-
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.
-
We have made minor adjustments to how the search bar works on ResetEra. You can read about the changes here.
Next gen PS5 and next Xbox launch speculation - Secret sauces spicing 2019
- Thread starter Phoenix Splash
- Start date
- Status
Or 50% less fat (quite a dirty trick, because then they pump it with sugar to improve flavor and then you end up with something even more caloric). Or 50% less sodium. :D
Wild guess is lower clocks and weaker FP64 perf (say, 1/8 FP32 vs 1/2 or 1/4).
Last edited:
Wait, you love the numbers because the arcade would be DOA? Or is that a separate thought? Regardless, I think those specs with that price difference would turn out fine for them. Game changing? Don't know, don't care but DOA seems like crazy hyperbole.
There is the possibility that Navi has more than 64CUs, lets say 80-96CUs for "Navi 10" and 104-128CUs for "Navi 20".
So having a cut down Navi10 design for APUs makes sense. 60-72CUs most likely, if full-fat Navi10 is 80-96CUs.
Though there is a small possibility that they might lower the amount of unified shaders per CU from 64 to 48.
If this happens, we'll need 72CUs@1450MHz to reach 10.02TFlops.
Last edited:
that user is saying he would love if the numbers would be so high, however they feel that it will hurt lockhart's sales.
For less than the price of a game you get a better system. For launch day buyers, $100 is the increment where people start to do the pros and cons about what to get. But $50 is nothing when you're already spending $350. It's like spending $500 000 on a house and getting squirelly over $2000 for hardwood in the basement. It doesn't make any sense. The 1s is routinely cheaper, it is weaker and even has a 4k BluRay player. It is always outsold by the ps4 slim. Bar the 4k, it is the exact situation you are outlining here. Cheaper and weaker....and yet it is outsold. When you especially consider the whole point of that SKU is to drive volume sales, $350 makes no sense. $300 and it's a debate.
Thank you. The specs across the board I would love. The price for the arcade makes no sense
Got it. I still think DOA is hyperbole and theatrics but that's our community; you're either number one or you're dead and buried. Sony will be going into next gen with even more momentum than they had going into PS4 but MS isn't about to fuck up every single possible part of the Scarlett launch like they did last time. So while I expect Sony to lead in sales from the gate I don't think that an Xbox Arcade specced/priced like that would be considered a failure or DOA.
How does he know Navi desktop GPU wuld run at ~2Ghz?? Sounds like pure speculation to me...
It's far more common for GPUs to keep clock speed increases small to nonexistent across manufacturing nodes while focusing on increasing parallel execution units. Thermal constraints are your ultimate limiting factor, and you invariably gain more by adding CUs than bumping clock speeds within the same thermal budget.
All subject to architectural limits and manufacturability, of course, so we'll see what the right balance is for Navi ... but I'd be shocked if they pushed clock speeds. That's the game you play when you've hit architectural limits or exploitable parallelism and have no other way to improve performance.
It is, but it's educated. Vega VII boosts to 1800MHz and Turing goes near that high already at 12nm. If Navi is the least bit more efficient than Vega, you're at 2GHz.
Last edited:
I fully agree with that. I'm going with what Richard on DF / Eurogamer has said, until a new video or article comes out with different expectations.
Anyway, I just cannot imagine a PS5 in 2020 (even at $399) with 2 more TFlops (8TF) and 4GB more RAM (16GB) than Xbox One X had in 2017 (even tho that was $499).
Sony knows full well what they did to Microsoft in 2013, and have to be expecting MS to come out with guns blazing in 2020. They'd have to be more than blind and stupid to come out with a PS5 that has the perception of underwhelming and disappointing.
If the rumors are true that Sony originally planned to have PS5 out for Holiday 2019, but pushed that out 1 year to Holiday 2020 (and that would not have been a recent move, but more like early or mid 2017), then I hope that provides a boost to final performance of PS5, even if we never get to know what happened documented.
Pascal and Turing both do 2000mhz relatively easily. Nvidia arch has been hitting much higher clocks than AMD for a long time now, but AMD has managed to somewhat keep up outside of nvidias halo chips despite the lower clocks (not really, as catching nvidias x80 series chips has meant for AMD huge 400nm+ dies and a lot of flops vs smaller Nvidia mid-range dies with lower flops). 980 tis could hit 1500mhz back in 2015. Since Pascal though nvidia has been somewhat stuck at 2000mhz, so it will be interesting to see what kind of clocks AMD can indeed achieve with a non Vega chip/possibly not GCN architecture.
