|
Oh man I cannot wait for the 980Ti, come to papa.
|
#
?
May 23, 2015 01:59
|
|
|
|
| # ? May 30, 2024 18:47 |
|
|
Don Lapre posted:Glad i didn't just click buy on a titan X. Someone leak a price Figure between $699 and $799, allowing for custom cooler shenanigans and early price gouging.
|
|
#
?
May 23, 2015 02:01
|
|
|
You can't sell it at that price, not without issuing a price cut on the 980 down to like... I feel $550 would be comfortable in carving out a niche for the 980Ti at that $699 price.
|
|
#
?
May 23, 2015 02:20
|
|
|
I think the 980 *is* $550 at most retailers in the US. It's here in Canada where our dollar is poo poo that it's an absurd $699 or more.
|
|
#
?
May 23, 2015 02:28
|
|
|
Kazinsal posted:I think the 980 *is* $550 at most retailers in the US. Yeah, I've seen them around $530 for decent models and I think I've seen the cheaper coolers down around $510 too, after rebate. I've seen open-box Strix 980s for $435 a couple times (normal warranty but no Witcher 3 code). You can make some pretty good deals on Tigerdirect - they run a 10% off coupon and Bank of America credit/debit cards often run a 10 or 20% cashback deal at Tigerdirect too. On the other hand, you're a "customer" of two of the least scrupulous companies out there, too. I've heard people say that TigerDirect sometimes splits the transaction into two, so you get the BoA cashback on 10% of $30 and then nothing on the rest of your order into two transactions. 980 Ti at $700 would be pretty tempting, particularly if the 390X is launching at $800+. It would put NVIDIA at basically the same performance-to-value proposition as AMD, plus you get access to G-Sync on premium monitors. NVIDIA would basically be at price-to-performance parity with AMD. That rarely happens, which actually makes me think they'll price higher. Hurry up Computex, I want to see how this thing shakes out already 
Paul MaudDib fucked around with this message at 02:49 on May 23, 2015 |
|
#
?
May 23, 2015 02:43
|
|
|
Something makes me strongly doubt that the 390x will launch at $800, it'd be pretty unusual for AMD to do so but I guess they are transitioning to a new tech. Price wise though it makes sense if they're trying to compete with the 980 rather than the 980ti.
|
|
#
?
May 23, 2015 03:00
|
|
cat doter posted:Something makes me strongly doubt that the 390x will launch at $800, it'd be pretty unusual for AMD to do so but I guess they are transitioning to a new tech. Price wise though it makes sense if they're trying to compete with the 980 rather than the 980ti. 390x will launch at $499 and be 50% faster than a 980Ti. 
|
|
|
#
?
May 23, 2015 03:06
|
|
|
cat doter posted:Something makes me strongly doubt that the 390x will launch at $800, it'd be pretty unusual for AMD to do so but I guess they are transitioning to a new tech. Price wise though it makes sense if they're trying to compete with the 980 rather than the 980ti. Rumors are (Sweoverclockers) that Fiji is going to be positioned as a halo product and branded separately from the rest of the lineup, so it's intended to compete with the Titan X. Which does make sense of the numbering scheme we have so far from the OEM lineup - an 8GB 290X becomes the 390X, and they launch a Fiji, Fiji XT, and eventually a dual-GPU Fiji VR as their HBM halo lineup. As such it makes sense to launch at $800 - it should perform like a Titan X at $300 less, and it should spank a 980 pretty well. But that's exactly the spot where NVIDIA would launch the 980 Ti. Which according to that chart will perform within margin of error of the Titan X, or in other words just like the 390X. It does have half the memory of the Titan X, but 12GB is overkill anyway, and it's 2GB more than AMD will have at launch (supposedly). So for the same money you get more memory, equivalent performance, and access to NVIDIA's walled garden of premium features (the XB270HU, Gsync, Gameworks, etc). Which is why I think it sounds suspiciously good given NVIDIA's usual pricing, but there's nowhere else for the 980 Ti to go except the $800-900 unless they raise the price on the Titan X, which seems unlikely. The problem is really that AMD has spent the last 6 months teasing a card that doesn't even officially exist while they try to sell their oversupply of 2 year old crap, while NVIDIA rolled out a new architecture, a flagship grade chip, a staggeringly large halo-grade chip, and now they're moving on to selling a bargain version of it. If the 390X had launched 2 months ago AMD could be rolling out a dual-GPU Fiji VR card right now. They're too late to the party and if they launch with pricing like that NVIDIA is going to eat their lunch. Paul MaudDib fucked around with this message at 03:27 on May 23, 2015 |
|
#
?
May 23, 2015 03:20
|
|
|
For a card that doesn't official exist yet you seem to know an awful lot about it's performance.
|
|
#
?
May 23, 2015 03:29
|
|
|
Tanreall posted:For a card that doesn't official exist yet you seem to know an awful lot about it's performance. It's almost like he read the article to which he linked.
|
|
#
?
