PC system impacts from tech like UE5? [Storage, RAM] *spawn*

PC SSDs this summer won't be faster than PS5's 5.5GB/s minimum / >8GB/s effective throughput. 5.5GB/s already close to saturating whatever you can extract from a 4x PCIe 4.0 bus, and then the PC has no decompression hardware to reach a higher effective throughput.

Isn't it 6x on PS5. Cerny called it "Priorities" though but I think he means lanes. Current gen top SSD's use 4xPCIe lanes as you mentioned, the PS5 uses a bit more because they can.

Ye, lol, today. The PS5 isn't out yet, so ain't the 7GB/s SSD's, but both will be at the end of this year.

Something you forgetting here though. If they are using compression, 9GB/s would be required. Then there's the driver and OS/filesystem overheads. Even current gen SSD's that are 5GB/s, real world results show actual speeds of 4GB/s or less depending on SSD/controller used/heat etc.

Tim did say the Intel NVMe Optane SSD's as the closest SSD's to the PS5 this gen. I would take it RAID NVMe SSD's would also have enough bandwidth to match the PS5, but that's going to extreme cost wise on the PC.
 
Isn't it 6x on PS5. Cerny called it "Priorities" though but I think he means lanes. Current gen top SSD's use 4xPCIe lanes as you mentioned, the PS5 uses a bit more because they can.

Something you forgetting here though. If they are using compression, 9GB/s would be required. Then there's the driver and OS/filesystem overheads. Even current gen SSD's that are 5GB/s, real world results show actual speeds of 4GB/s or less depending on SSD/controller used/heat etc.

Tim did say the Intel NVMe Optane SSD's as the closest SSD's to the PS5 this gen. I would take it RAID NVMe SSD's would also have enough bandwidth to match the PS5, but that's going to extreme cost wise on the PC.

Ok, I went to look at the video again. He doesn't talk about the lanes. He does however talk about the priority levels. He says current gen and seems to imply current 7GB/s gen SSD's only use 2 priority levels where as the PS5 uses 6 priority levels.
 
Ok, I went to look at the video again. He doesn't talk about the lanes. He does however talk about the priority levels. He says current gen and seems to imply current 7GB/s gen SSD's only use 2 priority levels where as the PS5 uses 6 priority levels.

Isn't it 6x on PS5. Cerny called it "Priorities" though but I think he means lanes. Current gen top SSD's use 4xPCIe lanes as you mentioned, the PS5 uses a bit more because they can.

Something you forgetting here though. If they are using compression, 9GB/s would be required. Then there's the driver and OS/filesystem overheads. Even current gen SSD's that are 5GB/s, real world results show actual speeds of 4GB/s or less depending on SSD/controller used/heat etc.

Yeah these are different things. Although Cerny says NVMe currently supports 2 priority levels, everything I've read says it supports 3 (High, Medium, Low), I've also seen reference to a 4th which is urgent. In any case the PS5 supports more which presumably makes things a little more efficient. Both PC and PS5 (and XSX) use PCIe 4.0 4x interfaces from the SSD to the rest of the system though which maxes out around 7.5GB/s. The difference is that the PS5 compresses more of the data it sends over that interface than a PC would so it can increase the effective bandwidth to 8-9GB/s. PC drives will be pushing 7GB/s uncompressed (vs PS5 5.5GB/s) by the end of this year, but it's not clear to me how much extra effective bandwidth PC's would get from compression, if any. And that extra bandwidth would have a trade off in CPU cycles which the PS5 doesn't have to contend with thanks to the hardware decompressor.

The PS5 has lot's of other enhancements to help out the data flow though, some of which may be solved/matched by DirectStorage and technologies like HBCC and some which certainly won't.

Tim did say the Intel NVMe Optane SSD's as the closest SSD's to the PS5 this gen. I would take it RAID NVMe SSD's would also have enough bandwidth to match the PS5, but that's going to extreme cost wise on the PC.

Optane is currently slower than standard NVMe Gen4 drives in pure throughput but has a massive advantage in latency. Gen4 Optane drives should equalize things in terms of throughput and I'd expect to see those within the coming months. An Optane SSD would have far lower latency than the PS5 SSD, but the rest of the data path from SSD to memory would have far lower latency in the PS5 than (current) PC's.
 
Yeah these are different things. Although Cerny says NVMe currently supports 2 priority levels, everything I've read says it supports 3 (High, Medium, Low), I've also seen reference to a 4th which is urgent. In any case the PS5 supports more which presumably makes things a little more efficient. Both PC and PS5 (and XSX) use PCIe 4.0 4x interfaces from the SSD to the rest of the system though which maxes out around 7.5GB/s. The difference is that the PS5 compresses more of the data it sends over that interface than a PC would so it can increase the effective bandwidth to 8-9GB/s. PC drives will be pushing 7GB/s uncompressed (vs PS5 5.5GB/s) by the end of this year, but it's not clear to me how much extra effective bandwidth PC's would get from compression, if any. And that extra bandwidth would have a trade off in CPU cycles which the PS5 doesn't have to contend with thanks to the hardware decompressor.

The PS5 has lot's of other enhancements to help out the data flow though, some of which may be solved/matched by DirectStorage and technologies like HBCC and some which certainly won't.

haha, I said as much in the Unreal thread.

https://forum.beyond3d.com/posts/2127122/

Optane is currently slower than standard NVMe Gen4 drives in pure throughput but has a massive advantage in latency. Gen4 Optane drives should equalize things in terms of throughput and I'd expect to see those within the coming months. An Optane SSD would have far lower latency than the PS5 SSD, but the rest of the data path from SSD to memory would have far lower latency in the PS5 than (current) PC's.

