Hacker News Reader

Comment by striking | Hacker News Reader

> But then, toward the end of the project, it became clear that the game would not fit in memory on the Xbox 360 (which had 512MB of physical RAM, but of course a game could not use all of that)

Wow. Less than 512MB. I can't imagine how they fit stuff like Destiny and Halo: Reach into that.

View on HN

I always wonder how saving a few dollars in RAM pays off for the effort game developers have to make to ship games. I mean the consoles sales are tens of millions, but I guess games titles are in the thousands and they spend a lot of money to cope with these limitations.

Interesting reading: How Naughty Dog Fit Crash Bandicoot into 2MB of RAM on the PS1 https://news.ycombinator.com/item?id=9737156

They actually lose money on the consoles and hope to make it back up with games. And to be totally fair, the entire 512MB was made of GDDR3.

That's right. The entire RAM bank was made of chips designed for graphics. So it's no wonder it was so expensive.

Edit: and this was necessary to decrease the latency between graphical operations and processor operations. You can load graphics directly into what is essentially VRAM, rather than having to pull it through regular RAM first. Because there is no regular RAM.

The tradeoffs made sense at that point in time, because DDR1/2 sucked a lot, and pulling data through multiple stages would drag down performance.

For the manufacturer, it really does make a difference to their bottom line. The "software tie ratio" of console games in the 360/PS3 era was modest - somewhere between 3 and 8 according to this graph [0]. Getting unit cost down matters a lot when you aren't selling a lot of additional content, and the hardware got optimized around whatever game developers could soak up the most.

As such it was conventional for game consoles to have fast-but-small RAM. The reasoning is that console games mostly bottleneck on the rendering of a scene at acceptable framerates, vs. simulating all aspects of a complex scene or achieving maximum detail as a movie would. Since ROM cartridges were fast and optical allowed data to be streamed in "fast enough", there were plenty of ways to achieve the right effect under tight RAM conditions. One exceptional case where tight RAM did not play out well is the N64's 4kB texture cache, which imposed a large burden on the entire art pipeline(if you wanted a high res texture, you had to resort to tricks like tiling it across additional geometry).

Today what is demanded from a console is much more in lock step with every other consumer device - they do more computer-like things, they can multi-task some and scenes are doing more memory-intensive things so they're more well-rounded, and get more RAM.

[0] http://vignette2.wikia.nocookie.net/vgsales/images/c/ce/Esti...

Answer: Mipmaps(+ LoDs) and streaming, lots and lots of streaming.

That would explain why Halo: Reach required a hard drive for local multiplayer. It couldn't stream fast enough from the DVD for two+ viewports.

AND that's 512MB shared with graphics, AND that was the higher of the two options Microsoft was considering for the 360:

https://forums.epicgames.com/archive/index.php/t-553976.html

Yikes. Makes bad console ports especially underwhelming. Just recently have I encountered any games that even pretend to use the abundant resources available to them on PC - and of course they are PC exclusives, like Total War Warhammer.

Yup, you wouldn't believe the amount of pain you see from doing a port from PC -> X360. Malloc is not heavily optimized since with limited memory you usually allocate up front. Lots of PC games are allocation heavy since it's fast on the desktop and large parts of major subsystems need to be rewritten to be console friendly.

I was thinking more the other way around. Slow, clunky, low-res console-first games that labor to get 30 fps while using a tiny fraction of the available CPU, GPU and RAM when ported to PC.

Ah, I misread it. There's a very good chance you're just seeing the development engine that has very little optimization since ports are usually an afterthought.

Either way they're very different environments. For instance you usually have a unified memory model on most consoles where on PC you have to shuffle textures across a memory bus to the GPU. Leads you to optimize for different things(although most ports I know of were done with a skeleton team).

Reply on news.ycombinator.com