May 23, 2015 03:34
|
|
|
Subjunctive posted:It's almost like he read the article to which he linked. I too read the article with zero performance numbers.
|
|
#
?
May 23, 2015 03:41
|
|
|
So if I understand correctly, their whole lineup for Rx 300 is going to be refreshes of existing architecture and the only new stuff will be "titan killers"? That seems like a really stupid move for AMD to make, they'd be competing in a space they've never done that well in and AMD has no premium features worth a poo poo, not to mention they can't even get game ready drivers out on time.
|
|
#
?
May 23, 2015 04:01
|
|
|
AMD is the AMD of cpus
|
|
#
?
May 23, 2015 04:35
|
|
|
Hey someone has a 980ti https://www.youtube.com/watch?v=BNrLiD47vvc&t=4703s Don Lapre fucked around with this message at 04:48 on May 23, 2015 |
|
#
?
May 23, 2015 04:45
|
|
|
Tanreall posted:For a card that doesn't official exist yet you seem to know an awful lot about it's performance. It's a pretty straightforward extrapolation from the rumored number of cores. If Fiji has 4096 cores vs 2816 on the 290X, clocked 5% faster, and the performance of the cores have not decreased generation on generation, then Fiji should be a minimum of 53% faster than the 290X, assuming the performance scales linearly with the number of cores. Which was supported at previous GCN revisions, at least, particularly at 4K. This whole thing assumes that performance-per-core has not actually increased, of course, which is a pessimistic assumption, so that's the pessimistic number The other way is the "designed performance envelope" technique, where we basically assume that AMD is somewhere near the optimum for their system, i.e. they're not stupid enough to drastically overdesign a single portion of their system (an optimistic assumption). If the memory bandwidth of the 290X is 320GB/s and the memory bandwidth of Fiji XT is estimated at 512-640 GB/s, then we can estimate the performance of Fiji XT's processor elements as being designed by AMD engineers for a high-end target of around 1.6-2x the performance of the 290X. This is an extrapolation technique that a USAFA instructor taught me - if you assume the designer isn't going to make a dumb mistake then the design of a performance-constrained system is specifically aimed at certain performance characteristics. For example, speed or lift. You can figure out the target top speed of an aircraft* by looking at the wing width of the aircraft relative to the width of the mach cone at various speeds. Wings too narrow, you get no lift and you fall out of the sky and die**. Wings too wide for the speed, you rip off the wings and you die**. They're related performance and design traits. Processor data consumption and memory bandwidth work the same way, barring a drastic change in architecture. N cores requires X in bandwidth, to hold performance constant per core then 2N cores requires 2X bandwidth or they get choked out. So Fiji XT is looking like somewhere between 1.5x to 1.6x the performance of 290X, most likely, with a high end estimate of like 2x. In synthetic benchmarks at 4K the 290X does 2595 in Firestrike Graphics score, the Titan X does 3953, so Fiji should be (1.5 * 2595) / 3953 = 98% to (2 * 2595) / 3953 = 131% the performance of Titan X. You could also say that AMD might have expected a performance increase per-core of between (1.6 / 1.5) = 5% realistic to (2 / 1.5) = 33% optimistic increase in per-core performance from architectural causes - the difference between what you'd get if nothing improved and an optimistic assumption from the bandwidth they allocated for those cores. At least from the gigantic pile of assumptions on top of the limited open source data we have available. For example bandwidth from a 1GB stack of HBM is open source, and we know that it works at 1GHz, 4 stacks yielding 512GB/s, if 1.25GHz is true then Fiji has 640GB/s. e*: now with 100% more spreadsheet since I managed to locate a file I'd been seeking for more than a decade! Share and enjoy, and design your own radio controlled rocket glider. I piloted it and our team won second place because our design rocked and I'd built and flown high performance R/C aircraft before  e**: fun fact: in extremely constrained systems such as certain aircraft the difference between "stall, fall out of the sky, overstress the wings, and die" and "overspeed, rip the wings off the airplane, and die" can be as little as 11 miles per hour, necessitating a second trained pilot spotting your landing and talking you down onto the ground while chasing you in a sick muscle car. Brought to you by your grandfather's tax dollars, your father's tax dollars, and your tax dollars! Lockheed, doing it better since 1912. Paul MaudDib fucked around with this message at 06:54 on May 23, 2015 |
|
#
?
May 23, 2015 06:04
|
|
|
I was wondering if anyone else has had crashing trouble after installing the latest 352.86 Nvidia drivers. I installed them and started experiencing crashes when opening Windows Photo Viewer and Firefox, so I tried rolling back to the previous 350.12 drivers. That didn't seem to help so I did a few rounds of uninstalling, removing old drivers with DDU in safe mode and reinstalling - no luck, I've rolled back to 347.88 but I'm still getting crashes. I guess my options at this point are to keep rolling back to find a stable driver or do a clean Windows install?
|
|
#
?
