It's a big problem if review sites get a cherry-picked card and just call it a '1080' in benchmarks, so people expect that performance when in reality they'll be getting significantly less on stock cards.Originally Posted by Marc_92
Oh and I still think 1080 is a terrible name, I thought it was just a place-holder name like Xbox 720 - not something that would actually be used! Then again, the name that replaced 720 wasn't exactly great was it? xD
Nvidia did it with the boost v1 which sites had to modify their testing methodology after the initial reviews. But the damage was done as the false scores made it look faster than it did.
FE will likely just be the early production cards - ie high cost, low volume, before bulk production. Looks like they'll have those fancy|ugly expensively made coolers because when you're paying a premium for early silicon the relative proportional costs of the other gubbins goes down.
It's quite possible that the volume production cards, when they arrive later, will be able to clock better as the manufacturing process matures - which is what AMD found and led them to release the GHz edition cards after production had matured.
All in all, a clever take on what to do to get early silicon out there into the hands of (a few) consumers and review sites, while not having the volume for bulk orders - this way a limited run will get an easily met premium and everyone is as happy as they could be.
edit: side note - they also took aim at AMD's PR with a point about supporting async compute "New asynchronous compute advances improve efficiency and gaming performance." Whether that's real or a software trick we'll have to wait and see.
Last edited by kalniel; 08-05-2016 at 09:04 AM.
Which means they finally admitted Maxwell does not!
Wasn't,Maxwell made for DX12??
Also, leaks from sources which have been unusually accurate in the past, indicated support is closer to GCN1.0 level.
Which is why you don't buy anything until Polaris high-end is here.
Society's to blame,
Or possibly Atari.
Yes, and they even listed async compute under the DX12 features that Maxwell had, so we can't really gain much understanding from this statement eitherWhich means they finally admitted Maxwell does not!
Wasn't,Maxwell made for DX12??
still waiting for Polaris 10 and its low priced release.
Has anyone speculated yet what the numbers for a couple of 1080s in SLI will be?!
Off-topic a bit but I've seen a rumour (based on pixel-counting an AMD slide, so it's entirely possible it's just a mock-up) that the Polaris 10 die is a bit smaller than the 1080 with people drawing performance estimates from that, however it's important to remember that unlike previous generations, AMD and Nvidia are sourcing their GPUs from different foundries with AMD going with 14nm from GloFo/Samsung. Remember that the Apple A9 was a bit smaller on 14nm than TSMC's 16nm so, although it could vary based on layout, it does look like GloFo/Samsung is denser than TSMC (you can't always draw conclusions from process node naming). Combine with that the nodes may differ in more than just density and you really can't just look at die size for anything more than ballpark estimates.
But yeah I agree it's very concerning that reviewers might get the highly-clocked cards, do the benchmarks once then not get any more free samples to re-do the reviews on standard cards. Reviewers really do need to keep benchmarks up-to-date - there are even a couple of sites who didn't bother updating their charts when the 280X came out because they already have 7970 results. Fair enough it's pretty much the same hardware but drivers had improved performance significantly by that point so comparing old 7970 numbers to years-newer competitive benchmarks is just plain misleading.
The difference is tiny between the two A9s though - process yield is a more important factor (and we'll never know how that compares).
But more relevantly to performance were AMDs comments that they were targeting mainstream while nVidia were high end.
True, but like I say it can vary depending on layout. In fact Nvidia could end up being denser on the 'less dense' node in theory.
The term 'mainstream' is very arbitrary though and AMD haven't to my knowledge elaborated on what they mean by it (plenty of rumour mills have done it for them of course). Going on die size/transistor count alone then you could say that about the 1080 too - it has 7.2B vs 7.1B of even the GK110, and lower than the 8B of the GM200. It looks like a lot has been gained in terms of clock speed though, but like I say if you just look at die size or even transistor count you'd be way off the mark in terms of performance.
What AMD might be considering as the 'high end' served by the Vega10, like the GP100, is still some way in the future. But that's probably sensible - I expect that sort of huge die on 14/16nm would be monstrously expensive at the moment.
The die size of the 1080 pretty much fits with what I've been expecting - much smaller die than the previous 'high-end' but with a huge die coming much later. The 28nm node has muddied the waters a bit as they wrung so much performance from a very mature node with reticle-size dies.
It was AMD saying that Polaris was in a lower market segment than the 1080/70 themselves though - they didn't say 'our product is this' and nVidia say 'our product is that', the comparison came all from AMD.
http://arstechnica.co.uk/gadgets/201...ainstream-gpu/
While they also said Polaris 10 targets "the mainstream desktop and high-end gaming notebook segment."
Ah I see what you mean. However you could read it as Pascal being only a high-end part for now, whereas AMD have two Polaris GPUs ready to go. Of course I'm just speculating though.
I didn't read too much into the mainstream desktop/high-end mobile part - don't forget GM204 made it on to laptops too.
The power usage figures are impressive, for sure, but the estimated performance increases not so much. To me this seems like Nvidia just want to get something into the market.
Indeed, can run a 1080 which is apparently somewhere close to 2 980s off of a single 8 pin connector and apparently their demonstration was running at over 2000 MHz while sitting on around 67c, which is presumably the stock nvidia cooler. So performance improvement will be nice but relatively standard, but the power consumption and therefore heat output seem to be the real selling points here.
Ofcourse, that does mean overclocking could be far better on these cards if it doesnt get hardlocked. No doubt manufacturers, gigabyte in particular tends to this will I think, will release some cards with 2x8 pin so you can push a whole lot of power through them. Boost should also be better, but boost should already be factored into the current statistics.
Still undecided whether to buy into this generation or not personally. Going from a 760 to a 1070 or even 1080 ($50 bump might knock that out of my price range once the manufacturers take their cut) would be a significant achievement and a drop in heat and power, but I still feel like this is a half baked release since HBM2 is right round the corner. Though I guess realistically the first time we see HBM2 will be the 1080ti at the tail end of the year, and presumably sometime towards the middle of 2017 for HBM2 to actually end up in the 70 and 80 cards.
Last edited by jag272; 08-05-2016 at 06:37 PM.
I'll be buying into 1080... I've just invested in new rig as want to get back into gaming. Retired my ageing 8800GTS and am currently running on-board Intel graphics, so bit desperate.
There are currently 1 users browsing this thread. (0 members and 1 guests)
Posting Permissions
LinkBack URL
About LinkBacks
Reply With Quote
