TR Forums

Chrispy_ wrote:

Captain Ned wrote:

And at this point it's time for everyone to retreat to a neutral corner and take a few deep breaths before anyone says something I might regret.

Thanks for listening.

This seems to have become awfully personal again and not really of any interest to someone who might want to read the thread based on the thread title.
Can we put this to rest please, it's clogging up the frontpage forum thingy.

Chrispy has a very valid point. This is an argument best taken to PM or email rather than splattered all over this thread.

Thanks for listening.

Captain Ned wrote:

Chrispy_ wrote:

Captain Ned wrote:

And at this point it's time for everyone to retreat to a neutral corner and take a few deep breaths before anyone says something I might regret.

Thanks for listening.

This seems to have become awfully personal again and not really of any interest to someone who might want to read the thread based on the thread title.
Can we put this to rest please, it's clogging up the frontpage forum thingy.

Chrispy has a very valid point. This is an argument best taken to PM or email rather than splattered all over this thread.

Thanks for listening.

Captain Ned:

auxy wrote:

Waco wrote:

I did, and yes, while you come to the same conclusion, your reasons for doing so are totally wrong (both the car and GPU explanations).

No, they aren't, and screw you for saying so! щ(ಠ益ಠщ)

And doubly for not even attempting to justify yourself! Your trollish behavior is noted! ಠ益ಠ)凸

Wow. Just wow. Where were you when Auxy replied with that? Auxy even gave Waco the finger!

auxy wrote:

I hope you recognize when you're outclassed, but you probably won't. Just get away from me. You disgust me.

What the heck?

I think I've decided to put Auxy on my foe list just so I don't have to read this kind of crap.

JohnC wrote:

lol

Good one.

The sad part is, I don't see anything terribly contradictory in the last page- just different starting points for the same perspective. But I wonder if auxy hasn't already earned a vacation.

Oh, one thing that did stand out: 'you need more memory bandwidth to take advantage of more memory' -- while innately true, it still depends entirely on the circumstance. Typically, you don't need more memory at a particular latency/bandwidth level to do more things at once- you need more memory at a particular latency/bandwidth level to prevent the action of switching to one of the several things you're doing at once from having to reach back to memory with latency that's at least an order of magnitude higher or whose bandwidth is at least an order of lower.

You fix the problem either by a) adding more memory or b) making that next lower-level of memory faster. More VRAM helps GPU's keep out of far slower main memory across the high-latency PCIe bus, and as an example of the latter, SSDs decrease the latency and increase the bandwidth of main memory's next lower-level.

So, if we had GDDR5 on the CPU and GPU drivers that could account for the latency, which should actually be fairly easy, then we likely could get away with not putting so much VRAM on graphics cards. But we don't. We have DDR3, with DDR4 realistically over a year out, and DDR4 is both higher latency (from an absolute measure of latency in nanoseconds, let alone massive increases in per-clock latency) and not terribly higher bandwidth than DDR3- likely around a factor of two, which doesn't begin to close the difference in bandwith with top-end GPU GDDR5 implementations.

Which means that, yep, we are going to need more VRAM on our GPUs- both on average, to the tune of 4GB-6GB, and on the high-end, to the tune of 6GB-12GB. And no, Titan isn't high end. It's not a consumer graphics card. But Nvidia could easily ship a GTX780 with 12GB bolted on tomorrow if they so chose, and it'd likely cost them no more than an additional $20-$30 to manufacture over the 3GB model.

DDR3 may be that cheap, but GDDR5 is still upwards of $10-15/GiB.

JustAnEngineer wrote:

DDR3 may be that cheap, but GDDR5 is still upwards of $10-15/GiB.

But does that invalidate the point? Apple will charge you $100 for 16GB of flash memory. When we imagine what GPU makers charge for a little more memory, is it even remotely related to the price? At $15/GB, for example, 12GB of VRAM is $180. If you were already paying $90 for 6GB, should the hypothetical card actually retail another $300 or $400 higher?

My point is that VRAM- even GDDR5- isn't that expensive. What's missing is the demand to produce said cards, and the competition to keep the pricing in check. I expect that issue to be alleviated as game engines and the games built upon them evolve out of the constraining conventions of the current console generation and into the artistic freedom enabled by the excess of memory available for graphics of the next.

If I could pay for extra GPU memory at near-cost levels I'd do it every time. Going with a 4 GB card over a 2 GB card for an extra $20-$30 would be a no-brainer.