Damn.No wonder sony doesn't want to talk about next gen yet.
The problem for me is not the Nintendo Switch shortage, its those Nintendo greedy guys selling twice the price because of the shortage with a pandemic in place.
But I'm blaming human, not Nintendo.That's not on Nintendo, they are constrained by supply chain and logistics issues. Blame human beings, they are shitty because they'll use any opportunity to make quick money. If you think Switch situation is bad, wait till you see some prices of PS5 on eBay in 9 months time.
Oh ok. Gotcha. I don't think it's about not waiting to get data from memory as often though, but rather decreasing wait time between cycles. You'll wait lightly more often but for much shorter periods of time, so to speak. But thanks for the clarification.That Wikipedia article is about error correction in caches. PS5's cache scrubbers are very different, they avoid the need to flush (i.e. erase) the entire GPU cache whenever the SSD is read.
Very different technologies. With this technology the GPU compute units don't need to wait to get data from memory as often.
The PS5 TE is not a standard CU, it's generally based on a CU architecture, for instance, it has 64 SU, but it is not an RND2 CU. When Sony or MS makes a GPU, they have no way of knowing in advance which CU will be defective so as a result, they have no idea which 4 CU they will disable, it's different for each APU that gets made. So Sony needs all 40 CU to be identical in order for the active 36 CU to be identical, that's why TE can't be one of the 40 CU. If anything, the TE will probably add more power to the PS5 because it's programable, which means some developers will "hack" it and use it for other purposes.
He said something along the lines that it was based on a CU to "benefit from GPU parallelism", but differed in some ways such as, for example, no caching, in order to resemble a CELL SPU more and make it more appropriate to audio than an actual GPU (this last comment should tell you that it's separate from the 36 active GPU CUs, btw).I am in agreement, however didn't Cerny say it will contain no GPU properties like a CU? That it's more an SPU like in the Cell, which then I suppose could be used for other things besides audio if they are allowed to. SPUs were able to do adaptive tesselation as an example.
Not happy with the mega PS4 sales. I can see Sony scratching their heads.....maybe we should wait an extra year.
Edit : I see you've added a bit about ssd sequential vs random access speed differences. I've never heard of this applying to ssd. It certainly doesn't make sense without ssd having mechanical restrictions, but I'm no expert.
I can't argue the ssd stuff because I simply don't know enough. But the internet downloading is not strictly comparable, there are overheads for data over the net. 10 small ones have 10x the packaging overhead vs 1x bigger one. And yes I also appreciate packet sizing matters.Manufacturers have taken to only posting max numbers and not providing a full performance breakdown. While there are no seeks, there is still the issue of the max #s being based on requests that can fully saturate the pipeline for the duration of the test. This is easy to do when testing against a large file that is large enough, but not too large to fill the dram cache. If you run a standard random read test, say 4kb random reads spread over a section of storage larger than the dram cache, speeds will plummet. This is similar to a internet bandwidth test, testing with one large file will saturate your connection offering the best result, while thousands of tiny files will not fully utilize your connection.
That Switch shortage is really hurting those numbers from Nintendo. I'm surprised at how badly the X1 is doing in comparison to it's direct competitor.
The problem for me is not the Nintendo Switch shortage, its those Nintendo greedy guys selling twice the price because of the shortage with a pandemic in place.
If DirectStorage is only an API, how do they magically get performance of multiple Zen cores? From your own link;
Microsoft's Goossen told Digitial Foundry that doing decompression on the 4K textures to match the speed of the SSD rate would have consumed three Zen 2 CPU cores, plus an additional two more just for the I/O overhead. With DirectStorage, Microsoft reduced that down to just a tenth of one core. All that CPU power can now be repurposed for other things.
I can't argue the ssd stuff because I simply don't know enough. But the internet downloading is not strictly comparable, there are overheads for data over the net. 10 small ones have 10x the packaging overhead vs 1x bigger one. And yes I also appreciate packet sizing matters.
