I swear you are the poster child for confirmation bias
-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
Inside the Scorpio Engine: the processor architecture deep dive
- Thread starter Space_nut
- Start date
I swear you are the poster child for confirmation bias
Leyasu
Banned
Except the new mrxmedia has changed his colors..
hydrophilic attack
Member
so basically, the typical gaffer would need a pc to be impressed
nekkid
It doesn't matter who we are, what matters is our plan.
You forgot to add that to achieve 8 inches, the Pro needs to have the girth of a chopstick.
VlaudTheImpaler
tl;dr
NSFL :Warning: NSFL
EDIT: I may have had a dream like this once... As a teenager of course...
What info?
I'm not going to pretend like I understand how full precision really works. It seems like FP32 is the route to go for better quality textures etc, going off those images posted by Ethomaz.
I'm going to stick to what I can physically see ie comparisons in 4k.
Games I really want to see patched are Witcher 3 and Fallout 4 with the ultra textures.
I'm going to stick to what I can physically see ie comparisons in 4k.
Games I really want to see patched are Witcher 3 and Fallout 4 with the ultra textures.
There will 100% be a Fallout 4 patch, one can only hope Witcher 3 gets one for both the Scorpio & the Pro.
Well it was more than five pages buy i thank you for your services.
Dee Dah Dave
Member
Yea it's time to lock the thread now.
Actual face off threads are going to be a minefield.
Can you provide any examples of items that would need fp32 and could benefit from fp16 instead?
It's just a parody of the claims MS made about the XB1... http://m.neogaf.com/showpost.php?p=80168873&postcount=708
I think these are, without proper context, somewhere between irrelevant and deceptive. These are images being produced by calculations that noticeably differ in their results depending on whether you use FP16 or FP32, but, without knowing what the calculations being used are, it is impossible to know whether it's a case of FP16 being insufficient, or FP16 being incorrectly used. More over, as you indicate, these are cases where one is purely using FP32 and on is purely using FP16, which wouldn't be the case with the PS4 Pro.
Your summary is correct though: the ultimate point with this whole debate is that sometimes you can get away with lower precision without noticeably lowering quality. There are two really relevant questions that stem from this: how often can you get away with it, and, how often will developers go the extra mile to do so? I can only speak from my perspective as someone who works on mobile games, where bothering to exploit lower precision is absolutely necessary, but I'd suggest you can use FP16 a surprising amount without anyone being able to tell the difference.
We don't know why Microsoft have decided not to include FP16. There may well be a very good reason that they didn't want this particular customisation. These kind of improvements, after all, are reflected in the physical hardware. It's not necessarily cost free, or down-side free. We'll probably never know. On the face of it, it does seem like an oversight, and it is certainly one area where the PS4 Pro's GPU one up's Scorpio's.
I think you're wrong on two counts really. First, no-one is arguing the PS4 Pro is more powerful in here. Second, I doubt these discussions will continue much beyond November.
As I said in my response to Ethomaz, I'd be careful in reading too much into those images. You certainly won't experience any differences of that nature between a game on Scorpio and the same game on PS4 Pro. Developers will only use FP16 when they don't believe there is an appreciable difference in quality, and where there is a non-negligible difference in performance. It's important to remember that both consoles support FP32 and FP16 - but only the Pro supports performing two FP16 operations at the same speed as one FP32 operation.
I'm guessing you mean wouldn't rather than would here. There are various examples that come to my mind.
1) Colour. Even with HDR, we don't use more than 16 bits to store colour. Calculations involving colour can thus often be done at FP16.
2) Texture coordinates. FP16 is typically sufficient for the position you want to perform a texture lookup. Not certain how well that holds up on a 4K texture - it's possible there is noticeable precision loss near 1. Certainly, I've never noticed any issues with precision for textures smaller than 4K using FP16. Worth noting that a game being rendered at 4K dosn't mean all of your textures are going to be 4K.
3) Directions. A lot of shader mathematics revolve around normalised vectors, which is to say vectors which represent directions with a magnitude of one. Floating point numbers are, by design, increasingly precise at smaller magnitudes. Hence, FP16 is still quite precise for a normalised vector because its magnitude is small. A basic example of where you could get away with FP16 would be in the common diffuse lighting calculation N dot L, where you determine how bright an object's surface is by dot-producting the direction of the incoming light (L) and the surface normal (N). N dot L is equivalent to N.x * L.x + N.y * L.y + N.z * L.z. My general experience is that, in this calculation, FP16 produces results that are indistinguishable from FP32. Other lighting calculations, like those involved in calculating specular reflection are a lot more sensitive to precision issues due to the relevant equations involve powers. Practically, in the PBR era, most lighting calculations require precision above that of FP16.
The general point with FP16 is that, ultimately, our intention with graphics is to output colours. Those colours are not stored at FP32, and, in fact, don't even use the full range of FP16, even in the HDR era. It is thus obvious that there will be areas prior to the output where you can use FP16 to store values, and areas where you can use FP16 to calculate values, without there being a difference, and an even wider range of areas where you can use FP16 to store or calculate values where there will be no noticeable or appreciable difference (i.e. you wouldn't prefer one output over the other). How well FP16 holds up in any given situation depends hugely on the magnitudes of the inputs, and the operation in question*. Obviously, each operation adds potential for imprecision, so, for example, repeated multiplications quite quickly see you lose precision.
