@jgf said:
@alexglass said:
I've already explained what I believe is bullshit. Taking a specific area such as the GPU where the PS4 has a nearly 50% increase in compute units, and using it a blanket statement by turning it into "the PS4 is 50% faster" overall.
It's just not supported by any of the specs we have. At most the PS4 has a 40%-50% advantage in GPU, Outside of RAM and GPU, all the other hardware points in favor of the X1. CPU, CPU bandwidth, audio chip, eSRAM, data move engines.
To be fair GPU is the most important part for the performance of games. Just look at what happens when you upgrade your PC with a new graphics card vs. when you put a better CPU in it. Putting a 50% faster graphics card in your PC will have far more effect then putting a 50% faster CPU in there. So this is not just "bullshit", the GPU is a very important part and when its faster its a big deal.
I would rather say that claiming the Xone is ahead in "all the other" departments is far fetched. We don't know the final cpu speed of PS4 yet, as it stands its 1.6ghz vs 1.75 - so roughly a 9-10% faster cpu. CPU bandwidth is the same afaik the Xone has 30gb/sec and the PS4 too, at least if you count the additional 10gb/sec of the onion bus. The PS4 has an audio chip too, granted it may be true that the xones chip is better - but what does that mean? Will it help to run games smoother, if so how much? Thats a bit too much speculation for me pinned on a vaguely "better" audio chip. Counting the eSRAM as an advantage is two folded. Its there to migitate the otherwise far to slow DDR3 memory. I wouldn't count the eSRAM+DDR3 combo as an advantage over the GDDR5. It leaves some wiggle room for future optimizations though and the PS4 doesn't have one, so in that way you may count it as an advantage. But its questionable at the very least. Finally the data move engines may count as an advantage, I don't exactly know what they do (asynchronous transfer I guess?) and I also don't know if PS4 has something similar, but I'll grant you the benefit of doubt there.
Yet you don't mention specific advantages of th PS4, e.g. no kinect and a leaner operating system. Kinect has an appeal on its own for some people, but performance wise its an additional task that the system always needs to dedicate compute time to. They have built in special chips to minimize the overhead, but its still something to consider.
I can only go based on the numbers we have. So if the specs turn out to be different then things change. Just like I think most of these quotes are probably from developers that don't actually have dev kits with the upgraded tools or specs MS has made.
For example 50% more CUs leading to 50% more computation power was actually accurate prior to the GPU upgrade when both GPUs were running at 800MHz. It's 12 vs 18. But that was months ago. Now it still has 50% more CU's but doesn't quite lead to 50% more computational power. The 12 CUs in the X1 run at 853MHz and the 18 CUs in the PS4 run at 800MHz. However that is still a significant advantage in favor of the PS4. So are ROPs and pixel fill rate.
But on the CPU end, based on what we have the X1's CPU is faster and has a coherent 30GB/s bandwidth vs 20GB/s on the PS4. That's a small advantage in favor of the X1.
Bandwidth isn't just speed. But acceleration + top speed. Here's where again the PS4 has an advantage but it's not all one sided. GDDR5 is better at top speed. DDR3 has tighter timing, better latency or acceleration. People want to say latency isn't a big deal with GDDR5 anymore but we'll see. So far we have seen more PS4 games suffer from pop-in than X1 games. Infamous, Drive Club's reflections, Killzone. When you're talking about pop-in from loading on demand, that's usually a ram issue. It's taking too long to stream and swap the objects in. If the PS4 software tools are more developed, isn't it at all odd that multiple games from multiple developers are displaying these issues? What's causing it? Are they all writing bad code? Let's just say it's an API issue, who knows but it could also have something to do with this.
Now when it comes to eSRAM it doesn't interest me as much that it mitigates bandwidth. The amount it mitigates if we're talking about moving asses is so miniscule that it won't matter. The PS4 still has a huge advantage in bandwidth to main ram.