Apart from that, Sony doing 12 priority channels on the ssd and all this wiz bang stuff could be exactly the motivation to alleviate traditional shortcomings of the tech.
Anyway, convo out of my depth lol. All I said is PS5 ssd has ddr3 speed which checks out if reviewing speed metrics
That 22GB is theoretical peak of the hardware. Probably only obtainable if the texture is 1GB of zeros... The real number is 9.6higher throughput from decompression unit (22GB/s vs 6GB/s),
800TBPS5 ssd is essentially 800 terrabytes of DDR3
Q. Why aren't people talking about the cache scrubbers secret sauce ?My authority has cache scrubbers, when refuted I don't have to wait for the next dubious source, just scrubbed and loaded up with some more juicy FUD.
Putting all the transfer speeds aside, there is one other important aspect here. You can read from and write to RAM bit for bit. For SSDs, you can't. You can only read pages, and only write to blocks.
RAM is cleared and reloaded frequently depending on the assets that will be needed in the near future. An SSD would not be able to do this, because they are by default hampered in terms of writing.
If DirectStorage is only an API, how do they magically get performance of multiple Zen cores? From your own link;
Microsoft's Goossen told Digitial Foundry that doing decompression on the 4K textures to match the speed of the SSD rate would have consumed three Zen 2 CPU cores, plus an additional two more just for the I/O overhead. With DirectStorage, Microsoft reduced that down to just a tenth of one core. All that CPU power can now be repurposed for other things.
SSDs aren't supposed to be ram, but if you remove the bottlenecks you won't need as much RAM because you can switch data quickly enough in the RAM itself. It's why we don't need such a big increase in RAM, why data streaming is only going to become even more important, and it's why these SSDs are a biG deal. It's not like we have on pc, because there's many bottlenecks there (which will be removed as well in the near future).
It's hard to grasp for some folks, even people like Alex from DF, but thankfully devs will know what to do with it as it was their main request.
According to Cerney cache scrubbers allow the deletion of discrete parts of the cache instead of all of it. The cache normally would all get blown away unnecessarily as soon as ram gets altered by streaming in new data from ssd. Cache scrubbers will only delete necessary parts of the cache where underlying ram gets changed. Obviously cache is the fastest part of the memory system so you want to preserve it to keep peak performance.That 22GB is theoretical peak of the hardware. Probably only obtainable if the texture is 1GB of zeros... The real number is 9.6
800TB![]()
Q. Why aren't people talking about the cache scrubbers secret sauce ?
A. Nobody know what they fuck they are or do.
I thought cache gets changed/overwritten when GPU access memory not in cache - .. so it'll get overwritten anyway ?According to Cerney cache scrubbers allow the deletion of discrete parts of the cache instead of all of it. The cache normally would all get blown away unnecessarily as soon as ram gets altered by streaming in new data from ssd. Cache scrubbers will only delete necessary parts of the cache where underlying ram gets changed. Obviously cache is the fastest part of the memory system so you want to preserve it to keep peak performance.
So, for you BOTH SSDs are useless due to the lowest common denominator HDDs? Don't forget SX will be much more affected by this problem cause even its first party games will have to run on old hardware for some time, while PS5 first party games can thrive from the start.I want to know something
How is the Playstation 5's SSD a "gamechanger" when all of the multiplatform games will be targeting HDD?
We'll only see a few improvements, the main one being load times, but then again, comparing SSD load times is like comparing bullet speeds.
The only games to take full advantage of PS5 ludicrously fast SSD are the First Party exclusives, which are going to be totally amazing.
At the end of the day, games are going to look better on the SXS and load faster on the PS5. Both consoles are going to be amazing.
If you say so... I still have questions. Like... The one I already asked. How can only an API change reduce CPU performance from 3 cores to less than 1/10th of a core? Where they simply lazy all these years? Why wasn't this done earlier on the Xbox One which is known to be extremely underpowered in the CPU department?"This newest member of the DirectX family is being introduced with Xbox Series X and we plan to bring it to Windows as well."
