Baseless Next Generation Rumors with no Technical Merits [pre E3 2019] *spawn*
- Thread starter Arkham night 2
- Start date
-
- Tags
- non sequitur
- Status
It should be though. I have largely struggled with this dual SKU business but if we take the numbers at 12 and 24 then it looks proper. The numbers aren’t rational, I get that but the positioning is. As a product manager 12 and 24 make a much better lineup.
Consider the following new consoles
SAD - 199
1S - 249
1X - 399
Lockhart - 499
xcloud - ?
Anaconda - 799+
The price point and power curve have something for everyone but more importantly there is very little overlap for the customers, the product line is simple. You pay more you get more.
We go back to 6-12 things get very crowded
2 products at 6 TF, 2 products at 12 TF. You introduced SAD this year that is only a little worse than the lowest product but is lower than 1S in pricing.
It’s just confusing to consumers
I don’t know if 12-24 is realistic, but the product line is much better spaced out and will hit a large variety of customers from the ultra low end to the ultra high end.
I would be okay with clustering around 6-12 if they had more products beyond 12. But consoles aren’t like GPUs where they are servicing the whole range. But 12-24 would be appropriate to service the whole range without having to introduce more consoles
And Xcloud being the same power as the base SKU makes sense. You need more of them, to service more players.
Last edited:
24 teraflops. Realistic. When is this supposed console coming out? It's something like twice the flagship performance of current GPUs which are at about the limit of cost-effective lithography so you aren't going to get 24 TFs on 7nm. So you're looking at a bleeding-edge, biggest-possible 5nm chip in 2021. The sort of chip sold for $1000 as a desktop GPU going into a console. Is the business strategy a $500 loss leader? Or to sell 30,000 of these in total, like exotic sports cars?
Or are we talking a 2024 release when 24 TFs will be affordable?
Silent_Buddha
Legend
It's a dual GPU...no...no...dual SOC SKU.
2 consoles in one, the ultimate local multiplayer machine!Regards,
SB
Sorry that was supposed to be rhetorical. The product line up is what mattered to me more. Don’t care if it’s 9-12. Or 12-15. Just don’t clog up the product line.
Whether 24 is realistic; perhaps not at launch? Or perhaps if they dual GPU/SoC.
Which they have the API stack to support.
But. Yea. I’m not putting any money down there.
From the perspective of product however, it’s desirable to go with 12/24. Provided you keep the line up all the way down to SAD for next gen and slowly phase out the backend.
When I see MS moving all their games to PC, steam, and XCloud the console itself still needs to compete in those areas. Let X1X and SAD handle your lower end markets. Have Lockhart and anaconda compete against the PC space.
Lockhart at the mainstream PC space, Anaconda at the enthusiast space. MS will be able to make a cheaper console than what PC builders can make and the play, “thanks MS, now I don’t need an Xbox” isn’t so straight forward anymore.
They have to keep the last gen alive for this to work though. They will need the full range for this to work.
Last edited:
If we consider transistor performance between 7nm and 16nm it is really possible.
7nm CMOS transistors have over 3x logic density and 39% power consumption of 16 nm transistors.
If x1x GPU expands 2.5~3 times at 7nm then the large GPU should have similar die size and power dissipation of 16nm x1x GPU. If AMD can further increase GPU efficiency around 40% then we may have 4x performance of x1x.
Considering 7nm transistor performance it seems that AMD hasn’t really taken full advantage of 7nm process. If NAVI really consumes 180W then the performance per watt may less than 1.5x of GPU of x1x. However Nvidia GPUs have much better efficiency and 7nm Nvidia GPU may really have 4x performance of Xbox one x GPU.
What's possible? 24 TF in a console at what price and what release date?
Last edited:
Since this is the baseless thread - has there been any word on which foundry MS or Sony will be using? I have been assuming TSMC 7nm or 7nm(+) - partial EUV. (The whole Nvidia going to EUV Samsung 7nm and an old rumor about a PS4 slim going to Samsung EUV 7nm is making me question my assumptions.)
TSMC i think is who AMD will be partnering with.
If we're talking numbers as large as 24 TF. Then you need to throw out the small features. It's going to be about big price or very far in the future.
My take on the 2x and 4x gains is that these are actual performance gains when taking the entire system into consideration for their internal benchmarks and they're not just GPU flop multipliers.
And if that's the case, I would still expect the lower end SKU to be faster then the X1X simply due to a much better CPU and a much more efficient GPU.
And if that's the case, I would still expect the lower end SKU to be faster then the X1X simply due to a much better CPU and a much more efficient GPU.
No one that's rooted in reality at all is talking about 24 TF.
D
Deleted member 13524
Guest
24TFLOPs would be easily achieved if AMD went back to VLIW5 Terascale 2.
The HD5870 Cypress XT did 2.72 TFLOPs with only 2.1B transistors at 850MHz. At ~1.5GHz the same GPU would do (2.72/0.85) * 1.5 = 4.8 TFLOPs.
