Unreal Engine 5, [UE5 Developer Availability 2022-04-05]

That makes no sense to me. SSD's access data directly, so position within the NAND won't affect seek time. With HDDs, or spinning disks, you have to move the head to the data with a significant influence on latency, so physically having the data closer by reduces seek times.

There should be zero reason to duplicate data on SSDs, unless I don't understand some underlying intricacies.

Because the sequential speed of SSD is faster on NTFS. They arranged the data to be sequential no duplication of data because it is a linear sequence but if you begin to have an open-world with data called often with very fast movement you will need to duplicate the data too for keeping the data sequential. Don't forget MS probably changes the I/O with Direct Storage and maybe the filesystem.

Like I said before PC filesystem is not ideal for SSD. It works like a PC benchmark.

2019-09-23-image-5.png
 
That makes no sense to me. SSD's access data directly, so position within the NAND won't affect seek time. With HDDs, or spinning disks, you have to move the head to the data with a significant influence on latency, so physically having the data closer by reduces seek times.

There should be zero reason to duplicate data on SSDs, unless I don't understand some underlying intricacies.
AFAIK nand memory is not exactly random access memory.
 
Depends of the filesystem, on PS5 they have file archive a fils format special SSD and on Xbox Series X Direct Storage it will arrive later on PC.
Fantastic explanation by Google:

https://www.elinfor.com/knowledge/w...ce-between-nand-flash-and-nor-flash-2-p-11162

What is the difference between NAND Flash and NOR Flash?

To overcome or reduce the limitation of the slower read speed, the data is usually read in the form of a page in NAND flash memory, which is the smaller portion of the erasing block. The contents of a page are read sequentially using the address and command cycles only at the beginning of each read cycle. Sequential access durations for NAND flash are generally lower than random access durations for NOR flash devices. When using the random access architecture of NOR flash, address lines need to be switched at each read cycle to accumulate random access for sequential reads. As the size of the data chunk to read increases, the cumulative latency in NOR flash becomes greater than that of NAND flash. As a result, sequential reads of NAND flash can be faster. However, the performance difference between the two types is apparent only when transmitting large data blocks that are typically larger than 1 KB because the initial read access duration of NAND flash is much longer.
In both Flash technologies, data can only be written to a block when it is empty. The slow erase operation of NOR flash makes its write operation slower. In NAND flash, similar to reading, data is usually written or programmed as a page (usually 2KB). For instance, using NAND flash S34ML04G2 alone requires 300μS to write to a page.
In order to speed up write operations, modern NOR Flashes also use buffer programming similar to page writing. For instance, NOR flash memory S70GL02GT mentioned above supports buffer programming, which enables it to write word-like timeout multibyte programming. For example, the buffer programming of 512-byte data can achieve the throughput capacity of 1.14MBps.


Energy Consumption
NOR flash memory usually requires more current than NAND flash memory during initial power-up while the standby current of NOR flash is much lower than that of NAND flash. The instantaneous active power of the two flash types is equal. Therefore, the effective power is determined by the duration of memory activity. NOR Flash has advantages in random reads, while NAND Flash consumes relatively little power in erase, write, and sequential reads
 
Last edited:
This demo went insanely viral. Just listening to latest joe rogan podcast and he briefly talks about unreal and this demo. Just nuts, pretty much everyone and their mother must have heard about this demo at this pace.
 
Well, Huang did say in a slide back in late 2019 that a "next gen console" is less than RTX 2080 on notebook.

5urok2y09m641.jpg
Well, he must be smoking something allucinogenic (Jen-Huan dixit) or AMD and RDNA 2 are some serious turds. But i dont believe so, as Volta and RDNA 1 already had the same perf/watt and RDNA2 suppossely will be here to neutralize the increment Amper will bring with IPC (rumored to be a 10% over Volta) plus node change.
 
To give some context around the laptop GPU, the Nvidia RTX 2080 Max-Q : https://laptopmedia.com/comparisons...-that-far-away-when-it-comes-to-heavy-stress/

Well, he must be smoking something allucinogenic (Jen-Huan dixit) or AMD and RDNA 2 are some serious turds. But i dont believe so, as Volta and RDNA 1 already had the same perf/watt and RDNA2 suppossely will be here to neutralize the increment Amper will bring with IPC (rumored to be a 10% over Volta) plus node change.