But how big boost?If PS5 was planned for Q4 2019 ,and maybe delayed for whatever reason, that means PS5 specs were locked long time ago.What can they possibly change after that with ~1 year delay?Only realistic option i see is to raise CPU/GPU clocks a bit,like MS did with 1X that launched 1 year after Pro.But that also required expensive cooling solution and i highly doubt Sony would use something like that in $399 box.
They might scale down the "shader engines" slightly, this might be one of the things they'll do to improve the scalability. So instead of 4 "normal shader engines", they'll go for 6 or 8 slightly smaller shader engines.
This would also make them easier to feed data, which is one of the issues that prevented them from going above 4 engines in the past. (according to an interview.)
Last edited:
so... i know insiders over here told us to ignore that user, but i could not help it and looked over at osiris's post again in the old forum about his PS5 "leak",
it sounded pretty crazy, but now i don't know anymore.
first, it accurately predicted the CPU speed from the gonzalo leak at 3.2GHZ.
but then the second part sounded unbelieveable, a 2.1GHZ GPU. even NVidia's fastest GPUs do not get that speed i think, but now we have komachi saying they think Navi 10 will be around that range... what if its real?
that last part also sounded very weird, 18GB GDDR5.
it sounded pretty crazy, but now i don't know anymore.
first, it accurately predicted the CPU speed from the gonzalo leak at 3.2GHZ.
but then the second part sounded unbelieveable, a 2.1GHZ GPU. even NVidia's fastest GPUs do not get that speed i think, but now we have komachi saying they think Navi 10 will be around that range... what if its real?
that last part also sounded very weird, 18GB GDDR5.
True - we'll have to see if AMD can push past the 64CU limit on GCN. So many questions remain unanswered right now until they outline Navi.
I'm basically with you on everything except the GPU. Even the 1tb SSD sounded crazy until we found out about the plans MS has. The Navi 10 clocks are most likely desktop, which are not going to happen in console
Yes, that's what it means in this context. It doesn't tell us whether it's monolithic or chiplet, however.
I am Team 12TF. With a twist.
PS5 ($500 console sold for $399)
CPU Ryzen 2 8C(16T)@3.2Ghz
GPU 72CU [email protected]
RAM 16GB GDDR6 @ 512GB/s + 4GB LPDDR4
STORAGE 240GB embedded Nand flash + 2TB HDD
XBA ($500+ console sold for $399)
CPU Ryzen 2 8C(16T)@3.5Ghz
GPU 72CU [email protected]
RAM 16GB GDDR6 @ 512GB/s + 4GB - 8GB LPDDR4
STORAGE 240GB embedded Nand flash + 2TB HDD
XBL ($400 console sold for $299 the 1080p/1440p console)
CPU Ryzen 2 8C(16T)@2.5Ghz
GPU 56CU 7TF@1Ghz
RAM 12GB GDDR6 @ 384GB/s + 4GB LPDDR4
STORAGE 120GB embedded Nand flash + 1TB HDD
PS5 ($500 console sold for $399)
CPU Ryzen 2 8C(16T)@3.2Ghz
GPU 72CU [email protected]
RAM 16GB GDDR6 @ 512GB/s + 4GB LPDDR4
STORAGE 240GB embedded Nand flash + 2TB HDD
XBA ($500+ console sold for $399)
CPU Ryzen 2 8C(16T)@3.5Ghz
GPU 72CU [email protected]
RAM 16GB GDDR6 @ 512GB/s + 4GB - 8GB LPDDR4
STORAGE 240GB embedded Nand flash + 2TB HDD
XBL ($400 console sold for $299 the 1080p/1440p console)
CPU Ryzen 2 8C(16T)@2.5Ghz
GPU 56CU 7TF@1Ghz
RAM 12GB GDDR6 @ 384GB/s + 4GB LPDDR4
STORAGE 120GB embedded Nand flash + 1TB HDD
PLEASE don't use anything to do with Blue Nugroho! He is Mister X Media's "Technical Gentleman" otherwise known as "mistercteam"...
Good to know, thanks! I have a question. Is their a number teraflop wise that studios in general seem to be expecting? Or even want? I completely understand if you can't answer.
