Dr. Claus
Vincit qui se vincit
You are so full of shit that you could fill to bursting the Hirakud Dam.
-
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.
- Status
You are so full of shit that you could fill to bursting the Hirakud Dam.
Imtjnotu
Member
Have faith.
You are so full of shit that you could fill to bursting the Hirakud Dam.
Imtjnotu
Member
Gonna go out on a limb and call bullshit here
StreetsofBeige
Gold Member
Considering PS3 was a 2TF machine that could do 1080p at 120 frames per second, I heard a rumour too.
PS5 will be 24TF, 8k, 240 frames
Last edited:
TLZ
Banned
Audiophile
Member
I do think however that Spider-Man is in dire need of ray-traced reflections on its skyscrapers, it's one thing that really breaks the illusion in that game..
ResilientBanana
Member
I would tend to agree with you there. Ray tracing would do that game a world of justice as GI may go unseen in that world.
Is there any way to get this guy verified to ensure what he says is truthful?
Last edited:
Negotiator
Member
Switch Pro is 7TF?
Fake
Member
Nvidia Navi 7TF yes.
svbarnard
Banned
Audiophile
Member
While I'm still of the opinion that devs will mostly favour eye-candy and spatial resolution over framerate... If you're aiming for 4K/30 or 4K/60, it might make more sense to go ~1440p/60 or ~1440p/120 and use reconstruction 4k reconstruction technique such as Insomniac's Jitter/Temporal Injection (the best there is imo)...
Such techniques are already getting very close to 4K with minimal artifacting; and with twice as many samples (frames) to reference that reconstruction is only going to be better, the increase in temporal resolution will also increase perceived clarity, you'll have the responsiveness and fluidity that comes with increased framerates; and finally if you stick in the 1440-1500p range you'll still have some change left over for additional fx/logic.
Such techniques are already getting very close to 4K with minimal artifacting; and with twice as many samples (frames) to reference that reconstruction is only going to be better, the increase in temporal resolution will also increase perceived clarity, you'll have the responsiveness and fluidity that comes with increased framerates; and finally if you stick in the 1440-1500p range you'll still have some change left over for additional fx/logic.
bhunachicken
Member
bhunachicken
Member
I've said it (many times) before, but:
Coder: It's running at 70fps, with everything enabled.
Management: Oh, so you've got frames to spare?
Coder: Well, yeah, but ...
Management: Add more bling!
Coder: Okay, we're at 65fps
Management: More bling!
Coder: 62fps
Management: MORE BLING!
Coder: 59fps
Management: Right, cap it at 30fps
You had your fun, let's dial it back on the "insider" trolling.
ResilientBanana
Member
I don't think that's the case. Framerate is mostly variable these days. So they usually decide early in the design phase whether or not they want to hit 30 or 60.
Tqaulity
Member
I believe you
. I've been saying this since April when Sony first unveiled the PS5 that the GPUs in the both will likely end up around 12TFLOPS. For marketing reasons, it makes sense to at least say that the next boxes are 2x the graphical performance of the Xbox One X (even though we now all now that Navi is waaay more efficient so 12TFLOPS of Navi is likely ~16TFLOPS of Polaris). In terms of raw horsepower, we know that the next high end Navi cards from AMD is aiming to deliver 2080 Super level of performance as configured for PC. In typical console fashion, the cards will have to be paired down a bit for yields and power considerations so expect the consoles to be roughly based on the RX 5800 (or whatever it will be called) but with slightly less CU count and under-clocked when compared to the PC equivalent. With that said, we will still be looking at 2070 Super level perf at a minimum but more likely closer to RTX 2080 level (I would not be surprised if one console is slightly less powerful (i.e 2070 Super vs 2080 standard)). Yes, RTX 2080 level performance in a dedicated closed box console ?! Some amazing things to come! U R NOT EBut here's the breakdown for those that may be confused: The RTX 2080 is ~10TFLOPs by Nvidia's calculations. However, we can extrapolate what an AMD Navi based card with similar power will measure in terms of TFLOPs. How? Just look at the specs vs perf between the RX 5700 XT and the RTX 2070 Super/ RTX 2080. The 5700 XT is ~9.7TFLOPs out of the box so let's say 10TFLOPs with an overlock (same at RTX 2080). Yet, according the the perf benchmarks at sites like TechPowerUp, the RX 2070 Super is ~10% faster than the RX 5700XT (assuming a slight overclock to reach the 10 TFLOPs i.e. the Anniversary edition). Meanwhile the RTX 2080 is roughly 20% faster despite having the same TFLOPS measurement. So that means that AMD's Navi architecture is still roughly 20% less efficient than Nvidia's Turing. Thus, for AMD Navi to reach the 10TFLOPs perf of an RTX 2080, the Navi card will need to be roughly ~12TFLOPs (10TFLOPs + (10*0.2)).