This is coming straight from Xbox's own glossary of terms under the term DirectStorage, what else do you need other than words coming from Microsoft's mouth? The newest member of the DirectX family. DirectX isn't a component, it's an API and DirectStorage will be one of its components once the Series X launches. It will find its way on pc sometime later down the line. I don't think i need to say more.
They can, but I doubt they will use it from the start. Most likely it will take at least two years into the cycle before we start seeing games that truly utilize the SSD speeds.Don't forget SX will be much more affected by this problem cause even its first party games will have to run on old hardware for some time, while PS5 first party games can thrive from the start.
You don't choose where to disable them unless the 40 ones are usable.The thing is any one of the CUs from SA's can be disabled, they can't predict were the defects will hit that's why im not sold on the idea, plus TE is a different design. Besides TE is tiny it wouldn't take much space at all.
But idk could happen.
They can, but I doubt they will use it from the start. Most likely it will take at least two years into the cycle before we start seeing games that truly utilize the SSD speeds.
You'll see them crawling out of the woodwork any time some new or positive PS news gets released.Because there are way more people attacking PlayStation than Xbox.
Afaik api's can help reduce overhead to a certain extent speeding up certain workloads depending on how they're implemented and utilised. One game engine might use it in a different way than another. No two engines are ever alike, especially proprietary ones coming from first party studios ,which are bound to use the api itself and its features in a much more streamlined manner since they're supposed to be the ones more familiar with the architecture itself compared to third party studios which could approach it in a different manner. You'll notice the effects of DirectStorage more on first party games than third party games.If you say so... I still have questions. Like... The one I already asked. How can only an API change reduce CPU performance from 3 cores to less than 1/10th of a core? Where they simply lazy all these years? Why wasn't this done earlier on the Xbox One which is known to be extremely underpowered in the CPU department?
There is something else going on here.
They can, but I doubt they will use it from the start.
Let me rephrase it a little bit. They will use the SSD speeds, but it won't be anywhere near optimal. In the beginning I expect developers to simply throw everything at the SSDs with little optimization. They have much more leeway compared to before after all. One of the things with too much freedom is that efficiency and optimization goes out the window.What makes you think that?
Let me rephrase it a little bit. They will use the SSD speeds, but it won't be anywhere near optimal. In the beginning I expect developers to simply throw everything at the SSDs with little optimization. They have much more leeway compared to before after all. One of the things with too much freedom is that efficiency and optimization goes out the window.
Imagine for a second if the PS3 had come out with 16GB of RAM. The last thing you're going to optimize for is the RAM, because the bottleneck is everywhere else.
A similar thing will happen in the beginning of the PS5/XSX gen. Developers are currently used to running games from extremely slow hard drives. The SSDs provide such fast speeds that their optimization will not go into the SSDs for now, but into the rest of the system. And since games take a few years to develop, the games that really utilize the SSDs efficiently will release in about two years, which is actually quite short. If compared to the PS3 Cell, it took way longer than 2 years to see games that actually took real advantage of it.
I'll assume that by gaming pc you mean something that can play Farmville!Most pc's used for gaming (Not necessarily gaming pcs) match that
Unless you're counting my grandma's pc
This must be why Rockstar releases on the PC 2 years after they release GTA on the consoles. If what you say is correct shouldn't rockstar release first on the PC and then port to consoles ?Uh?
You act as if the target customers are the Master Race, a little hint, they arent, devs will optimize for the largest common denominator
That's just complete conjecture and doesn't really make sense at all
I mean, why exactly would Spiderman 2 or HZD 2, which we can assume both will be out by 2021/2022, take this inefficient route? And God of War 2 or Days Gone won't? They're not going to be rushing these things.
Damn.No wonder sony doesn't want to talk about next gen yet.
