores which supposedly have to take care for RT effects so fps penalty doesn't make sense? or these RT cores doesn't do all the work, meaning cuda cores also do heavy calculations on RT effects, which means nvidia turing is not all HW RT based?
 

ethomaz

Banned
bit off topic, but can someone explain why nvidia's turing cards has such heavy perf penalty with RT on vs RT off? even if these cards has dedicated RT cores which supposedly have to take care for RT effects so fps penalty doesn't make sense? or these RT cores doesn't do all the work, meaning cuda cores also do heavy calculations on RT effects, which means nvidia turing is not all HW RT based?
Ray-tracking is that heavy in real-time.

You will probably need a 1 PENTAFLOPS GPU to have full ray-tracing in real-time... and we are nowhere close to that... so you will need to be happy with ray-tracking being used in some effects with dedicated hardware in actual GPU power.
 
Last edited:

SonGoku

Member
I think consoles makers are unfairly squeezed by consumers in this industry.
I'd gladly pay $1000 for state of the art console tech
If they deliver a conservative mediocre product (<8tf) im not biting for more than $299
Well, expectations of what AMD could deliver where much lower prior to Navi 5700/5700XT launch.
Now it's all sunny.
Precisely! 12 was the expectation of what could be doable on GCN tech inside a console, it was never the benchmark for next gen
Vega64 is not even 2x the real world performance of Scorpio GPU because its severely bottlenecked.
 
Last edited:

Darius87

Member
Ray-tracking is that heavy in real-time.

You will probably need a 1 PENTAFLOPS GPU to have full ray-tracing in real-time... and we are nowhere close to that... so you will need to be happy with ray-tracking being used in some effects with dedicated hardware in actual GPU power.
i know these RT effects in current games is nowhere near full RT scene like CG animations, but question is, does turing cards use cuda cores for RT effects in conjunction with RT cores?
 

ethomaz

Banned
i know these RT effects in current games is nowhere near full RT scene like CG animations, but question is, does turing cards use cuda cores for RT effects in conjunction with RT cores?
What is actually CUDA cores? I think maybe it is the FP32 untis (the graphic/compute units)? If so then yes... these FP32 unit are used for render together with RT cores the scene and ray-tracing effects.

image11-601x1024.jpg
 
Last edited:

Darius87

Member
What is actually CUDA cores? I think maybe it is the FP32 untis (the graphic/compute units)? If so then yes... these FP32 unit are used for render together with RT cores the scene and ray-tracing effects.

image11-601x1024.jpg
then turing is hybrid RT hardware? like RDNA2 will be.
 

ethomaz

Banned
then turing is hybrid RT hardware? like RDNA2 will be.
I'm not sure what that term hybrid means... Ray-tracking is just an algorithm like rasterization... both can be used to render graphics... both uses shader units to be rendered... so in both cases you use Stream Processors.

The different is that Ray-tracking is too heavy to be processed (in fact it has way way way more processing calcs than rasterization)... so to speed up some calcs for ray-tracing nVidia added a hardware unit specialized in speed up these ray-tracking calcs... when there is ray-tracking calcs that can be speed up by the RT core it will run in the RT core instead the SP unit... that way the that specific calc will be done faster.

What ray-tracking calcs are fixed in the RT core I don't know... nVidia didn't disclosed that too... I can only guess that that RT core has some fixed functions/maths that are most used in ray-tracking render to speed up the processo but it doesn't have all the RT functions/maths... there not found in RT core are done by the SPs like in any previous GPUs.

If hybrid is that then nVidia solution is hybrid.

In simple terms... let's say the ray-tracking algorithm has 100 most used calcs... from these 10 most used calcs are speed up by RT Cores while the rest runs in the SPs in Turing.
* Remember I'm guessing here... I don't know how what instructions/calcs are fixed in the RT unit but I believe it is less than 10% of ray-tracking tasks *
 
Last edited:

Aceofspades

Banned
I'd gladly pay $1000 for state of the art console tech
If they deliver a conservative mediocre product (<8tf) im not biting for more than $299

You are concentrating on one side of the equation, the TF number only, ignoring all arch advancements of Navi.

Titan X is a $1000 GPU only that has 6.6TF of performance. Would you be happy if Sony or MS put Titan X inside PS5 + SSD and everything else and sold it for $1399 and call it a day?
 

SonGoku

Member
You are concentrating on one side of the equation, the TF number only, ignoring all arch advancements of Navi.
I think Vega64/1070Ti performance should be $400 at most
Though personally i won't be an early adopter of such mediocre and conservative tech, I'd wait till it hits $299 and there's plenty of games available
Titan X is a $1000 GPU only that has 6.6TF of performance. Would you be happy if Sony or MS put Titan X inside PS5 + SSD and everything else and sold it for $1399 and call it a day?
Titan X is an obsolete card from 2015, why would i be happy with obsolete tech?
What does this DUV mean?
Standard 7nm
 
Last edited:

mckmas8808

Mckmaster uses MasterCard to buy Slave drives
We dont know.
on 7nm DUV the absolute best they can do is 11TF
on 7nm EUV anywhere from 11 to 14 teraflops is possible

Haven't we been talking about how RDNA brings efficiency to levels GCN just can't compete with?
 

