LinkBack URL
About LinkBacksRead more.Epic Games touts card's performance.
Read more.Epic Games touts card's performance.
It was running a much less intense form of AA than on the GTX580!! Its stated in the actual press release!! There is no info about the framerates and the previous demo was made over a year ago and IIRC was meant to be unoptimised anyway. We have no clue how well a single GTX580 will run it today.
"Epic is able to achieve the greatest performance from the upcoming architecture as opposed to AMD's Radeon HD 7xxx series, an indication which bodes well for NVIDIA."
Correlation does not mean causation though. It could be a case of Nvidia dumping more money at Epic games,to optimise the engine for Kepler. Until we get some actual data,it is just supposition.
Last edited by CAT-THE-FIFTH; 09-03-2012 at 11:19 AM.
your summation of the original article is a littlebit misleading - the previous 3x 580 demo was at a much higher AA level to the single kepler (the original was 4xMSAA and the new single kepler demo was on NV's latest version of FXAA), they also ran the old demo on 1.5GB cards so there's no telling if VRAM limitation was an issue which also wouldn't be affecting the kepler card
I was actually at the event and personally asked Mark Rein about the differences in rendering year to year.
The three-way GTX 580 was run with 4x MSAA. The same demo was run yesterday on a single 'Kepler' card, at the same settings, but with MSAA substituted with FXAA. The end result was a comparable image but with far less computational penalty for running the pixel-shader-based AA over conventional MSAA.
For what it's worth, a 'Kepler' card is fundamentally faster than a GTX 580 at exactly the same settings, though I can't say by how much.
That's why I said 'indicating' - I have no doubt NVIDIA's throwing some level of money at the firm and there will be some level of bias, however, factor this with Epic stating that it can achieve more from the card with UE 4 and it's not wrong to make such a suggestion.Originally Posted by CAT-THE-FIFTH
I'm not attempting to say "this is exactly how much faster the card is", the tech demo was a teaser and nothing more, however likewise, I can't verify what level of impact a move to FXAA had and if it was the bottle-neck preventing say three 580s from becoming two, so it's no more or less misleading as a teaser of performance.
Indeed, but for all we know, three 580GTXs would have still been required but FXAA was the enabling factor for the Kepler, which, whilst not performance bound could have been memory bound and this kept it down to a single card, we don't know either way.
Last edited by Scribe; 09-03-2012 at 11:55 AM.
Really? Doesn't it just indicate that nVidia paid/donated to/provided something for Epic to use one of their cards instead of the competition?
I mean, Thierry Henry is sponsored by Gillette - that doesn't make them the best performing razors for playing football.
Agent (09-03-2012)
FXAA is very fast compared to MSAA; MSAA can easily eat over a quarter of your framerate while FXAA can be basically free to around 10-15% depending on the engine. MSAA is also incredibly memory bandwidth intensive.
You can't really draw any reasonable conclusions from the demo running at different settings, and the 580 SLI setup may have been pulling the framerate down for all we know. Sure, Kepler will be faster than Fermi on release, it would be a disaster if it wasn't, but this doesn't prove it.
Such a bunch of moaners. Who cares who paid who. It's always the case, promotion is needed, somebody has to do it. The point is we have a tech demo that hopefully will be run in real time when GTX 680 comes out and will be free for download. I'm bored to death with other NVidia or ATI/AMD tech demos for that matter. This particular one is a nice eye candy. That is all.
Last edited by CAT-THE-FIFTH; 09-03-2012 at 12:20 PM.
It has nothing to do with AMD vs Nvidia, people are complaining because a game running at much less intensive settings is running faster, and a conclusion has been drawn from that somewhere along the line. I'm sure all of the posters would be saying the same thing if it was 6970 vs 7970 in the same situation.
And about 'who paying who', a conclusion drawn that a single application runs faster with one card than one from another brand again means nothing; I could write or give examples of code that will run an order of magnitude better one one brand or the other. The point here is that Nvidia are probably cooperating with Unreal to optimise the code so it comes as no surprise that a newer card running optimised code is performing better than one from the other brand.
CAT-THE-FIFTH (09-03-2012)
No one's drawing a conclusion because there aren't enough facts to draw one from; it's also highly likely that Epic has worked equally hard on optimisation for AMD graphics, especially given that we're expecting to find them in the next-gen consoles, which is where Epic states it intends to focus with UE4.
the kepler could have been memory bound? but it has more memory than the GTX580's they used (1.5GB models), so erm, I'm even more confused by what you are trying to say now
Well surely we can be forgiven for thinking that?
What Epic did confirm was that the demonstration of UE4 shown behind closed doors was indeed powered by NVIDIA Kepler technology, strongly indicating, that at this stage, Epic is able to achieve the greatest performance from the upcoming architecture as opposed to AMD's Radeon HD 7xxx series, an indication which bodes well for NVIDIA.
What's more, is to tease the public as to the actual performance of the Kepler GPU, Epic ran, a live, real-time demonstration of its previously released Unreal Engine 3 Samaritan tech demo running on a single Kepler GPU, which last year, required three GeForce GTX580 cards in SLI to achieve equivalent frame-ratesIt's not just about the amount of memory, bandwidth is incredibly important for GPUs* and as I said earlier, MSAA is very bandwidth-intensive. For all we know, the memory bandwidth of the architecture may play a large part in the need for 3 cards to run the demo at a decent framerate; running FXAA a single 580 may be able to handle the demo fine.
*As an example, look how much of a difference choosing a faster DDR3 speed makes for AMD's Llano IGP and this is only a small GPU.
Because there's three of them in SLI 1.5GB x 3. Assuming an optimal set-up, both memory usage and bandwidth usage could have been spread across the cards to a certain extent, especially when deferred rendering and MSAA are framebuffer heavy.
OFCLike I say, it's all conjecture, hopefully we'll find out soon enough for sure.
You can't share framebuffers between cards. Doesn't matter how many cards you have in SLI, the 1.5gb would be the limit, especially for MSAA.
There are currently 1 users browsing this thread. (0 members and 1 guests)
Posting Permissions
Reply With Quote