I think first parties will work to maximize both systems as quickly as possible. Third parties are in a different position. Most of those profitable $60 game sales are in console, but the PC market is still worth a good chunk of change (the mid tier being the most profitable). They could go all-in on the SSD and just forget the majority of the PC market, or they could build a game that can run on a lower standard and have all the markets covered. I know where I'm putting my money. Even if new technology hit the PC space tomorrow (better APUs, GPUs, SSDs, Memory, etc.), it will take years for that tech to become the average baseline for PC.
Yes, completely different concepts. The 6 priority levels tailored for gaming Cerny mentioned are in the memory controller and have no correlation with nand chipsReally? Ok...
Do we know for sure random reads have a significant impact in gaming, and how frequent are they? keep in mind they also implemented customizations to minimize its impactThen why did he compared it with HDD's sequential speeds? Doesn't make any sense...
Thanks for breaking it down, I know TE used a CU as its frame but they redesigned in such a way the micro architecture closely resembles a SPU. As such TE wouldn't be able to dub as a traditional CU to work in tandem with other CUs since its different on the micro-architecture level, it wouldn't contribute to the advertised 10.27TF eitherThe PS5 TE is not a standard CU, it's generally based on a CU architecture, for instance, it has 64 SU, but it is not an RND2 CU. When Sony or MS makes a GPU, they have no way of knowing in advance which CU will be defective so as a result, they have no idea which 4 CU they will disable, it's different for each APU that gets made. So Sony needs all 40 CU to be identical in order for the active 36 CU to be identical, that's why TE can't be one of the 40 CU. If anything, the TE will probably add more power to the PS5 because it's programable, which means some developers will "hack" it and use it for other purposes.
True but remember he said its main purpose its 3D audio and it can help with other audio related calculations since it excels at themI am in agreement, however didn't Cerny say it will contain no GPU properties like a CU? That it's more an SPU like in the Cell, which then I suppose could be used for other things besides audio if they are allowed to. SPUs were able to do adaptive tesselation as an example.
Agreed, was merely comparing the peak capabilities of the decompression units.That 22GB is theoretical peak of the hardware. Probably only obtainable if the texture is 1GB of zeros... The real number is 9.6
I think first parties will work to maximize both systems as quickly as possible. Third parties are in a different position. Most of those profitable $60 game sales are in console, but the PC market is still worth a good chunk of change (the mid tier being the most profitable). They could go all-in on the SSD and just forget the majority of the PC market, or they could build a game that can run on a lower standard and have all the markets covered. I know where I'm putting my money. Even if new technology hit the PC space tomorrow (better APUs, GPUs, SSDs, Memory, etc.), it will take years for that tech to become the average baseline for PC.
Yes, completely different concepts. The 6 priority levels tailored for gaming Cerny mentioned are in the memory controller and have no correlation with nand chips
Do we know for sure random reads have a significant impact in gaming, and how frequent are they? keep in mind they also implemented customizations to minimize its impact
4K random reads benchmarks were posted earlier to show how SSD throughput drops but what about current consoles laptop hdds? In 4k random read benchmarks the average 5400rpm drive scores 80 iops or 0.3 MB/s. I don't think these tests are indicative of gaming workloads
Thanks for breaking it down, I know TE used a CU as its frame but they redesigned in such a way the micro architecture closely resembles a SPU. As such TE wouldn't be able to dub as a traditional CU to work in tandem with other CUs since its different on the micro-architecture level, it wouldn't contribute to the advertised 10.27TF either
True but remember he said its main purpose its 3D audio and it can help with other audio related calculations since it excels at them
I don't think it would be worth it to sacrifice audio (defeating the purpose of TE) to offload visual calculations with only a fraction of GPU power
Agreed, was merely comparing the peak capabilities of the decompression units.
Damn.No wonder sony doesn't want to talk about next gen yet.
Damn.No wonder sony doesn't want to talk about next gen yet.
Lol...we are so hungry for news other than pandemics and politics, most of us would watch a four hour special on the power cable! If it was a call in show, I'd even call in to join the discussion.