As a basic example, you can expect a lot more precision when adding than when multiplying. I mentioned earlier that floating point number precision is higher with smaller magnitudes. The reverse of this is that floating point number precision is lower at higher magnitudes. The highest value (other than infinity) that you can store in an FP16 is 65,504, which is obviously big enough for a great many things, but, the next highest number is 65,472, and from there to 65,440. Hence, if I use FP16 to multiply two numbers together, say, 255.7 * 256.0 (= 65,459.2), I'm going to lose a lot of precision with FP16 when it gets rounded. With adding, the precision loss is far smaller. 255.7 + 256.0 = 511.7, as FP16 is capable of storing 511.75.
BriareosGAF
Member
That's a good question, actually, I'm not sure I remember the original rationale (e.g. why not patch on the PPU via write combine?), but it's not entirely uncommon to do such things, really, just usually you have a custom front-end processor on the GPU that does DMA, etc., while on the RSX the real cost was the mode switch. Needless to say one of the earliest tasks leveraging the SPU was to move shader program patching there. (Another fun fact: stock Unreal 3 had only one SPU module, and it either did shader program patching or EDGE culling depending on the arguments passed. On Call of Duty we had in excess of 64 SPU modules by the time we finished MW3.)
I think Michal Drobot might discuss our up-sample approach at SIGGRAPH (other primary author here is Jorge Jimenez), but we don't use entity ID data as part of our custom up-sample in Infinite Warfare. DICE's is much closer to the "stock" up-sample arrangement.
Like anything sufficiently interesting, it's complicated, and workloads vary depending on requirements, context, etc. That's really my point most of the time posting in threads like this--it is essentially impossible for the consumer to ever understand the environment in which decisions like this are made or the real underlying reasons and rationale. And really, there's no reason to--that's why I like when DF just focuses on the end result *as observable by the consumer* e.g. doing analysis of frame pacing, resolution, etc. It avoids them getting into the dangerous territory of speculating about how hardware is designed, leveraged, and how software is written, which is far more complicated than possible to express accurately in web articles.
I know, I was here when it happened, lol.
Crazy times back then. It's about to repeat soon and when both companies launch their next gen consoles.
Dee Dah Dave
Member
If history proves anything, all these minor 'bespoke' features never amount to much. In the end the main hardware tells the real story. CPU, GPU, RAM and bandwidth.
The Scorpio is better and that's all there is to it. XB1 was good with PS4 being even better. Now Pro is good and Scorpio even better. No big deal.
The Scorpio is better and that's all there is to it. XB1 was good with PS4 being even better. Now Pro is good and Scorpio even better. No big deal.
But all those people that started gaming on the ps3 are not used to Sony not having the most ~powerful~~~ toy console out there. Won't anyone think of those people?
On topic: I just want to see some games. Games are way more important than all those numbers no one even understands. And then there's the price problem.
On topic: I just want to see some games. Games are way more important than all those numbers no one even understands. And then there's the price problem.
Well said
An important thing to remember too is that FP16 won't make up for the ram and bandwidth advantage Scorpio has. This will allow 4k assets which will be a big differenterator between Scorpio and pro
If its ok to ask (not sure of NDA's and such so ignore me if it'll get you in bother) but do you have any personal thoughts on Scorpio and developing for it?
Always curious to hear from third party developers
Cheers!
If it doesn't matter than Microsoft should stop talking about all these customizations. It seems constantly we're reminded of them and how they are the thing that will help the Scorpio kick the 1080 in its nuts.
I'm exhausted looking for actual technical data rather than buzzwords so yeah it does feel bad.
Exactly.
Wow, how did DF miss this detail?Beyond3d
BriareosGAF
Member
There's nothing specific I can discuss (and my comments should not be construed as the official position of my employer, etc.), but I will say it's one of the more fun things in the job to be on the leading edge of disclosure with new hardware. And where I am in my career now one of the few challenges left is to participate in the design process for hardware, so it's been fun seeing tastes of that through time since we work so closely with IHV partners. We also have some incredibly talented engineers on staff here who take it as a personal challenge to discover features and nuances in the hardware before first party does, which is always fun.
That's the promise at least.
Fredrik
Member
I have no worries this time, I'm sure we'll see a few brilliant MS exclusives at E3 as usual and then there is the new third party games situation too where everything most likely run really great on Scorpio. E3 will be fun for sure this year.
SolidChamp
Banned
Has it been determined if this will provide an excellent alternative to the rigs needed to support the VIVE/Oculus stuff?
If I were rich, I'd go all out on a VR-dedicated gaming rig. But I'm not rich, so I often entertained the idea of getting PSVR, but could be swayed towards this if it supports the kind of VR tech that PC is seeing.
If I were rich, I'd go all out on a VR-dedicated gaming rig. But I'm not rich, so I often entertained the idea of getting PSVR, but could be swayed towards this if it supports the kind of VR tech that PC is seeing.
