I never said anywhere that AMD don't have working optimizations in place. In fact I said the direct opposite. All I'm contesting, blatantly, is HBM's uniqueness or the relevance of its bandwidth. And yes, you very much did imply so.
Originally Posted by Mahigan
It's possible to do on GDDR5 but GDDR5 isn't as fast as HBM..
Like here, where you very much imply the speed of HBM vs GDDR5 is some aiding factor.
I said "HBM is different than GDDR5". <--- notice the period? Then I mentioned "Its bandwidth is used like an embeded cache". <--- notice the period.
And here. If we're picking at writing structure, a period does not inherently signal the end and isolation of a train of thought... especially not when the sentence it ends parrots a statement made in the preceding quote and quite significantly affects the meaning of the following sentences. If a period completely closed off individual thoughts from one another, then coherent essays and paragraphs would be impossible unless they were just one, nonstop sentence. Notice the quote from AMD you were repeating immediately before making that statement. The implication of HBM's difference to GDDR5 making size comparisons murky is clearly there whether you intended it or not.
I mean what, you just decided to randomly throw it out there that HBM is different than GDDR5 as the opening sentence of a new paragraph with no meaning behind it whatsoever? Okay...
I never made that claim. Where in my quotation do I mention its capacity? Nowhere. What you fail to grasp is that you misunderstood what I said and have created this enormous strawman that you then write 5 paragraphs slaying.
Of course you're talking about working around capacity limitations
. That's the entire point of these driver optimizations: dealing with limited VRAM capacity and you've implied more than once that HBM's speed or bandwidth is part of that in the quotes above. Unless you're just randomly throwing facts out there in the middle of your paragraphs, for no reason, that coincidentally imply very specific meanings. In which case, a misunderstanding is perfectly reasonable.
Is HBM different than GDDR5? Yes, it ia stacked DRAM with many connection points over an interposer. These connection points or "pins" are in greater number than GDDR5 and thus allow for a wider memory interface (4,096 bit). This translates into higher bandwidth. So yes HBM is different than GDDR5.
Again, random point thrown into a sentence with no regard for context? AMD said and you paraphrased:
Main point being that HBM can be thought of as a giant embedded cache, and is not directly comparable to GDDR5 sizes.
That sentence of yours that immediately followed makes absolutely no sense in a forced vacuum. It clearly stemmed from the AMD quote.
But on the matter of how effective these optimizations are regardless, HardOCP's take on RoTR is a very limited sample to base an overall, conclusive judgment of the tech on. HardOCP themselves have had mixed things to say about Fiji's VRAM (or, well, mixed things to say in general) depending on when, such as their interpretation of Dying Light's VRAM demands way back at Fiji's launch (of course, before the supposed optimizations):http://www.hardocp.com/article/2015/07/26/amd_radeon_r9_fury_x_4k_video_card_review/8#.VtXtW_krLIU
And there are these videos showing what appears to be VRAM-related stuttering/freezes back then as well:https://www.youtube.com/watch?v=8hnuj1OZAJshttps://www.youtube.com/watch?v=k_wNEOviaeE
If these are from last year, games are only getting more demanding. And if they also demand game-specific optimizations from AMD, then Fiji's already walking on thin ice given the investment required for individual tuning. Not to mention that the most increasingly-large graphical assets continue to be things like textures that absolutely tend to demand bandwidth (system RAM through PCI-e has very harsh limits; especially in a theoretical Xfire scenario as you mentioned both due to double the power with the same VRAM and additional PCI-e limitations). But most importantly, that "hand-tuning" that needs to be done for specific games has its limits in implementation time and scope (i.e. which games get the treatment and when?). HardOCP's latest review showing VRAM not to be an issue in RoTR is very different from their previous one from nearly 3 weeks after the game's release:http://www.hardocp.com/article/2016/02/15/rise_tomb_raider_video_card_performance_review/11#.VtXxTfkrLIU
And HardOCP noted in their latest RoTR that it was the first time they actually saw AMD's claimed optimizations in action:
If you remember a good while ago that AMD was making claims to this effect, but this is the first time we have actually been able to show results in real world gaming.
Game specific, takes time, first time HardOCP have actually seen it in action, limited testing so far, etc.; the approach obviously works to some degree, but has some severe drawbacks and it's all obviously because Fiji's VRAM isn't enough under standard, non-tuned scenarios. That's no myth given the attention AMD themselves have publicly given the issue.Edited by Serandur - 3/1/16 at 12:07pm