You want to talk about PS4 advantages? More bandwidth means you can load a lot more of them at once. But since they both have roughly an equal number of usable RAM, around 4-5GBs, this really should only lead to being able to transfer a larger number of higher res textures or objects through bandwidth. In addition to that though there's a lot of other things in game development that are very important
But that's not what makes the eSRAM important or interesting to me. To understand why we have to ask ourselves what RAM and bandwidth targeting in terms of actual game assets or techniques. Well it's mainly responsible for storing textures and assets, characters, objects, etc. Right? So more RAM means more textures and assets. But that's not what eSRAM's primary purpose is. It's on chip and it's capable of offering blistering speeds for a small amount of RAM. Doesn't seem like it would be built for that.
So how can this be useful? We have a lot of new software development techniques that appear to me could favor the eSRAM set up.
Procedural textures. We know for a fact, they don't require lots of RAM or lots of bandwidth. They require calculations. Now knowing what I know about procedural textures, and I have been keeping up with this for months, there is no doubt in my mind they will play a huge role this generation. Search for Allegorithmic's Substance engine, and go look at how many studios bought this middleware for next gen development. Partial resident textures. Again we've seen example where a texture as large as 3GB can be stored in 16MB of RAM. Granite was on stage with Microsoft showing of their Graphine PRT middleware at the DirectX 11.2 build. We now have compute units. Now bandwidth may be important here, but do they need a lot, or are they more suitable to a fast access on chip eSRAM? Some of these may be helped by an on chip scratch pad.
So I'm particularly interested in finding out what exactly the eSRAM's purpose is as it relates to software techniques because I don't believe for a second MS put in the X1 to mitigate main bandwidth, when it's obvious that's not it's strong point in the first place. They see what predominant techniques are taking over development. They also know what type of features they're working on for DirectX11.2.
For example during the Xbox and PS2 days it was multi-texturing in a single pass, where both GC and Xbox had it, and PS2 didn't. The PS2 could push around a whole bunch of flat shaded polygons, but it dropped like a rock in a pond when it came to textures. It made a huge difference in graphics. From bump mapping to displacement to multi-texture effects, that ended up becoming a very important difference. It wasn't so much raw power that produced the big visual gap between Xbox and PS2. In large parts, it was that. Multi-texturing in a single pass. That ended up being the largest defining factors between Xbox and PS2 games.
So my questions are, does the eSRAM provide any benefits to procedural textures and procedural generation? We know procedural textures just aren't going to care very much about bandwidth. They're tiny. But they care more about compute. Now if I'm doing compute on my GPU, and it turns out I have to go back and forth to RAM, I think I'd much rather have blistering fast on chip eSRAM to write and read from.
How do the move engines and eSRAM benefit partial resident resources? Once again, they take up little space, and streaming tiles isn't going to require a whole lot of bandwidth. But when it comes to streaming things from RAM, latency will play a bigger role than top speed. Low latency will probably affect texture pop-in more.
How does the eSRAM affect GPGPU?
Because to me everything is pointing towards those areas will end up making a big impact on the way games will look this generation just as much as resolution and the number of unique assets you can store and stream. I've said it before and I'll say it again. Specs and raw power don't decide graphics. It's the software techniques of the times, and which chips are better suited to running them. They go hand in hand, and often times manufacturers will build chips to suite particular techniques that developers are using.
Finally, and this one is a bit off topic here, but to me, Kinect is not a disadvantage. I feel the exact opposite. How can you possibly see a system where it isn't capable of running an entire line-up of games as something advantageous? A system where voice recognition and gesture recognition is an accessory dropped to the side? I think that's actually going to end up creating the largest rift. One console is capable of giving you both a Wii like and core gaming experience. And one isn't. In addition, voice recognition and gesture recognition will probably end up affecting core games in a big way by the end of this gen. Now that it's standard for an entire generation it's going to end up making its way into core games, and who know what type of games that's going to end up leading to. That's a big advantage in favor of the X1 if you ask me. HUGE. That's a much bigger impact than a small difference in graphics. And I'd also hold off on labeling the PS4 as having " a leaner operating system". It wasn't too long ago that the article came out where the PS4 will also reserve a large part of its ram for its OS and relate apps. And considering how many non-gaming related things it's also trying to do, we'll see how that pans out.
Log in to comment