Is this in relation to the 40fps claim ?

Note that the engineers said, "Actually for the scene you saw previously, we were able to run it @ 40fps on our laptop inside the editor tool where the assets were not completely cooked".

I don't know what he meant by "not completely cooked". As ultragpu pointed out, he also didn't mention the resolution.
 
Last edited:
Those laptops cost upwards of £€$1600. I don't think it's a fair comparison no matter your perspective (and would they even have the same IO or decompression tech? Certainly not).

I don't even think many PC gamers would suggest buying one of those laptops, they're hardly bang-for-buck.
 
Is this in relation to the 40fps claim ?

Note that the engineers said, "Actually for the scene you saw previously, we were able to run it @ 40fps on our laptop inside the editor tool where the assets were not completely cooked".

I don't know what he meant by "not completely cooked".

Usually the editor performance is not as good as in-game, so that looks even better for pc gpu performance.
 
No one here is talking about value. We're talking about the technology and how it scales.
 
Usually the editor performance is not as good as in-game, so that looks even better for pc gpu performance.

He was referring to the scene near the beginning where light shone down at the rocks. He didn't specify the resolution. Not sure if all effects are on.

He seems determined to hit his framerate targets (in the context of nextgen consoles) though.

EDIT:
You should probably wait for the final thing as they still have work to do, and may be using non-GM devkits. Remember, this is an early look. UE5 itself is due early next year.

They were going to show the demo at GDC 2 months ago but Coronavirus happened.
 
Last edited:
https://www.eurogamer.net/articles/...eal-engine-5-playstation-5-tech-demo-analysis

"The vast majority of triangles are software rasterised using hyper-optimised compute shaders specifically designed for the advantages we can exploit," explains Brian Karis. "As a result, we've been able to leave hardware rasterisers in the dust at this specific task. Software rasterisation is a core component of Nanite that allows it to achieve what it does. We can't beat hardware rasterisers in all cases though so we'll use hardware when we've determined it's the faster path. On PlayStation 5 we use primitive shaders for that path which is considerably faster than using the old pipeline we had before with vertex shaders."

Regarding software rasterising using CUs, I caught up reading the Eurogsmer article, and for PS5 at least, Epic are using its primitive/ geometry shaders for this area instead of CUs. So that presumably leaves the CUs for lighting and shading.

Also, RDNA2 has 4 Primitive/ Geometry units?

If they take a billion triangles in a frame and cull them down to 20 million lossless triangles, then render at 1440p/ 30. Those 4 primitive/ geometry shaders are spitting out roughly:

1440p ~ 3.6 million pixels per frame

PS5 GPU ~ 2230 MHz/ 30 FPS = 74.3 million cycles per frame

1 pixel rasterised ~ 74.3/ 3.6 ~ every 20 clock cycles using 4 primitive/ geometry units

So, their REYES algorithm can process a pixel in 20 cycles - is that quite cheap? Then use the CUs to shade and light for dynamic GI.

I believe you are misunderstanding how this works. The output of primitive shaders feed the rasterizer hardware. The UE5 engine software rasterizes and bypasses the raster hardware. The majority of the triangles are software rasterized. The minority of triangles, where the hardware rasterizes would be faster (probably related to cases where polygons are bigger than one pixel), the engine falls back to primitive shaders instead of the old vertex shader pipeline. For the majority of the scene the primitive/raster units are not used.
 
Really curious to see how DLAA 2.0 would work with UE5. Likely it'll be DLAA 3.0 by that time. DLAA is typically competing with highly-aliased image output, or image output that is highly blurred(TAA) to remove the aliasing. UE5 should not have any edge aliasing, so 1440p UE5 should look much cleaner than a typical 1440p game. Also makes me wonder if DLAA would even work with it, or could be massively simplified. You've pretty removed the AA part, and now you're just upscaling an image. There will be new types of visual noise from shaders, or maybe texture sampling. You can move slightly and change the angle of your view which would make the polygon per pixel change, which means you're now sampling a new texel which could be a different colour. I'm imagine like ray-tracing they need to de-noise because of shimmering in motion.
 
Back
Top