Coincidence!?? I think NOT
As someone who has worked with PC hardware, software, and games professionally for over 10 years, I assure you it's not rocket science
Last edited:
svbarnard
Banned
This user has been removed from thread. Ignored prior warning.
I believe you
But here's the breakdown for those that may be confused: The RTX 2080 is ~10TFLOPs by Nvidia's calculations. However, we can extrapolate what an AMD Navi based card with similar power will measure in terms of TFLOPs. How? Just look at the specs vs perf between the RX 5700 XT and the RTX 2070 Super/ RTX 2080. The 5700 XT is ~9.7TFLOPs out of the box so let's say 10TFLOPs with an overlock (same at RTX 2080). Yet, according the the perf benchmarks at sites like TechPowerUp, the RX 2070 Super is ~10% faster than the RX 5700XT (assuming a slight overclock to reach the 10 TFLOPs i.e. the Anniversary edition). Meanwhile the RTX 2080 is roughly 20% faster despite having the same TFLOPS measurement. So that means that AMD's Navi architecture is still roughly 20% less efficient than Nvidia's Turing. Thus, for AMD Navi to reach the 10TFLOPs perf of an RTX 2080, the Navi card will need to be roughly ~12TFLOPs (10TFLOPs + (10*0.2)).
Coincidence!?? I think NOT
I know what you mean I have worked at 989 Studios for the last twenty years, who here liked syphon filter?
Panda1
Banned
I believe you
But here's the breakdown for those that may be confused: The RTX 2080 is ~10TFLOPs by Nvidia's calculations. However, we can extrapolate what an AMD Navi based card with similar power will measure in terms of TFLOPs. How? Just look at the specs vs perf between the RX 5700 XT and the RTX 2070 Super/ RTX 2080. The 5700 XT is ~9.7TFLOPs out of the box so let's say 10TFLOPs with an overlock (same at RTX 2080). Yet, according the the perf benchmarks at sites like TechPowerUp, the RX 2070 Super is ~10% faster than the RX 5700XT (assuming a slight overclock to reach the 10 TFLOPs i.e. the Anniversary edition). Meanwhile the RTX 2080 is roughly 20% faster despite having the same TFLOPS measurement. So that means that AMD's Navi architecture is still roughly 20% less efficient than Nvidia's Turing. Thus, for AMD Navi to reach the 10TFLOPs perf of an RTX 2080, the Navi card will need to be roughly ~12TFLOPs (10TFLOPs + (10*0.2)).
Coincidence!?? I think NOT
The problem with this story is that today a 2080RTX
2080 rtx - Google Shopping
www.google.com
Is about £600
You wont be able to get this cards power even if half price next year and the parts for the rest of the console especially with SSD, controller, memory etc for £500. No one is going to be stupid enough to launch a base console at 599. (even though I would not mind). Im thinking they will use another metric rather than FLOPS to compare the machines to make it seem more of a jump - but 8TF is more realistic.
svbarnard
Banned
Lol 8TF!!!! I am telling you the next Xbox will be 12TF, Navi will do that.
Panda1
Banned
Then you must believe high end PCs would be able to get 40TF cards around early 2021?
