Next Generation Hardware Speculation with a Technical Spin [pre E3 2019]
- Thread starter AlNom
- Start date
-
- Tags
- 5 bytes non sequitur
- Status
If we take what Cerny said at face value the SSD speed would need to be north of 3GB/s to be in range of "faster than pc". Getting the speed via some kind of cache would not be faster than pc as for example samsung has software where one can use large amounts of ram as cache and the speed for stuff in cache is ridiculous. Getting faster than pc speeds on low capacity drives is near impossible. The individual nand chips are only so fast and to get 3GB/s+ speeds you need many and more chips and do concurrent read/write operation to many nand chips. That is at least if traditional off the shelf nand is used. Low capacity drive wouldn't have enough chips==parallelism to be as fast as what cerny claims ps5 solution to be.
If sony somehow designed custom nand chip that doesn't obey the established rules or if sony is using intel optane all bets are off. Completely custom solution might lead to really fast small capacity drive.This could catch microsoft by surprise if microsoft is using off the shelf parts.
This is same reason why you would see same drive have different performance if you look at very small capacities like 256GB versus larger capacity like 1TB.
There is point of diminishing returns as typical controller can only do so many parallel operations. Very large drives might not be able to fully utilize what is possible in theory as there is more chips than what controller can use at any given time. Also pci-e bus might be limiting factor even if controller was up to the task.
If sony somehow designed custom nand chip that doesn't obey the established rules or if sony is using intel optane all bets are off. Completely custom solution might lead to really fast small capacity drive.This could catch microsoft by surprise if microsoft is using off the shelf parts.
This is same reason why you would see same drive have different performance if you look at very small capacities like 256GB versus larger capacity like 1TB.
There is point of diminishing returns as typical controller can only do so many parallel operations. Very large drives might not be able to fully utilize what is possible in theory as there is more chips than what controller can use at any given time. Also pci-e bus might be limiting factor even if controller was up to the task.
Last edited:
RTX 2080 only has 448 GB/s VRAM. Will 400 GB/s become bottleneck for a console GPU which is not as powerful as RTX 2080?
anexanhume
Veteran
Posted earlier. A Digitimes report has emerged that alleges ASE will handle packaging, in direct contradiction to this rumor.
https://www.digitimes.com/news/a20190416PD205.html
Architectures can have different sensitivities to bandwidth.
Oh it's a nice one!
Good homework on technology and trying to fit with the most recent rumors and speculations. It's a perfect average of all rumors without anything new added, which is either a tell tale sign of a fake, or a sign we are amazing at predicting everything (which never happened)
Good homework on technology and trying to fit with the most recent rumors and speculations. It's a perfect average of all rumors without anything new added, which is either a tell tale sign of a fake, or a sign we are amazing at predicting everything (which never happened)
That’s purely for the video card however. It’s not a shared pool and doesn’t suffer from async bandwidth issues. (Not even sure if this is a thing next gen tbh). So arguably there needs to be a lot more bandwidth on a console to feed both CPU and GPU sufficiently in the worst case performance scenarios.
Without knowing the specs of CPU and GPU. It’s hard to tell. But expect it to be a lot
Eh mechanical drives are all but gone from pc purchases. They exist as outliers in the market mostly in desktops or as secondary drive . The majority of ssds give 500MB/s speed . But over the next few years with NVME m.2 drives hitting $100 bucks for 3GB/s drives at 1TB capacities they will take over the gaming sector.
4k is almost pointless for native rendering, but 8k rendering would be the dumbest choice ever. Something sub-4k will still be the best option for performance, quality and to match the tvs and screens people use.
Last edited:
Only questioning your statement about what screens people use, consider that the PS5 won’t be launched until the end of 2020, and remain the base platform for much of that decade. 4k TV sets have already dominated retail for years here and is in a position now that is shifting from being a step up from HD to being a step down from 8k.
I don’t think that the PS5 will be connected to many HD-TV sets in 2025.
DuckThor Evil
Legend
Well my point and critique of your previous calculations wasn't just based on taking some stuff out the Vega 20 die. The Navi CUs, their capability and what they include were and still are quite unknown at this point. It's a new architecture that looks to have a different focus than Vega. The CUs could have seen a major structural change, completely changing the equation that should be used for the calculations. We simply don't have a valid information on the capabilities, power consumption, and transistor amounts of a Navi CU. Who knows if they are even called CUs anymore?
The 52 CU number alone doesn't really say much without having more specific information about them. Let's say for arguments sake that Navi CU has 50% more transistors than a Vega 20 CU, but the number in the console chip ends up being 52 CUs, was your math correct then? IMO not. If the Navi CU has double the transistors, but the console has 26CUs, then your original math was mostly correct even though the CU count is quite off.
I just think there are a few too many moving parts to put your chips on a number.
Love_In_Rio
Veteran
Wow!. I was a doubter, this should confirm it.
So torn. On the one hand, I have to 'like' if the ND dev is saying there's hardware RT. On the other hand, he's a lighting artist, not a lightning artist, and people keep making that mistake and it does my head in!!!
Leave the lightning to the costumers, please. ;-)
Last edited:
That assumes 2080 is optimal for BW. What if 2080 is actually lacking BW and could do with way more to not bottleneck on some workloads? If you look at BW per flop over the years, its reducing:
"Memory bandwidth has always been a challenge for video cards, and that challenge only continues to get harder. Thanks to the mechanics of Moore’s Law, GPU transistor counts – and therefore the quantities of various cores – is growing at a rapid pace. Meanwhile DRAM, whose bandwidth is not subject to the same laws, has grown at a much smaller pace."
Considering the tech of HBM is supposed to provide ludicrous amounts of bandwidth and solve this issue, to have an implementation that's only so-so is definitely a disappointment.
But a memory bank can't be accessed simultaneously by the CPU and GPU, There will be memory contention and the overall memory bandwidth available to the GPU will decline disproportionately higher than the amount the CPU accesses.
The Tradeoffs of Fused Memory Hierarchies in Heterogeneous Computing Architectures:
https://ft.ornl.gov/~lees2/papers/CF12.pdf
Its not a limitation unique to PS4 and Xbox One, its a fundamental downside of unified (aka fused) memory.
It can be mitigated somewhat by increasing the number of memory banks, increasing cache sizes, increasing the bandwidth to compensate.
Rootax
Veteran
The thing about the loading time, it's often taxe the cpu too, no ? At least I saw that on PC, I guess a lot of decompression, shaders compilation, etc. So my guess is loading time this fast is a best case scenario, where everything is already "available", without decompression or compilation needed...
Naughty Dog must know Sony's plans and hardware. It could as well be that the guy was told to post a retraction because Sony aren't ready to talk about the hardware beyond what's been said. It's not really proof nor disproof, but I'm inclined to believe that the initial response had more insight than just the Wired article one-liner.
True but if the trade off is lower BW for less cost perhaps we're seeing the projected savings used to subsidize the SSD or CPU/GPU budget.
- Status
Similar threads