Yeah, not following Optane but Sweeney did mention that Optane was better, and now I take it he was actually talking about the latency as you mentioned. Good to know. :)

*editted* to fix all the quotes and my forum post.
 
The barrier to brute-forcing it on PC is you have to brute-force more than half of the PC architecture to accommodate all the bits of hardware in the I/O chain - including the SSD, the controller, the Southbridge, the CPU and the Northbridge - and that doesn't do anything to mitigate the inefficiencies of the software stack, nor overcoming two RAM pools.

What hardware developments do you think are on the horizon that will allow a PC to brute force it? Because I don't see anything and new standards are public a good 4-5+ years before that are implemented widely.

Can AMD start to release an APU with a similar I/O design as in the PS5?

Is there legal barrier for AMD to release an APU with 64CUs RDNA3, 12-core Zen3, 32GB HBM3, with a soup up PS5 I/O in 5nm process in 2022?

Can APU be the future of PC?
 
Can AMD start to release an APU with a similar I/O design as in the PS5?

Is there legal barrier for AMD to release an APU with 64CUs RDNA3, 12-core Zen3, 32GB HBM3, with a soup up PS5 I/O in 5nm process in 2022?

Can APU be the future of PC?

AMD did release some APUs for desktop called the A-Series. Right now the closest you'll find are CPU with integrated graphics.

https://www.amd.com/en/products/apu/7th-gen-a12-9800e-apu

The reason AMD will (probably) never release a APU similar to what PS5 has is that most people that are interested in high end graphics, are those who usually upgrade their components frequently. An APU limits your flexbility if you want to upgrade in the future. If I'm not mistaken you could pair some AMD APUs with another dedicated GPU but the support was very poor.
 

Attachments

  • upload_2020-5-20_13-26-39.png
    upload_2020-5-20_13-26-39.png
    91.7 KB · Views: 3
I think they exist, APU's atleast, in the desktop. But they rarely are for anything serious gaming.
of course they exist, but he is asking if those are the future of PC's. Of course they are not. PC have a lot of space to cool down components, no need to put GPU and CPU togheter
 
I listened to the conversation where epic said that it would run on a modern PC.
Was it just one press conference? or did they speak with multiple outlets independently?

A modern pc could be with or without sata. Their claim was that epic noted not even sata ssd could suffice.
 
Was it just one press conference? or did they speak with multiple outlets independently?
As I understand it, it was an open* remote meeting. Multiple outlets reported on it, each with their own take-aways it seems.

* Journalists only. Wasn't public.
 
As I understand it, it was an open* remote meeting. Multiple outlets reported on it, each with their own take-aways it seems.

* Journalists only. Wasn't public.
Yeah that was precisely how it was - I listened to a recording of the meeting that DF and other outlets had representatives at! IIRC it was the PC Gamer representative who asked questions about the nature of the PC (RTX 2070 was the example used) that could run it.
 
Agreed but the distinction of there being two buses with their own bus handling drivers continues. You can put a ton of functionality into a single ASIC but that does not remove the need for different drivers hooked into the right part of Windows for it all to work.

Is it two buses though? These are the 24 PCIe lanes on the Zen2 chip itself. They are simply split as 16 for the GPU, 4 for the NVMe SSD and 4 for the rest of the chipset. So surely these would all be controlled by a single driver, especially as the split of the 20 non-chipset lanes isn't fixed.

I don't follow AMD hardware in the PC world, but I wonder what the reason is for this? I can think of many but it would be speculation, from a for starters you really don't want one driver managing two very different types of hardware resources. There are different type of kernel driver optimised towards different types of device handling and GPU and SSD seem as far flung as it's possible to get in terms of I/O priority, IRP priority. Windows manages this itself, so if you do want to do this you now have to second guess the Windows driver management system. Also do you want a hickup with your graphics card impact your storage system? I'd argue not.

My understanding is that the ability to use the SSD (as well as network storage) as an extension of the GPU's virtual memory address space is more targeted at the server market than the gaming market, at least at the point when Vega was launched anyway. It may also be the case that AMD were aware of the console IO design at that point and wanted to ensure it's PC gaming parts could leverage the same capability when the time was right. Certainly at the moment the capability would be essentially useless as a gaming feature, it didn't really add anything when extending the memory space to system memory, never mind past that into NV memory because the software isn't designed to take advantage of it. But software designed around that paradigm coupled with potential improvements to the whole IO stack that Microsoft may be bringing with DirectStorage could potentially change all that.

It's also worth considering that the Radeon Pro SSG already sets a precedent for this in the PC space, i.e. the GPU communicating directly with an SSD (actually multiple SSD's in a RAID configuration). However I've no idea whether they were controlled by the graphics driver or their own driver given the SSD was attached to the GPU itself rather than being on the other side of a PCIe bus.
 
In zen2 do the pci-e busses connect straight to cpu or first to io-chip and then from io-chip to cpu?

They connect directly to the on chip IO die.

image-1-893x444.png
 
That io chip in middle of everything would allow for all kind of neat things if it could do compression/decompression. It seems to be capable of routing traffic very nicely without first having to bother dgpu/cpu.
 
Back
Top