-
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
AMD's keynote at CES 2019 about 7nm products
OP
OP
There you go.
Tom's Hardware coverage and ongoing analysis of the event
Thanks for this link too SuperHans
Tom's Hardware coverage and ongoing analysis of the event
Thanks for this link too SuperHans
Also I think the main reason Sony has for subsidizing the PS5 is as an insurance vs Microsoft subsidizing the Xbox2, and viceversa.
Neither of them can afford not doing it and ending up with an sku with $100 less features/power
Indeed, I don't think they will go over the $100 though. Also, this time not only are they in a better position both financially and have momentum...but also they have the revenue stream of PS+, whilst it's lost money, I think if they can get from loss to even in about 1 year that is what they will aim for...essentially I'm expecting Sony to use PS+ to fund the first year of PS5 hardware losses.
Well, maybe not the final retail specs but yeah, one should assume they have a solid idea of what Sony plans to deliver. Also, Sony approached quite some people/studios/developers during the PS4 design proces (through surveys and also with Cerny visiting), and also had post-mortems; it would make sense that for PS5 they stick to this proces (since it worked out well for the PS4).
I doubt they pay more than $125-$150 for the CPU and GPU combined. I believe most industry experts pegged the PS4 APU at $100-$110 in 2013 to make
I can't see them spending more than:
$100-$150 on both CPU/GPU
$100 RAM
$30-$40 storage (1 or 2TB HDD)
$30 UHD BR Drive (already this cost more or less)
$20-$25 controller (so they can sell it at $70 like Switch Pro Controller, better rumble too)
While this seems low and like specs would suffer, Sony will get better deals than consumers since they can guarantee 50M+ orders and potentially 100M+
Xbox One X has a 40 CU (44 w/ 4 disabled) GPU @ 1172MHz on TSMC 16nm FinFet giving a 6TF peak performance.
It'd perfectly reasonable to see a PS5 with an 80 CU (88 w/ 8 disabled) GPU @ 1368MHz on TSMC 7nm giving a 14TF peak performance.
If anything I believe it'd come in a bit smaller and likely run a little cooler. Price is of course the main sticking point but 7nm is looking good, there is likely 4-16mths between now and the start of PS5 manufacture and the economies of scale will be very much in Sony's favour.
As for AMD's 64 CU limit, it's less of a hard limit and more a bottleneck; or a point of diminishing returns inherent to GCN as it currently stands in the 'PC' space. I doubt it'll be an issue when you take into account Navi/Next-Gen likely being the basis for a PS5 GPU, the single-spec closed box allowing optimisation for a specific task and the hardware customisations that are possible in this space.
Saying that, it'll be in AMD's interest to push past the 64 CU soft limit as a whole as soon as possible, when it comes to bang for your buck, there's a lot more to be gained in the GPU space by going wide as opposed to faster.
It'd perfectly reasonable to see a PS5 with an 80 CU (88 w/ 8 disabled) GPU @ 1368MHz on TSMC 7nm giving a 14TF peak performance.
If anything I believe it'd come in a bit smaller and likely run a little cooler. Price is of course the main sticking point but 7nm is looking good, there is likely 4-16mths between now and the start of PS5 manufacture and the economies of scale will be very much in Sony's favour.
As for AMD's 64 CU limit, it's less of a hard limit and more a bottleneck; or a point of diminishing returns inherent to GCN as it currently stands in the 'PC' space. I doubt it'll be an issue when you take into account Navi/Next-Gen likely being the basis for a PS5 GPU, the single-spec closed box allowing optimisation for a specific task and the hardware customisations that are possible in this space.
Saying that, it'll be in AMD's interest to push past the 64 CU soft limit as a whole as soon as possible, when it comes to bang for your buck, there's a lot more to be gained in the GPU space by going wide as opposed to faster.
I'm not sure he is. In a recent Kinda Funny interview he said he was creatively stuck on his next project, although he was hopefully coming out of it. But he also said last time (after GoW2) he remained stuck for 1.5 years, unable to work, in a "creative fog".
https://youtu.be/ZAmQkhOohp0 ,minute 40
Yes. Vega is in the lands of diminishing returns. You can see it when Vega 64 doesn't perfectly linearly scale up from Vega 56 results.
Navi is designed for "scalability", which presumably would address this.
Is anyone going to create a thread, for the AMD CES keynote? In context of our community (as a whole) and the upcoming console launches, this is way more important than Nvidia's Turing reveal.
Show starts 20 odd minutes from posting this.
OP
OP
I'd rather see someone making a thread after the keynote is over and the info is out. This could end up being a big blue balls show where the only thing we hear about is 7nm products arriving to the mainstream consumer market while the rest is info about enterprise devices, self-driving cars chips, machine learning and A.I. stuf... which is actually more important BUT we're not here to talk about that, at least not this section, haha.
I made a dedicated thread for the AMD keynote: https://www.resetera.com/threads/amd-keynote-at-ces-2019-january-9-9-00-–-10-00-am-pacific.92328/
I'd think a BOM of ~$500, a retail of $449 and a loss per SKU after shipping/retail of ~$100.
PS2 was a loss of ~$180, 360 a loss of ~$75 (ps3 was an oddball) , PS4 apparently a loss of $60. So initial losses have been dropping overall and node shrinks are now much more uncertain, so those all go against us.
But revenue from services, subscriptions and exceptional first party performance may provide the opportunity to push the initial loss per SKU back up a little and provide a stronger system.
The catch-22 of the uncertainty surrounding process node shrinks and memory prices at this time is that this generation could end up being longer as a matter of necessity. So do you save the money up front and gimp a whole generation or take a risk so that it's truly future-proofed?
