-
Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.
Next-gen PS5 and next Xbox speculation launch thread - MY ANACONDA DON'T WANT NONE
- Thread starter Phoenix Splash
- Start date
- Status
Why have one billion when you can have...Okay, this thread has really gone places with completely insane and impossible Tflops number.
Let's calm down and lower our expectations, shall we??
I expect no less than ONE BILLION TERAFLOPS
ONE MILLION
You missed my point completely. Sorry.
Still talking strict PC here.
a) I was not talking about the TDP in my response at all as you can't derive any clock speed from it or the resulting peak performance.
b) I never mentioned 2000Mhz, I mentioned 2280MHz. You seem to ignore that fact completely and make up a completely new context.
c) To me the 2280MHz is a crazy clock speed looking at the Radeon 7 that already is on a 7nm process and running at 1750Mhz boost clock.
d) Following your suggested math just for the first line already contradicted your idea how the performance column is to read by putting the most expensive card behind the 48CU GPU in peak performance.
e) The Xbox One X example was a general assessment that we instead looking just at the peak performance we should also look into it from an efficiency standpoint. We all know that Nvidia produces a better in-game performance with the same peak performance. I would assume that AMD has worked on that issue to close that gap to Nvidia with Navi. So instead to have to increase clock speeds which reduces yields why not just build a more efficient GPU and have some headroom to finally compete again with Nvidia at in-game performance and power-efficiency. I agree based on the table they will achieve the better power-efficiency target if those numbers are accurate.
I repeat: My point is that the performance column in the table we talk about is not referencing increases in peak performance (TFs) by percentages, it compares to expected in-game performance benchmarked for those existing GPUs and the mentioned percentages is how much we should expect as an increase. The only way you could derive the peak performance from that table would be if the table would contain the actual clock speeds for the Navi GPUs.
Maybe it would help you if you remake the diagram and put the clock speeds and the resulting peak performances of the Navi GPUs and the GPUs compared to into the new diagram. And then have fun to explain all the inconsistencies.
Edit:
This is not about what you think the peak performance of the PS5 or Anaconda are.
Last edited:
Colbert, you're gonna be shocked when your low ball estimates are going to be exceeded.
Heck, we already have a proven insider claim that both Sony and MS will exceed Stadia's 10.7 TF
You have gone to great lengths over the months setting the bar low.
Heck, we already have a proven insider claim that both Sony and MS will exceed Stadia's 10.7 TF
You have gone to great lengths over the months setting the bar low.
Alright, finally unbanned from thanos's snap, i missed A LOT of excitement here, haha.
I did see some confusion about gonzalo and ariel, from my understanding gonzalo is the codename of the APU itself, while ariel is specifically the GPU part? I might be wrong but i think ariel is basically the navi 10 lite.
As to expectation, i now think PS5s gpu will be 1800mhz with 48 cu (basically the navi 10, as i think navi 10 lite is a navi 10 with different features for being semi custom).
I did see some confusion about gonzalo and ariel, from my understanding gonzalo is the codename of the APU itself, while ariel is specifically the GPU part? I might be wrong but i think ariel is basically the navi 10 lite.
As to expectation, i now think PS5s gpu will be 1800mhz with 48 cu (basically the navi 10, as i think navi 10 lite is a navi 10 with different features for being semi custom).
To be fair Jason said that as far as he knows both Sony and Microsoft were aiming for specs above Stadia. He also clarified he doesn't know the actual specs.
Ah well, not much longer until Sonys fiscal year report. Let's see if anything of note comes from it as far as next gen tidbits.
Probably nothing more than "we are planning to release the next generation of the PlayStation console in 2020"
Maybe. But maybe we'll get fiscal projections that could give us hints at what could come. Not expecting them to provide a teraflops number or anything, lol, but just some subtle clues here or there might be nice.
Aye, maybe.
A big drop in profit forecast for the next fiscal year might mean a $500 console?