So a GPU 5x larger than Cypress XT with ~10.5B transistors at 1.5GHz using Terascale 2, AMD would be able to do 4.8*5 = 24 TFLOPs.
AMD is going back to TeraScale 2. Case closed.
Good luck getting any modern shader running efficiently in that, though.
The HD5870 Cypress XT did 2.72 TFLOPs with only 2.1B transistors at 850MHz. At ~1.5GHz the same GPU would do (2.72/0.85) * 1.5 = 4.8 TFLOPs.
So a GPU 5x larger than Cypress XT with ~10.5B transistors at 1.5GHz using Terascale 2, AMD would be able to do 4.8*5 = 24 TFLOPs.
AMD is going back to TeraScale 2. Case closed.
Good luck getting any modern shader running efficiently in that, though.
It's a GPU that doubles as a whole home heater and BBQ set.It's a dual GPU...no...no...dual SOC SKU.
Regards,
SB
D
Deleted member 11852
Guest
I am really interested to see what magical unicorn RAM and memory bus will provide sufficient bandwidth to feed a 24Tf GPU.
Did I miss a memo? Was today do-drugs-and-post day?
cheapchips
Veteran
4x S = 4.8Tf
2x X = 12Tf
Would gel with one of the Lockhart/Anaconda rumours.
2x X = 12Tf
Would gel with one of the Lockhart/Anaconda rumours.
Pretty sure the base og XO was 1.3 TF?
If so 1.3 * 4 = 5.2 TF
+ 15% efficiency = 5.98 TF
15% is my own figure but I suspect that's on the low side, also it's not going to be precisely 4x, so I expect the GPU on Lockhart to be more powerful than 1X even without taking advantage of new features.
Even at 0% would just need to be 4.6x
That's the reason the 4x sounds more like gpu, than overall system performance.
Zen, storage, add those in would be able to say much more than 4x.
Bored, thought I search some Reddit pastebin speculation and nonsense.
Speculation wise, I could see another Sony/Rambus partnership happening, not so much the HBM3 (maybe HBM2 for such a memory setup). And the last pastebin seems not out of this world, however, the supposed APU sizes don't seem plausible towards reasonable wattage/TDP in the console space (I could be wrong though).
https://pastebin.com/dd1WUpjb (2 or more of these)
No salt required.
- Both are monolithic APU designs.
- Midrange CPU desktop clocks.
- Navi is an architecture. RDNA is a philosophy towards a more flexible and customizable core implementation design. RDNA can be implemented in prior or undisclosed architecture designs outside of the Navi architecture.
- MNXB: CPU is multitier architecture oriented for large AI and machine learning datasets. Larger L1 cache.
- MNXB: CPU takes up a good portion of the APU design.
- MNXB: GDDR5 memory (Hynix) and PCIe 4 busing system.
- MNXB: GPU sparse rendering acceleration.
- SPS5: GPU has a distinct advantage in feature sets and performance.
- SPS5: HBM3 memory and busing system (Rambus partnership)
- SPS5: SSD advantage in terms of performance.
https://pastebin.com/12Yj1KT8
System 1 APU ~525mm²
System 2 APU ~534mm²
More to come.
prior info https://pastebin.com/09ddeXvn
https://pastebin.com/09ddeXvn
Both the core systems have monolithic APU designs. One favoring more CPU space usage for larger cache sizes and the other favoring more GPU space for an additional but smaller non-AMD GPU based co-processor. More to come.
Speculation wise, I could see another Sony/Rambus partnership happening, not so much the HBM3 (maybe HBM2 for such a memory setup). And the last pastebin seems not out of this world, however, the supposed APU sizes don't seem plausible towards reasonable wattage/TDP in the console space (I could be wrong though).
https://pastebin.com/dd1WUpjb (2 or more of these)
No salt required.
- Both are monolithic APU designs.
- Midrange CPU desktop clocks.
- Navi is an architecture. RDNA is a philosophy towards a more flexible and customizable core implementation design. RDNA can be implemented in prior or undisclosed architecture designs outside of the Navi architecture.
- MNXB: CPU is multitier architecture oriented for large AI and machine learning datasets. Larger L1 cache.
- MNXB: CPU takes up a good portion of the APU design.
- MNXB: GDDR5 memory (Hynix) and PCIe 4 busing system.
- MNXB: GPU sparse rendering acceleration.
- SPS5: GPU has a distinct advantage in feature sets and performance.
- SPS5: HBM3 memory and busing system (Rambus partnership)
- SPS5: SSD advantage in terms of performance.
https://pastebin.com/12Yj1KT8
System 1 APU ~525mm²
System 2 APU ~534mm²
More to come.
prior info https://pastebin.com/09ddeXvn
https://pastebin.com/09ddeXvn
Both the core systems have monolithic APU designs. One favoring more CPU space usage for larger cache sizes and the other favoring more GPU space for an additional but smaller non-AMD GPU based co-processor. More to come.
Last edited:
- Status
