While ultimately frame rate is king. Surely those hexus guys should have graphed the memory usage throughout the benchmarks, when the review focuses on two cards from each camp whose main difference is memory allotment.
I assume you have tested 2 of each card in SLI on a selection of PCIe lane bandwidths running different resolutions, with a comparison chart to back that statement up? Happy to argue the toss with you Harry, if it gets bit to pull their finger out and finally do a proper SLI round up for high end users.
sli doesnt swamp 8x pcie let alone 16x pcie so pcie lane bandwidth wouldnt come into play; if you want ` an ultimate` round up , then do - 2gb and 4gb gtx 670/680 and 3/6gb 7950/7970 in sli/cf up to 3 x 30" screens - you know only about 6 grands worth of kit.....
I'm sorry but both Simon and I have demonstrated a marked difference between SLI @ 8x and SLI @ 16x with larger resolutions. http://forums.bit-tech.net/showpost.php?p=2904810&postcount=21 The difference is greater still at lower resolutions, as CPU can be the limiting factor. Statements like "SLI doesn't swamp 8x" are clearly misleading.
That vega video was with 4 x 2GB cards. The bus was getting choked, but only when the vram ran out on pcie 2, though admittedly it is a super extreme example. Edit: you know what would really clear this up? A thorough labs test with a selection of GPUs and platforms
If I can demonstrate a difference between 8x and 16x with a synthetic benchmark that only used 1200MB of VRAM, then I fail to see why a game that uses 2700MB of VRAM would deliver any different results.
currently no game uses 2700mb of vram - the most and `max` situation on hexus was 2204. which , as shown, means its pointless getting a 4gb or 6gb card.
Crysis 2 with very high levels of AA and texture pack etc uses 2.7 @ 1600p or 2.4 @ 1440p BF3 on max at 1600p can go upto 2.4
Alot of people are too busy trying to justify purchases rather than been helpful here. if you game above 2560 x there might be a situation where it would help but they are few and far between ( crysis 2 with high end texture pack at max settings is the only current game ) by the time most require more than 2gb of ram on this site they would of upgraded again so as baz has said its nigh on irelivent. id still personally save £200 and get 2 670 2gb cards and sell them on next year and buy the next gen cards which will undoutably come with more memory. £200 is half the cost of the next gen card in savings even if you get £200 for each 670 thats still enough to buy 1 and half of the next gpus.
Maybe. But then again having people with real life experience of a card you are considering buying is really helpful. I was in a similar quandry myself a little whole ago, trying to choose a gpu to drive my 1440p res. But i decided to go for the wildcard option (a 2nd hand 7970 for £250). Im pretty sure that with amd the additional vram did make a difference, but with the nvidia cards its not so clear cut!
The difference seems to be approximately 1 frame per second on min and average frame rates. Max was up by 6 frames on dual 16x. It may be marked but hardly significant
The link to the other thread I posted clearly shows Crysis 2 and BF3 both using over 2GB, with Crysis 2 using 2.7GB of VRAM. ...and BF3! I've seen 2.3GB being used with that. It's up to the purchaser of course, but because something is adequate now, is never a good reason to buy. I'd like a little headroom.
The memory usage of the GPU is all estimated, and usually very wrong, as each GPU architecture manages the memory completely differently, and the GPU has no system in getting the memory usage. Some card the memory sage is 100% all the time, as it uses technologies similar to Windows superfetch, just for example. So do programs that measure this consider this or not? What if the GPU uses a compression system on the texture to load more textures? What if the GPU pre-loads future textures? So it's all estimated, and guesses.
and as numerous reviews have shown - anything more than 2GB for triple 1080p displays is pointless. just as 16x > 8x pcie yields 1 `avergae` fps , which is within the margin of error anyway.
This may be true at the moment. My question is, can the 670/680 effectively utilize more than 2GB ram due to its lower memory bandwidth compared to its predecessors and AMD equivalent.
I've been studying Vram behaviour with EVGA Precision and my conclusions are as follows: I think some games will detect how much Vram you have and stream textures/data appropriately. Others will load as much as they can into the Vram and loose the bits that aren't needed when new textures/Data needs to be loaded in, much like windows cache. BF3 isn't a good benchmark for Vram because you never need much more than 1.5GB as it is the Textures that take up the space, and they are the same no matter what resolution you have. The only difference is when you turn them down and the lower res textures start popping in and out. There may be a few rare circumstances where you might want double the video memory like with GTA4 (when most cards were 512MB) and turning the settings much above the recommended settings would mean that the data would have to be swapped back and forth from the system memory over the PCIe bus, causing jerkyness.
Sadly EVGA Precision can only guess RAM usage, as I explained. GPU's don't work like CPU's, and nothing is standardized. The only way you can really know, is if you are an engineer that worked on the memory management part of the GPU, and have all the tools to monitor and scan memory usage, and even then.. you come down to: how do you measure this.