Darius87

Member
I'm not sure what that term hybrid means... Ray-tracking is just an algorithm like rasterization... both can be used to render graphics... both uses shader units to be rendered... so in both cases you use Stream Processors.
what i though hybrid is that both RT hw accerelation cores and CU or cuda cores working in tandem on RT calculations.

The different is that Ray-tracking is too heavy to be processed (in fact it has way way way more processing calcs than rasterization)... so to speed up some calcs for ray-tracing nVidia added a hardware unit specialized in speed up these ray-tracking calcs... when there is ray-tracking calcs that can be speed up by the RT core it will run in the RT core instead the SP unit... that way the that specific calc will be done faster.
from what i read that only thing which halts performance with RT effects is ray intersection calculations, maybe denoising also? for which needed extra HW aka "RT cores'. other operation related with RT could run on CU or CPU cores, without halting performance heavily.

What ray-tracking calcs are fixed in the RT core I don't know... nVidia didn't disclosed that too... I can only guess that that RT core has some fixed functions/maths that are most used in ray-tracking render to speed up the processo but it doesn't have all the RT functions/maths... there not found in RT core are done by the SPs like in any previous GPUs.
so there's basically no difference between nvidia RT and what described by Amd next RDNA RT which will next-gen consoles have, except RDNA2 RT HW seems more closely integrated with CU's, both companies have to off load Ray intersection calculations to specific HW, everything else can run without heavily impacting performance.

In simple terms... let's say the ray-tracking algorithm has 100 most used calcs... from these 10 most used calcs are speed up by RT Cores while the rest runs in the SPs in Turing.
* Remember I'm guessing here... I don't know how what instructions/calcs are fixed in the RT unit but I believe it is less than 10% of ray-tracking tasks *
i don't know also but it seems weird to me from what i read nvidia is considered as full fledged RT hardware and AMD next RDNA solution considered as hybrid, so what is what?
 

Aceofspades

Banned
Titan X is an obsolete card from 2015, why would i be happy with obsolete tech?

That proves my point, you focused on only the TF number ignoring everything else.. I agree with you that I wouldn't pay $299 for 8TF (GCN) its not worth it, but we are getting ~9 -11 TF (Navi) that punches like a 12-13.5TF GCN which is a bargin at $499 with cutting edge CPU and state of the art storage system.
 

SonGoku

Member
Haven't we been talking about how RDNA brings efficiency to levels GCN just can't compete with?
Yes
That proves my point, you focused on only the TF number ignoring everything else..
Im focused on performance
Vega64/GTX1080 its not impressive by any metric as a next gen leap
10-11TF NAVI would indeed be impressive and worth $500 (RTX 2080 tier of performance)
 
Last edited:

Bani

Member
Ray-tracking is that heavy in real-time.

You will probably need a 1 PENTAFLOPS GPU to have full ray-tracing in real-time... and we are nowhere close to that... so you will need to be happy with ray-tracking being used in some effects with dedicated hardware in actual GPU power.
That you will surely need such a strong GPU, only if you take that the whole Earth globe is going to be illuminated in RT all the time, whether you see it or not, for the purposes of the game you will never need such a strong GPU with RT,never
 
Last edited:

mckmas8808

Mckmaster uses MasterCard to buy Slave drives
That proves my point, you focused on only the TF number ignoring everything else.. I agree with you that I wouldn't pay $299 for 8TF (GCN) its not worth it, but we are getting ~9 -11 TF (Navi) that punches like a 12-13.5TF GCN which is a bargin at $499 with cutting edge CPU and state of the art storage system.

I agree with this too, but it does matter to me the size of this SSD. PLUS.....will we get a 1 TF removable mechanical HDD too? If so $500 is a bargain.

Yes

Im focused on performance
Vega64/GTX1080 its not impressive by any metric as a next gen leap
10-11TF NAVI would indeed be impressive and worth $500 (RTX 2080 tier of performance)

I think we are finding out that it's possible a 9 TF NAVI GPU may be similar to an RTX 2080 tier GPU. A 7.5TF Navi is equal to a RTX 2070.
 
Last edited:

DJ12

Member
8/9 Navi TF is not going to be powerful enough to deliver strong titles until the end of the generation.

You should feel short changed with anything less than 12 Navi TF.

People need to up their expectations.
Exactly, Microsoft and Sony be reading these threads thinking we're all buying the BS being spouted, probably about to underclock what they have to cheap out on cooling lol
 

vpance

Member
Is this really true ? Anyone doing the mathematical calculations where is SonGoku SonGoku
We dont know.
on 7nm DUV the absolute best they can do is 11TF
on 7nm EUV anywhere from 11 to 14 teraflops is possible

The absolute best on DUV will have them taking a loss at $499 and 250W or more. So it's pretty unlikely.

I don't believe in DUV anymore. I think that was for the initial plan: 2019/early 2020 release with a 5700 equivalent.
 

mckmas8808

Mckmaster uses MasterCard to buy Slave drives
The absolute best on DUV will have them taking a loss at $499 and 250W or more. So it's pretty unlikely.