They do matter it's said in the article. All the customizations done were changes done from the extensive simulations from dev engines and finding all the botttlenecks they were experiencing. All the custom work done are changes to even the minute detail in the gpu such as queue caches and such. They even went into the CPU core and changed things too. They also took some Vega features that they knew would give benefit.
Risk Breaker
Member
Er... When performing FP16 operations the PS4P would already do them at double rate, it's a hardware feature, you don't program in "double-rate FP16" lol.
Cheers mate! understand that you can't really go into specifics, appreciate the insight though : )
This Guy gets it.
Crayon
Member
Some random hot take a few pages back.
12 page thread boils down to it's most obvious simplest post. Congrats.
Yeah that's right I made up all this fp16 stuff before Scorpio & PS4 Pro specs was even leaked just because I knew that I could use it for my agenda.
dr_rus
Member
Yes, you are.
Usage of FP16 inside the GPU for shader processing doesn't mean jack shit for memory and bandwidth. I generally fail to think of any game which is using 128 bits precision for render targets and frame buffers. Most games are outputting to R10G10B10x2 surface formats these days (which is 32 bits), some games (and only on PC mostly) are using FP16 for more precise tone mapping. D32F is a single component depth buffer format which is 32 bits. What you're saying simply doesn't exist.
...WOW.
This is possibly the fastest turnaround I have ever seen on GAF.
"None of these extra features really matter, UNLESS Microsoft says they matter."
I actually laughed when I saw this play out.
blu
Wants the largest console games publisher to avoid Nintendo's platforms.
It's been a trainwreck of a tech thread.
Crayon
Member
This marketing push to style xbox as the go-to brand for Real High Performance Gaming Enthusiasts has my full attention. It's going to be hilarious.
SimonGosejohann
Banned
Matt, you are saying this is not true? Are you saying that PS4 PRO is more powerful than the Xbox Scorpio thanks to "FP16"?
There's a difference in a bespoke feature trying to give impression of 2x performance then features that allow the gpu to be more efficient in handling bottlenecks. At the end of the day these are just features to help use full advantage of the hardware. There's not special magic to make a gpu be more powerful then what they do. 6TF vs 4.2tf that's all there is to it. Like I said MS did custom work that allows the gpu to be more efficient and devs can take advantage of
Are you joking here?
I don't know...
http://www.neogaf.com/forum/showthread.php?t=1358475
First line...
And then... is something on every XBO.
dr_rus
Member
Wait, what? Let me explain it differently. There are two ways to get some benefit from using FP16 on modern GCN. The first way which is applicable to all GCN3+ is to pack two FP16 registers instead of one FP32 one and process them at FP32 speed. This saves on chip memory (registers, LDS) and allows you to generally get two results faster than you would otherwise. The second way which is applicable to Neo and GCN5+ is to pack two FP16 registers and process them at double FP32 speed. This adds the advantage of a higher speed processing to the previous advantage of getting two results faster.
In both cases you're looking at the same bandwidth usage as two FP16 numbers will consume the same bandwidth as one FP32. You're winning in latency of shader operations which basically means that you're winning in performance.
Buffers is generally something which reside in device memory (VRAM) and they have no relation to shader processing as you're writing results from the shader execution to some surface format which in 99% of cases is either 32 or 64 bits already.
Here's another analogy: most data used in modern games are stored in 16 or 32 bits image formats (meaning either 4 or 8 bits per component). GPUs are processing this data in shaders with FP32 precision though (which is 32 bits per component) because this is the only way to make sure that there are no rounding errors (which will result in noticeable visual artifacts on shaded surfaces) during shader processing. This is basically similar to what's happening on the output of the pipeline - after you've processed the data with FP32 (or FP16) precision you're writing results to a memory buffer with significantly less precision because there is no need to use 128 bit precision for something which will be shown to a human eye - eye just won't see any difference between 32 and 128 bits.
SimonGosejohann
Banned
No, some people actually believe this:
According to him: PS4 Pro Peak Performance is 8.4TF. Scorpio is 6TF. You do the math.
lol, no. That would be ridiculous.
Just read the new article,
One thing i've read a few times now is that each MB is tailored to match the processor that its given.
Does this effectively mean that there will be a silicon lottery when buying a Scorpio much like we see with PC GPUs and CPUs? ie: good and bad consoles
Or has this always been the case to an extent with consoles as well? Ive just never heard about it applying to consoles before.
One thing i've read a few times now is that each MB is tailored to match the processor that its given.
Does this effectively mean that there will be a silicon lottery when buying a Scorpio much like we see with PC GPUs and CPUs? ie: good and bad consoles
Or has this always been the case to an extent with consoles as well? Ive just never heard about it applying to consoles before.
I think the whole point is to avoid that, to fine tweak each console to make sure it's as good as that specific console can be in terms of reliability?
No there isn't.
Using FP16 is just another method for tying to make graphics hardware more efficient. The basic logic behind everything here is the same, only in one case you are dismissing it, and in another you are calling it meaningful.
They are equally important. They are small features that help to squeeze a little bit more juice out of these machines.
I really do hope that you're trying to be funny
Yup I just said that in your quote good reading