Switch is having shortages. And media create confirmed that weekly number for Japan and this is correct . These are weekly numbersIs this a weekly or monthly total. ~300,000 per month ~= 3.6m per year ignoring seasons, but 14.4 m if weekly - seems high. Is the data confirmed legit.
Xbox still getting owned though.
[edit] - Doubt - PS4 outselling Switch in Japan - why?? this aint right ?
It's a weekly chart, so PS4 could be selling more than 1kk this month thanks to the quarantine. Switch is in shortage everywhere, Nintendo's low unit stock policy is surely hurting them now.Is this a weekly or monthly total. ~300,000 per month ~= 3.6m per year ignoring seasons, but 14.4 m if weekly - seems high. Is the data confirmed legit.
Xbox still getting owned though.
[edit] - Doubt - PS4 outselling Switch in Japan - why?? this aint right ?
FF7R.[edit] - Doubt - PS4 outselling Switch in Japan - why?? this aint right ?
I was thinking Animal Crossing.FF7R.
You doubt PS4 outselling Nintendo Switch. FF7Remake was launch plus Nintendo Switch shortage.I was thinking Animal Crossing.
Yes, TE is programable so developers might use it for other things just like they did on PS3 with the SPUs.I am in agreement, however didn't Cerny say it will contain no GPU properties like a CU? That it's more an SPU like in the Cell, which then I suppose could be used for other things besides audio if they are allowed to. SPUs were able to do adaptive tesselation as an example.
Not all parts of the GPU suffer from fail rates as high as CUs. Cache for instance has a very hight tolerance for manufacturing defects. A very high percentage of PS5 consoles will have 40 fully functional CUs and most of the PS5 consoles with defective CUs will only have one defective CU so even if the TE has the exact same fail rate as a CU, it will be much lower than 2.5% out of usable PS5 APUs. That's why GPU yields go up the less CUs a GPU has, a single CU has a very low fail rate but when you have 40, there is a defective CU in a high percentage of the GPUs.If it is a separate and totally unique component relative to the standard CUs, then how can Sony assure tolerable yield rates for the APU, since the odds of producing an APU with one fully functional Tempest Engine out of only one Tempest Engine would be low? Do you suppose that there are redundant Tempest Engines in the APU, so that the odd of producing an APU with at least one fully functional Tempest Engine would be high?
If all 4 defective CUs are in the same SA the GPU isn't usable. In RDNA 1 and 2 (and GCN too) the SEs have to remain balanced, so if you deactivate a CU in one, you have to deactivate one in the other. If all 4 defective CUs are in the same SA, you need at least 8 deactivated CUs in order to make the GPU pass certification and statistically you actually need to disable more more than 12 CUs.You don't choose where to disable them unless the 40 ones are usable.
You disable the defective ones and that means there 4 defectives can be even in the same SA.
It is random.
That is how you increase the yields with redundant units.
Of course it is... there is RX 5500 with all disabled CUs in the same SA... one SA with 14CUs and the other with 1Yes, TE is programable so developers might use it for other things just like they did on PS3 with the SPUs.
Not all parts of the GPU suffer from fail rates as high as CUs. Cache for instance has a very hight tolerance for manufacturing defects. A very high percentage of PS5 consoles will have 40 fully functional CUs and most of the PS5 consoles with defective CUs will only have one defective CU so even if the TE has the exact same fail rate as a CU, it will be much lower than 2.5% out of usable PS5 APUs. That's why GPU yields go up the less CUs a GPU has, a single CU has a very low fail rate but when you have 40, there is a defective CU in a high percentage of the GPUs.
If all 4 defective CUs are in the same SA the GPU isn't usable. In RDNA 1 and 2 (and GCN too) the SEs have the remain balanced, so if you deactivate a CU in one, you have to deactivate one in the other. If all 4 defective CUs are in the same SA, you need at least 8 deactivated CUs in order to make the GPU pass certification and statistically you actually need to disable more more than 12 CUs.