Current Generation Hardware Speculation with a Technical Spin [post GDC 2020] [XBSX, PS5]

Status
Not open for further replies.
Does anyone else think that when DirectStorage comes to PC's we'll start getting "DirectStorage certified SSDs" with special hardware to reduce the CPU I/O overhead??

DirectStorage could be a way for MS to easily indicate SSD hardware requirements and standards.

DX12U certified clearly indicates a min featureset your GPU supports and sets a standard
DirectStorage certified clearly indicates SSD features/capability and sets a standard
I don’t believe that you need a direct storage certified SSD to use it. It’s just an API to with less overhead to access storage.
 
Enthusiasts easily get disconnected from reality ;)
Even with it's flaws in certain areas, Steam Hardware Survey should give quite realistic picture of gaming machines on easily measurable factors like installed memory amounts.
Currently 16GB is barely more popular than 8GB, under 7 % of machines has more than 16GB of RAM.
8GB VRAM is more realistic, it's 2nd after 6GB, but under 4% have 11GB, even 3GB is more popular than that, not to mention 4, 2 and 1 GB options which are the next most popular amounts after 6 and 8 GB.

Steam can be installed on any old machine. That doesn't mean it's typically used for current console market gaming.

From the point of view of considering what kind of port the PC will get from a given next gen console game, those lower spec'd machines shouldn't factor into the equation as they're below the majority of today's minimum spec requirements for a non indie multiplat game. They basically wouldn't factor into developers decision making process with regards to what kind of compromises to make for the PC port.

Rather than look at the percentage split of the entire steam hardware base, developers are more likely to look at the raw numbers at or above a given minimum hardware level. E.g. how many systems out there have at least say 16GB RAM, a fast quad core and a 4GB DX12 capable GPU, and how does that number compare to the potential next gen console install base in the target timeframe?

The port will have to scale down to those machines and then the level of effort they put in to provide an equal or better experience to the base console game would be determined by what percentage of that already highly limited subset of machines is capable of re-producing or exceeding the full console experience.

So rather than the decision being based on say 2% of all PCs on steam featuring 32GB system RAM and an SSD, the equation may be more like 20% of the actual target market which might equate to in actual unit terms say 50% (purely made up number) of the expected console target base at the time of the games launch.
 
I don’t believe that you need a direct storage certified SSD to use it. It’s just an API to with less overhead to access storage.
Well we know that it's an API with less overhead.. but I feel like MS could use this as an opportunity to set some kind of standard going forward for storage which is easily communicable to gamers.

They gave an example of 3 full Zen 2 cores for software decompression as well as 2 for I/O overhead. On a PC you'd need 13 cores (7 game + 3 decompression + 2 I/O + 1 OS) On the Series X with their Velocity Architecture with DirectStorage and their decompression block it reduces that decompression and I/O overhead to just a fraction of a single core.
 
Yea I’m not sure where the decompression hardware sits. If it’s on the Hard drive itself than I think you have a point about needing certification.
 
Steam hardware survey numbers aren't reflective of the type of consumers that would be buying a next gen console. The PC market is filled with entry level machines that are designed to surf the web and read emails, and steam will install on them just fine, and they will play old titles or things like seek and find games as well. But that doesn't make them reflective of the current premium gaming market.
 
Yea I’m not sure where the decompression hardware sits. If it’s on the Hard drive itself than I think you have a point about needing certification.

Not much point in the decompression hardware sitting on the SSD itself unless it's an older/slower drive since by the time the XSX launches there will already be SSD's that can near saturate PCIe 4.0 bandwidth with raw throughput alone.

So why buy a slower drive with decompression hardware that won't always work when you can buy a faster drive without it at a potentially similar price that will perform as well or better.

Also wouldn't it be difficult to manage this kind of super high compression in the PC space as the install itself needs to accommodate systems without it? So unless I'm talking rubbish here wouldn't you need multiple install options depending on whether you have hardware decompression capabilities or not?
 
Last edited:
How much latency takes decompression in hardware? How many mm^2?
The ideal solution would be (talking about pc) keeping it compressed on the main ram, and unpacking on the fly when it moves through the cpu and pcie4 link to the gpu.
Do you have no hardware zlib/shrinkray? No problem! Finally, you'll find a use for that 16 thread cpu that you brought for gaming! (Compatibility fallback)
Do you have hardware shrinkray? Pay attention, please. Any misuse will not be covered by the guarantee.
 
Pretty sure the decompression hardware sits as a PCIe device, that is after the SSD.

Maybe a transition period where the chip is external to the CPU and sits on the PCIe bus between the CPU and a dedicated "DirectStorage" (or a more open standard equivalent) slot on the MB? Eventually, I think it probably gets integrated into the CPU, though.
 