Tqaulity
Member
Actually the problem with your point is that you are comparing Nvidia products and pricing to AMD's AND you are comparing console component pricing to PC which is absolutely not the same. Fact: AMD's next high end Navi card will deliver RRTX 2080/2080 Super performance and will likely cost less than the current Nvidia cards on the market. Also, Fact: console manufacturers purchase the components with a much different pricing model since they are purchasing in bulk for a multi-year term of the contract. Example, the PS3 was based on Nvidia's 7800 series of GPUs and the 7800 GTX for PC launched at $600 back in 2005. Yet, the PS3 BOM estimate for it's GPU was only $83 per unit at launch.
Rumors and leaks have been pointing to the next gen console have 11 TFLOPS or more since last year this time. I don't know why people are so resistant to the idea now?
Tqaulity
Member
Here is another way to think about it: IF the next gen console were to be only 7-8Tflops, that would put them at the current RX 5700 (non-XT) level. Does anybody here really believe that the next gen consoles will not be at least as powerful if not significantly more so than the current RX 5700 XT (~10TFLOPS)? Does anyone think Microsoft and Sony would invest billions of dollars to build a machine to last 5-7 years and be the future of gaming to launch with a entry level - mid-range GPU released over 1 year earlier than their box? With the slew of more powerful Navi cards still to launch between now and end of 2020, they will settle on a small mid-range card from 2019 that is barely a step up from the Xbox One X? Does that even make sense?
Negotiator
Member
Turing uses 12nm FinFET. Roughly the same process as 14-16nm.
7nm offers 3.3x scaling on the exact same die size. 3 times more transistors:
TSMC 7nm HD and HP Cells, 2nd Gen 7nm, And The Snapdragon 855 DTCO
Update and analysis of TSMC 7-nanometer node low-power and high-performance cells, 2nd generation 7nm, and the design technology co-optimization (DTCO) effort that went into the Snapdragon 855 SoC.
fuse.wikichip.org
That means a 754mm2 chip (RTX 2080 Ti) can be 1/3 (252mm2). Not sure how many people realize this.
Don't believe me? Well, just wait for Ampere next year and you'll see firsthand what I'm talking about. A high-end/enthusiast card from 2018 will be 2020's mid-range card.
It's the same reason the "weak" RSX was able to be as fast as 2 x GeForce 6800 Ultra (2004 high-end/enthusiast card). 2 years of lithography progress can do wonders. Console makers are experts on this.
Also: https://www.tsmc.com/tsmcdotcom/PRListingNewsAction.do?action=detail&language=E&newsid=THHIHIPGTH
"The N7+ volume production is one of the fastest on record. N7+, which began volume production in the second quarter of 2019, is matching yields similar to the original N7 process that has been in volume production for more than one year."
Negotiator
Member
Arcturus Navi is a thing now?
I thought Arcturus is based on the Vega series, while RDNA is a new microarchitecture.
Munki
Member
To be fair, the information in this video is stuff we have already seen in this thread (Zen2, Acturus, X1X with beefier CPU, new contoller for Scarlett with haptic feedback).
Last edited:
DeepEnigma
Gold Member
Next-Gen PS5 & XSX |OT| Console tEch threaD
Not true. RT reflections are much more expensive than RT AO or shadows. In control enabling a few RT features doesn't hurt performance all that much but once I enable Reflections, or if I only enable reflections, the performance takes a punch to the face. Not all RT techniques have the same...
Next-Gen PS5 & XSX |OT| Console tEch threaD
Not true. RT reflections are much more expensive than RT AO or shadows. In control enabling a few RT features doesn't hurt performance all that much but once I enable Reflections, or if I only enable reflections, the performance takes a punch to the face. Not all RT techniques have the same...
llien
Member
Guys, please realize how POOR today's RT performance is. A ray per pixel (at best current gen could do that, and even than, at low resolution) scene looks like this:
Nvidia blurs/denoises it alot, of course losing details, and gets something good enough for some reflections/shadows, but that's it.
We need RT performance that is dozens of times ahead of what we have now, to start thinking about it being rendered as RT only.