I only hope we will get some teases or confirmation of a super slim revision happening, i dont expect details about the PS5 unless it releases in spring this year, but after jason saying its fall 2020 i gave up that idea (he also reiterated that in his podcast yesterday)
i choose an even 2000 mhz to prove that you can hit 14 tflops on a Navi 10 SE without going up to 2280 like you did.
You and i both have been in these threads long enough to know vega 7 comparisons are simply not valid for so many reasons.
Again, i asked you specifically to ignore the nvidia comparison because we all know nvidia tflops cannot be compared to amds. At least not yet.
What do you think the overall tflops is then? For the rx 3080. Or the Navi 10 SE?
You are making assumptions about Navi matching nvidia tflops performance due to architectural gains and yet completely ignoring the physical gains of going to a 7nm die. This is pc gpus we are talking about here. Why would they settle for a sub 2.0 ghz clockspeed on a 48 or 56 or 64 CU gpu? What's stopping them?
If the navi architectural improvements are really that amazing, they come on top of the clockspeed and cu improvements. Not at their expense.
How I think about CUs and clock speeds you can derive from my baseline prediction that is on the most recent state.The prediction is threadmarked on the first page of this thread. Please keep in mind that I think it can end up 10% higher if they "go for it". Please also keep in mind that my prediction is based on my educated guesses and information that is available to me. As we can see, other educated guesses can be quite different but I looked into it from several angles (cost, power draw, common technology patterns. current GPU market and available peak performances).
Last edited:
Don't know why people dislike Colbert's estimates lol. Like he said, it's based on what we know.
I want a 12tf console myself, but there's a lot of things that have to go right for it to happen. That 1800mhz gonzalo for instance has to take a battering from tests and still be in working order. What type of cooling do you need for that to happen?
I want a 12tf console myself, but there's a lot of things that have to go right for it to happen. That 1800mhz gonzalo for instance has to take a battering from tests and still be in working order. What type of cooling do you need for that to happen?
Christ what a last ten pages! To repeat myself, PS5 will not be 8tf...Im comfortable with my 12tf prediction, with Anaconda within 10% +/-. Just keep telling yourselves the tf number is not even half the story! We have all fancy new architecture in these things! Can you imagine some of the games we are going to be seeing?!
okay reading through the financial report, it sounds like the next gen console will still be in development this fiscal year for all those that hoped a march launch.
also sony will drop the price of the playstation 4 which indicates a super slim revision might be releasing this year.
also sony will drop the price of the playstation 4 which indicates a super slim revision might be releasing this year.
Guys, dev kits from both Sony and MS are now out in the wild with third party studios. It is only a matter of time before someone hits up Jason Schirer or Eurogamer/DigitalFoundry. Sony's "Hey, the PS5 exists and this is what we're aiming for" interview with Wired was just them heading off the inevitable leaks in the coming weeks. We'll know a lot more pretty soon.
Yeah, I meant to say dev kits ARE out in the wild right now. Typo on my part.
I don't remember but one of the last Digitalfoundry video has a slide with a strange 18 Gb RAM Don't remember if it is the Cerny reveal one or the AMD Gonzalo one.
Yeah I know, no worries
Still hoping for that HBM + DDR4 rumor to be true though.
Already plead guilty that your studio already has a dev kit ;) Don't be so shy ... ;)
Yes. This fall, $199 was the leak
That would be fantastic tbh...you thinking $499 for the PS5 though?
I'm still on the $399 camp, 10-12TF, 16-20 RAM, with Anaconda being 20% more of those and 25% more expensive.
I think, as Jez said, both Sony and MS know what the other guys are doing so thats why they're confident in being most powerful, and Cerny's comments about price were just a bit of misdirection.
$399 is to good of a price to pass up, the"sweet spot" of power/price some have alluded too.
Welcome back mate!
Yeah, I remember that. Do you have a link to it? There was much more about PS5, wasn't it?