I don't believe in DUV anymore. I think that was for the initial plan: 2019/early 2020 release with a 5700 equivalent.

NO WAY do either MS or Sony use DUV for a late 2020 console at this point. It makes literally NO sense. I'd rather miss the holiday and come out in March 2021 with a 7nm EUV chip that can net me 20% more transistors with a 10% heat savings per watt.
 
How come 1.8TF GCN1.0 with mobile CPU made it to 2020?

Some might say astro-turfing campaigns are trying to persuade people that performance doesn't matter.

I can see how that would make sense for the weakest console.

Influence expectations lower so that people aren't disappointed when your product is revealed.

On the other hand, the strongest console will likely do a full read out of its specifications to really ram the point home of the performance gap.

There's going to be no hiding place. The reveals are going to be very entertaining.
 
8/9 Navi TF is not going to be powerful enough to deliver strong titles until the end of the generation.

You should feel short changed with anything less than 12 Navi TF.

People need to up their expectations.
I agree but that comes with massive disappointment. I am just gonna go with typical console conservation specs that these console makers usually go with for the past years. If anything I blame the fucking pro and Scorpio releases, they skewed our expectations. If they never came out an 8tflop console would have been huge for us.


I hope I am surprised and Sony blows me away.
 

mckmas8808

Mckmaster uses MasterCard to buy Slave drives
I agree but that comes with massive disappointment. I am just gonna go with typical console conservation specs that these console makers usually go with for the past years. If anything I blame the fucking pro and Scorpio releases, they skewed our expectations. If they never came out an 8tflop console would have been huge for us.


I hope I am surprised and Sony blows me away.

Do you realize that we never got to see the full power of the PRO or Scorpio though? They were held back by their original consoles.
 

llien

Member
How do we know whether 5700/XT are in fact DUV (deep ultraviolet) and not EUV (extreme ultraviolet, shorter wavelength than DUV?
TSMC was promising 7nm EUV mass production starting March this year. If they are on track, that is just about right for AMD 7nm products, bar Vega VII.
 

vpance

Member
How do we know whether 5700/XT are in fact DUV (deep ultraviolet) and not EUV (extreme ultraviolet, shorter wavelength than DUV?
TSMC was promising 7nm EUV mass production starting March this year. If they are on track, that is just about right for AMD 7nm products, bar Vega VII.

ammd-rdna2-2020.jpg
 

LordOfChaos

Member
You have a point but the average gamer/ consumer will always look at the bigger number. To you and me we know the difference but Google will celebrate that 10.7 over that 9tf

I think it'll be fine with as much as comparison videos are all the rage. Remember how much XBO die hards were wishing for some secret sauce that would put it ahead, but when the actual real world comparisons come out it's pretty conclusive. Same as the Switch often being reported by the FP16 number vs everyone elses FP32.

Google can tout a bigger number, but it won't last long if it's behind in real world graphics, it'll just end up looking silly.
 

mckmas8808

Mckmaster uses MasterCard to buy Slave drives
Google can tout a bigger number, but it won't last long if it's behind in real world graphics, it'll just end up looking silly.

And throw in the "over the internet/lag" portion of Google Stadia for comparisons and things could look downright ugly. We've ALL seen those blotchy videos while watching YouTube if things move too fast in the video. That never happens with locally powered games.
 

R600

Banned
Some might say astro-turfing campaigns are trying to persuade people that performance doesn't matter.

I can see how that would make sense for the weakest console.

Influence expectations lower so that people aren't disappointed when your product is revealed.

On the other hand, the strongest console will likely do a full read out of its specifications to really ram the point home of the performance gap.

There's going to be no hiding place. The reveals are going to be very entertaining.
Why dont you reply to my post? What happened in 2013 when STRONGER console was made up of 1.8TF GPU (when 4TF was top in PC market) and mobile CPU?
 

pawel86ck

Banned
bit off topic, but can someone explain why nvidia's turing cards has such heavy perf penalty with RT on vs RT off? even if these cards has dedicated RT cores which supposedly have to take care for RT effects so fps penalty doesn't make sense? or these RT cores doesn't do all the work, meaning cuda cores also do heavy calculations on RT effects, which means nvidia turing is not all HW RT based?
RT cores basically speaking are good enough only for 1080p calculations, but are too slow for higher resolutions, and because many gamers play in 1440p/4K so performance penalty is extreme due to RT cores bottleneck.


Y49I2hy.png

You see, when RT cores aren't a bottleneck then performance penalty is small.
 
Last edited:

ANIMAL1975

Member
Why dont you reply to my post? What happened in 2013 when STRONGER console was made up of 1.8TF GPU (when 4TF was top in PC market) and mobile CPU?
It was already point out in this thread by other people, but you only compute the information that suits you. What happened in 2013 was the previous generation with 300 dollars taken lost per console, rrod and red light of death, no mandatory ps plus... This time around is all different picture!
 
Status
Not open for further replies.
Top Bottom