Even Pixar, which uses server farms to render it's movies, at first combined raster and RT, only later on switching to pure RT, as it is simpler for them to model and their srever farms grew powerful enough to handle that. This is how their "render farm" looks like:
jonnyp
Member
Colt the Dolt is one of the biggest fanboys there is. He is almost misterxmedia level.
There was nothing to copy. Unified RAN was in 360, it was pretty straight forward and logical way IF you didnt care all that much about absolutely having to have 8GB of RAM.
In 2010, when consoles where designed, that was only way to guarantee 8GB of RAM being there on day one.
Critical strat mistake by MS, but it was more of a gamble duo to Kinect and TV. Sony where led by gaming first design, and won.
LostDonkey
Member
Ffs man im eating.
Negotiator
Member
What do you mean there was nothing to copy?
They abandoned the eSRAM+DDR3 combo and went with a conventional GDDR5 setup. No huge scratchpad memory wasting transistors that could have been GPU ALUs. That's what the PS4 has had since 2013.
There's a reason for that as I said. It made sense when fast DRAM was expensive and resolution was low (PS2 was an SD machine, XBOX 360 was sub-HD, since 10MB eDRAM was not enough for 720p).
It doesn't make sense anymore, because eSRAM requirements for native 4K (let alone 8K) would be insane. Imagine having 128MB of eSRAM (4k = 4 * 1080p)!
Last edited:
sonomamashine
Member
New Sony Patent Shows Network-Connected PlayStation Controller for Cloud Gaming
Sony Interactive Entertainment has recently filed a patent for a new network-connected PlayStation controller for cloud gaming.
wccftech.com
Mass Shift
Member
MS and Sony do NOT pay retail prices for any of their processors. WE are the ones who pay msrp.
I'll bet they won't even be paying $150 USD for their next gen APUs.
Panda1
Banned
Then every single crypto miner will buy the consoles and mine crypto if they can access more power per dollar than buying a card straight up - like the ps3 a lot of people got them to run non gaming applications.
Im just saying, copying unified memory arrangement sounds dubious. It was either Esram + DDR3 or GDDR5 from single pool, for both if them.
Designing console, with main requirement being it having 8GB of RAM back in 2010, there was only one choice. If MS knew 100% that GDDR5 could deliver 8GB in 2013, they would have gone with that thus putting more compute on chip without raising BOM.
Last edited:
iamvin22
Industry Verified
Why do people still think it works like this? Every gen we get this quote.
Panda1
Banned
name me a gen where the console surpassed the strongest pc graphics card 18 months old?
If the consoles were sold below cost per dollar it would be like
PlayStation 3 cluster - Wikipedia
Last edited:
Mass Shift
Member
Good for them. It's certainly their option to do so. But the average person playing video games doesn't possess such skills. So I don't think the console industry has anything to be concerned about.
Panda1
Banned
If i can mine crpto on it for 100 less than a pc alternate and resell the ps5 after 18 months at 10-20% loss then i will save around $200 per ps5 i purchase over 18 months,
If I buy 100 then I save 20.000 plus I also have my cryto - tell me how the hell you gonna get your hands on one if they are always going to be sold out in the first 2 years and sony wont recop any profit on any person doing this - it makes no sense in todays world when the ability to compute can actually be used to make money.
There was no cryto mania in 2005 with ps3/x360 but with x1 and ps4 being under powered it was not an issue.
If the new consoles are overpowered per dollar than a pc comparable then its open season for mining. T
Because Stadia is GCN. 10.7TF of Stadia is ~ 7TF from RDNA.
8-9TF RDNA + RT cores should significantly outperform it. In close box, not server stack.
Mass Shift
Member
Uh, what can I say? I wish you well on your new business venture.
But this is the console industry's business model and what differentiates from other gaming platforms. If you're okay with MS, Sony and Nintendo making the hardware choices for you and you're willing to buy into their cloistered eco systems, they pass along their bulk purchase savings to you as apart of their exclusive install base.
You get multiplatforms at decent specs, and optimized exclusives at their best possible performance. There are trade offs but it's quite mutually beneficial.
- Status