I was reading an article the other day that highlighted something about chip design - GPUs, especially - that I hadn't really thought a lot about: the power (and heat) cost of moving data around. Per article, the energetic cost of shunting stuff around has been decreasing slower than the power used by the logic itself, while datapaths have been increasing in width and length. The net result is that a proportionally larger part of the power budget is being consumed not by logic, but just by moving data around.
So, in thesis, what you want to do is to keep data as local as possible when working on it. And that's one area where something like Super SIMD will shine, because it will make it possible to queue a bunch of operations on some data that will be executed sequentially without the data ever leaving the ALU at all - and, even further, since ALUs are paired, you can have one feeding the other directly, say, one running a multiply and the other evaluating the result, with no round trips. That would mean that Super SIMD is not only more efficient from a workload perspective, but also from an energetic perspective.
Also makes me wonder if the current logic arrangement for AMD GPUs isn't partially to blame for suboptimal power usage.
So, in thesis, what you want to do is to keep data as local as possible when working on it. And that's one area where something like Super SIMD will shine, because it will make it possible to queue a bunch of operations on some data that will be executed sequentially without the data ever leaving the ALU at all - and, even further, since ALUs are paired, you can have one feeding the other directly, say, one running a multiply and the other evaluating the result, with no round trips. That would mean that Super SIMD is not only more efficient from a workload perspective, but also from an energetic perspective.
Also makes me wonder if the current logic arrangement for AMD GPUs isn't partially to blame for suboptimal power usage.
Mmm, someone high up at AMD? Maybe less than 100 highly paid engineers and their managers. Possibly c-suite at EA/AB/Ubi, etc.
Yup. Trace resistivity is not scaling, so IO becomes more expensive. Intel tried to innovate here on 10nm with Cobalt but other vendors have said Cu is still good until 7nm. The issue is the buildup layer around the Cu to combat electromigration is not as conductive as the Cu itself.
The spirit of your post is somewhat reflected in the latest Cerny patent. It repeatedly talks about manipulating data in local caches. The Turing architecture also upped cache sizes from Pascal.
https://patents.justia.com/patent/20190035050
Last edited:
Please delete everything you get from this idiot blue whatever ... He is the guy came up with the idea that the Xbox One has a hidden 2nd GPU.
Last edited:
GDDR5 is not really weird since they can get almost 1tb/s of bandwidth by using a 512 bit bus. I'd imagine procuring GDDR5 would be far cheaper than GDDR6 or GDDR5x but then again, that assumption could be wrong.
I wonder how long current gen consoles will still be supported, it will be a shame for games that could work on the mid gens don't come to them.
Also I hope next gen any Bluetooth headphones or earbuds will work with the console.
Also I hope they both can turn on my TV when I turn on my console like the xbox one does now.
Also I hope next gen any Bluetooth headphones or earbuds will work with the console.
Also I hope they both can turn on my TV when I turn on my console like the xbox one does now.
Yeah,Putty just told me,i had no idea who that is.
Btw,how is your new power hungry toy?
I am surprised that I can see his tweets here as I have blocked this guy on twitter to not see the nonsense he is usually spreading ...
Nope, max 512 gb/s on 512 bit bus. GDDR5x is max 640gb/s on 512 bit bus.
Well Sony likes to have their 10 year plans, so I wouldn't be surprised to see the PS4 still receive some titles up until around then. But I'm sure both companies will push the incentive of playing their next gen only titles via cloud streaming.
could be that 8tf was an overestimation then. a first party studio can get more out of 8TF than other studios. I agree that on paper it doesn't sound like a big jump from Scorpio. There would definitely be a visible difference but likely not big enough.
Oh, I see. Thanks for the explanation, it makes total sense. I had read about Intel going for Cobalt for the traces and didn't really get the reasoning behind that (and also recall reading it didn't go so well for them, they had lots of issues with that, like failures due to thermal cycling).
Oh, shiny. Hadn't seen this latest patent yet. Gotta take some time to read and digest it, looks dense.
Also, cool to have some validation that my reasoning isn't completely busted :D
Because Phil said so.
No, really, the reasoning behind that is that Phil said so.
Forgive me I haven't been following the next gen hardware news as much recently but did he explicitly say it'll be the most powerful console next gen?
- Status