Yes, about the HBM2 RAM iirc. Sorry I'm about to get on my car on the way to work but I'm sure someone else will :)
O, I know which one it was. It had lots of details about RAM.
Thanks for your answer and have a nice day:) Hopefully, this Day will be Gone before you know it;)
Regardless of actual TF numbers, having a new generation with new CPU's and SDD's will be a massive leap for the industry. Feels like we have been held back by poor performing CPU's for years now, even if they did make some cool stuff with it. Looking forward to a new generation that does not have to worry as much about the CPU being the weak link.
Anexanhume did a nice post about it;
https://www.resetera.com/threads/ne...anaconda-dont-want-none.112607/#post-20024749
for a watts limited, die space limited APU approach in a closed box - I think actual RT cores is a terrible use of precious real estate. Especially with XB1X and ps pro making it more difficult to impress with pure TF next gen, you'd potentially lose too much in regular rendering performance for the sake of bragging rights for a nascent feature. And risk that not being fully leveraged if it isn't adopted widely
I'm just thinking back to last gen teraflop predictions I saw on places like B3D, even after the XBO's 1.3 teraflop number leaked, almost nobody believed it until the very end, there was still quite a few 2.5 and 3 teraflops predictions.
If we have a proven insider telling us that some people are going to be disappointed, that pretty strongly implies we might want lower our expectations just a smidge.
Sony - No PS5 launch before May 2020
So let's say there's a trade off based on die space for hardware RT vs extra GPU CUs - with the extra TF being used for software RT in the latter, what sort difference in results would you get?
Ah well. At least we have something more concrete now. I just hope the wait brings us something good.
Well thanks a lot and a nice day to you too, always nice to wake up to kind words like yours. :)
This weekend is certainly shaping up to be full of excitement ;)
Question
Is It possible to use hardware RT cores for something different than pure ray, tracing ?
Is It possible to use hardware RT cores for something different than pure ray, tracing ?
You could do that by just tagging him but I have sent him a PM just to ruin his day.
This... this is where I am.
Well put, Lady Gaia.
RTX cores provide major benefits in accelerating BVH intersection tests, i.e. a part of the RT algorithm that the rest of the GPU isn't well suited to do.
So on that basis, while I agree with your general point that RT on the whole is primarily a general compute function, there may yet be other interesting areas that can be accelerated with fixed function hardware.
That 331mm² is for the whole GPU, including caches, i/o, FP64 gubbinz and HBM memory interfaces.
Tbf, however, if we compare PS4 to Pitcairn (HD 8870 with 20CUs), you're looking at a die size of 348mm² vs. 212mm² -- both on 28nm. I think Sony's 8 Jaguar cores at 28nm was around < 70mm², so that still leaves around 66mm² for all the uncore IO and interconnect gubbinz.
As such, I'd be surprised to see as much as 64CUs in PS5 or Anaconda's APUs, provided they stick with a max die area of < 360mm².
If they go with a chiplet design, however, it's a totally different story. Die-size limits on the more complex 7nm process may mean that both PS5 and Anaconda are going with a chiplet design.
The reddit rumour that talked about HBM for PS5 mentioned a 7nm PS4 made by Samsung.
hey Phoenix Splash
70 pages in less than a week. I don't think that this thread will hold until the start of E3.
70 pages in less than a week. I don't think that this thread will hold until the start of E3.
Hey, that is my job LOL.
No seriously, as I said earlier, I expect Navi to more efficient in games than their current GCN counterparts by around +10% to +15%, hence a Navi 9TF would perform like a 10.7TF AMD Vega 56. And if my estimation is too low, who cares, all will be happy it is better including me. Also: I expect a variance on my prediction, so I absolutely can see if somebody "goes for it" it can 10% higher for example.
With that info we know it's fall 2020 now. If 8TF seems low now, it will feel really low then and be a PR nightmare if Anaconda is much higher. I'm gonna go with 11-14TF for PS5 Nov 2020.
- Status