It's not possible to have this decoder chip build into the SSD?

PCIe 4.0 4x is not fast enough for that, it would also increase latency as you ideally want to send the data over the slowest busses as compressed as possible and only decompress it once you push it to your fastest bus.

Maybe a transition period where the chip is external to the CPU and sits on the PCIe bus between the CPU and a dedicated "DirectStorage" (or a more open standard equivalent) slot on the MB? Eventually, I think it probably gets integrated into the CPU, though.

I am sure it'd be built into the APU sorry if that wasn't clear, what I meant is that i except it to be its own device on the PCIe bus that you can write data to, and then the decompression block by itself can write to main memory.
 
PCIe 4.0 4x is not fast enough for that, it would also increase latency as you ideally want to send the data over the slowest busses as compressed as possible and only decompress it once you push it to your fastest bus.



I am sure it'd be built into the APU sorry if that wasn't clear, what I meant is that i except it to be its own device on the PCIe bus that you can write data to, and then the decompression block by itself can write to main memory.
right, yea it's gotta be built into the area that expects to receive teh data. Totally forgot about transport, which is funny considering about 90% of the pain in my life is working with uncompressed data types. 1 TB of data doesn't fit into a nvidia GPU lol. I just sort of forgot where things are being decompressed.
 
Another happy camper for the Next Gen Consoles with a long discourse on game scope vs graphics along with talk of variable frequencies and such.
Lengthy Reddit Post on Next-Gen

I will add that my questions on SSD lifetime issues and OS memory reserves regarding the PS5 were answered before I asked them ..
Most of the discussion for the SSD is about read traffic. Writes should be much better as well, but the systems are going to try to minimize that traffic since writes will be more difficult to handle and can produce more expensive disk operations due to wear leveling and remapping. An OS running every buffer there is through the SSD can also compromise the longevity of the drive.
So instead of say 15GB for games it would seem both consoles will have similar set asides.
Not as bad as when we only had 6 Gigs for the PS4 at first that was gut wrenching :-x..:LOL:
 
Probably just arbitrary numbered levels [example: 1, 3, 5, 7, 9, 11] with requests on one side being prioritized over ones on the other. So pick 1 to mean highest and 11 to be lowest.

*shrug*
 
That post just seems like a collection of various other posts and tweets from the past week or two. No original thought from what I've skimmed.
Not only that, there seams to be some conjecture in there, like the PS5 will have a constant temp with variable clocks while the Xbox will have variable temps but constant clocks. Logic would suggest that the variable clock rate will affect temperature in some way, and we know nothing about Sony's thermal solution. In fact, one thing we do know based on Cerny's presentation is that PS will not use temperature as a variable when changing clock speeds. Also, we know that Cerny's expressed pride in their thermal solution, and reports have indicated that it's more expensive than normal. One would think an elaborate cooling solution wouldn't be needed is temperatures were constant, unless maybe they were consistently high. But I don't think they would humble brag about their cooling solution if it could have been solved by a bigger piece of aluminum and an axial fan.
 
like the PS5 will have a constant temp with variable clocks while the Xbox will have variable temps but constant clocks. Logic would suggest that the variable clock rate will affect temperature in some way, and we know nothing about Sony's thermal solution. In fact, one thing we do know based on Cerny's presentation is that PS will not use temperature as a variable when changing clock speeds.
Yea,
the PS5 will have fixed power.
How much the clocks will fluctuate and how high the temperature rises as a result of that is unknown to us.
As I've stipulated repeatedly, my concern is the asymmetric difference in power consumption between CPU and GPU. With a fixed amount of power, under heavy load for the GPU to sustain it's frequency, it will draw significant resources from the CPU. This is my concern with those holding onto that 10.32 TF number. Frankly, I would have preferred the GPU to be a lower number and have the CPU fixed at 3.5Ghz. I suppose developers have a choice on locking one or the other. So perhaps my concern is a nothing burger when it actually comes to delivery. But in the case of discussion seems slightly dishonest to say PS5 is running 10.32 TF. It's also dishonest to say they have faster rasterization and ROP power when they have less bandwidth to feed it.

The assumption made by that writer is that there is lots of waste so not to worry they will maintain their boost clocks. But we know that's not true, Sony games are notoriously good at saturating their hardware or else their fans wouldn't be spinning up so loudly for all their titles. Thus, power draw. I'd also caveat that having a lot of wasted cycles on a high clock rate means you're not really doing much more. It's like having a 5GHz GPU with bandwidth of only 40Mb/s. Your silicon is running fast and idle twiddling it's thumbs waiting for work to do. Even Mark Cerny mentions that the high GPU clock rate combined with the slower RAM is an architectural imbalance during his presentation.
 
Status
Not open for further replies.
Back
Top