OP
OP
PHIL SPENCER AMD EVENT APPEARANCE ALERT
Edit: PHIL SPENCER LEFT THE AMD EVENT ALERT
Yeah, this means the next Xbox is a lock to be fully AMD powered... now I wonder if it's Vega 2 instead of Navi powered hmmm
Edit: PHIL SPENCER LEFT THE AMD EVENT ALERT
Yeah, this means the next Xbox is a lock to be fully AMD powered... now I wonder if it's Vega 2 instead of Navi powered hmmm
Last edited:
On one hand, it's MS, not Sony, present as "Gaming" part of the AMD keynote. On the other hand: nothing about next gen hardware, just cloud-cloud-cloud. Surprisingly, nothing but general "thankyou-thankyou". I would expect at least some technologies/projects mentioned as a part of the collaboration.
Oh, now it's getting interesting...
Oh, now it's getting interesting...
HBM2, though. The MI50 scales to 32GB HBM2. The prior Vega cards have 8GB of HBM2. AMD doesn't have an existing Vega product with a GDDR5/X/6 memory controller, so its still up in the air.
There is no chance next gen is going to have Navi like I predicted it is going to be Vega and if the Radeon 7 (2nd gen Vega) only matches the 2080 which is 10.1Tflops I also predicted that next-gen consoles wont be 10Tflops as there is no chance there will be a full fat GPU that needs 3 fans and massive heatsink to cool it in a small console.
*edit* Yikes, $699.
*edit* Yikes, $699.
Like PS4 which is Polaris with Vega features, PS5 is going to be custom Vega, there's only been one rumour that Sony may be working on Navi with AMD.
Yeah, the fact they aren't talking about Navi in this conference worries me that it won't even tape out until next year.
Yeah this is a stopgap GPU for and to show off what 7nm can do. Navi is 100% going into next gen consoles.
Looks like 7nm Zen 2 consumer cpus won't launch until mid 2019
https://www.anandtech.com/show/13829/amd-ryzen-3rd-generation-zen-2-pcie-4-eight-core
Looks like we are getting chiplets though.
https://www.anandtech.com/show/13829/amd-ryzen-3rd-generation-zen-2-pcie-4-eight-core
Looks like we are getting chiplets though.
I hope radeon 7 put people in reality with next gen consoles.
14 tflops for $699
A $399 console with 10 tflops will be a miracle and the best we can have. Probably more around 8-9.
14 tflops for $699
A $399 console with 10 tflops will be a miracle and the best we can have. Probably more around 8-9.
Where are you getting 14Tflops. It's on par with RTX2080 which is 10.1 Tflops.
AMD tflops ≠ NVIDIA tflops
Radeon VII has around 14 tflops.
Sure... But never a console had a gpu equivalent to a $699 desktop one.
So 8-10 tflops is more likely.
If that's genuinely true, I guess that gives them an opportunity to talk about it and do previews at Computex. Let's hope that ends up being the case.
Radeon 7 with 4 stacks of HBM2, 16GB at 1TB/s bandwidth. That's driving a lot of the cost there.
Also note the pricing is very strategic. They mentioned 2080 as the competitor in their performance graphs, and that's exactly where they priced it. This is also 60 CUs instead of 64. Very deliberate.
The 25% performance boost for the same power is also exactly what they've been saying all along.
Also interesting that Ryzen 3 is 8C/16T chiplet with IO die. Next gen consoles could use off the shelf CPUs with custom GPU and memory controller integrated inside.
Also note the pricing is very strategic. They mentioned 2080 as the competitor in their performance graphs, and that's exactly where they priced it. This is also 60 CUs instead of 64. Very deliberate.
The 25% performance boost for the same power is also exactly what they've been saying all along.
Also interesting that Ryzen 3 is 8C/16T chiplet with IO die. Next gen consoles could use off the shelf CPUs with custom GPU and memory controller integrated inside.
How Flops are measured is not the same between AMD and Nvidia. Vega R7 can have higher TF number but still perform in games comparable to RTX2080 with 10TF. It all comes down to how games makes use of the architecture and different bottlenecks inherent to a given architecture.
That is unlikely just look at it PS4 was 1.8TF and 4 years later XB1 was 6TF.
Now with another 4 years , new node shrink and architecture we should see a bigger jump.
Not huge or anything but a little bigger .
Also this depends on price of next gen systems .
Vega 2 isnt the same as Navi. Vega 2 is just a cash grab attempt at getting people to pay until Navi releases in july.
Navi was made with an architecture to fit all these improvements, the performance will be much better for the price.
Flops are measured the same. It's a theoretical maximum throughput of the arithmetic units in the architecture. The issue is that architectures are much more than ALUs, and efficiency is a huge deal.
Yeah,it seems tech is not ready yet.
No word at all today about Navi is,frankly,disappointing.
It looks to me that 7nm Zen2 is also way off...
I think the fact that Ryzen 3000 uses an IO die is good news in the sense that there's no performance degradation should Microsoft or Sony decide to do a GPU + memory controller die. They can feed the CPU over IF at up to 100 GB/s.
It could certainly save engineering cost to decouple the CPU/GPU development, especially if they can use off the shelf Zen 2 chiplets.
I think it's more likely PS5 delayed Navi, not the other way around.
I think the fact that Navi and Ryzen 3000 are both slated for mid 2019 shows that's where the economics for mass market consumer products make sense for 7nm HPC.
It could certainly save engineering cost to decouple the CPU/GPU development, especially if they can use off the shelf Zen 2 chiplets.
I think it's more likely PS5 delayed Navi, not the other way around.
I think the fact that Navi and Ryzen 3000 are both slated for mid 2019 shows that's where the economics for mass market consumer products make sense for 7nm HPC.
- Status