May 23, 2015 06:33
|
|
|
Paul MaudDib posted:(1.5 * 2595) / 3953 = 98% to (2 * 2595) / 3953 = 131% The range is huge. You could just as easily say it's a TitanX killer and the 980Ti is a wet fart at the same price. I don't understand why you would use Hawaii(GCN 1.1) when Tonga(GCN 1.2) is more recent.
|
|
#
?
May 23, 2015 06:59
|
|
|
http://www.kitguru.net/components/graphic-cards/anton-shilov/amd-radeon-r9-390x-to-cost-more-than-700-report/
|
|
#
?
May 23, 2015 07:16
|
|
|
Tanreall posted:I don't understand why you would use Hawaii(GCN 1.1) when Tonga(GCN 1.2) is more recent. Because I think there's some design and testing differences between a flagship chip (even 2 years old) and a midrage-tier chip like Tonga. Scaling up a 2816-core processor by 1.5x is different from scaling up a 1792-core processor by 2.28x. The bandwidth difference is 512-640GB/s vs 176 GB/s, so I'd guess on the high end by that metric you're looking at 2.9x to 3.63x total performance envelope for a Fiji XT vs a 285. I'm just too lazy to find a Firestrike benchmark of a 285 running 4K against a Titan X when there's problems like that. quote:The range is huge. You could just as easily say it's a TitanX killer and the 980Ti is a wet fart at the same price. Assuming the 980 Ti performs about the same at about the same price it would be a fantastic Paul MaudDib fucked around with this message at 07:41 on May 23, 2015 |
|
#
?
May 23, 2015 07:18
|
|
|
Paul MaudDib posted:Let's say AMD aimed in the middle. (2x / 1.5x performance increase) = 1.75x average, with a spread of .25x. .25/1.75 is a represents a 15%. Yeah, I'm definitely too lazy to do stats, but I daresay most "tock" HPC chip designs take a risk of at at least a +/- 15% delta in performance from their design target, especially when integrating novel technology. 100% low end performance target, 130% high end target. Or you could look at it as a realistic plan plus a small wager that your architecture hits the lottery and scales up really well. What's your problem bud? Let's just say they aimed for the upper range and kill the Titan X or let's say they aimed for the bottom range and are a wet fart. Both true! Your range is ambiguous and no real conclusion can be pulled from it. I also like the "U Mad Bro?" Tanreall fucked around with this message at 07:58 on May 23, 2015 |
|
#
?
May 23, 2015 07:45
|
|
|
The main difference between HBM with regards to nVidia and AMD is that it *has* to work for AMD. nVidia can afford for it *not* to work, plus they have the luxury of letting AMD effectively 'soak test' HBM in a consumer product for them. One has a safety net, the other is working without one, which is likely why only the 390 series will be getting the special sauce, and all the other 3xx-series will be re-brands of existing silicon.
|
|
#
?
May 23, 2015 07:51
|
|
|
Paul MaudDib posted:Assuming the 980 Ti performs about the same at about the same price it would be a fantastic I strongly doubt the reason for AMD's delays is "because they wanted to burn their stock of Hawaii down". Isn't this the first HBM product anywhere? They likely have run into production issues -- probably a variety of different ones, I would guess. Besides, to burn down existing stock they just do a rebadge, no need to hold off the new hotness for that.
|
|
#
?
May 23, 2015 12:40
|
|
|
Dream Attack posted:I was wondering if anyone else has had crashing trouble after installing the latest 352.86 Nvidia drivers. I installed them and started experiencing crashes when opening Windows Photo Viewer and Firefox, so I tried rolling back to the previous 350.12 drivers. That didn't seem to help so I did a few rounds of uninstalling, removing old drivers with DDU in safe mode and reinstalling - no luck, I've rolled back to 347.88 but I'm still getting crashes. Computer crashes or driver crashes? My updated driver kept crashing until I turned off hardware acc in chrome (still not sure it's resolved, but no issues for a few days now) as someone suggested earlier in the thread. What does it say in the log?
|
|
#
?
May 23, 2015 15:07
|
|
|
cat doter posted:So if I understand correctly, their whole lineup for Rx 300 is going to be refreshes of existing architecture and the only new stuff will be "titan killers"? That seems like a really stupid move for AMD to make, they'd be competing in a space they've never done that well in and AMD has no premium features worth a poo poo, not to mention they can't even get game ready drivers out on time. I don't think they've announced what we get for upcoming releases? Only thing I've seen is the list of what they're giving to OEM shitboxes. Surely they can't try and re-sell the 7970 /again/.
|
|
#
?
May 23, 2015 15:21
|
|
|
It doesn't look like they have anything new for laptops either. It's official that the M375/M360/M330 are ancient GCN 1.0 rebrands, and the rumor that the X-suffixed parts would be new was shot down when the M370X in the new Macbook Pro turned out to also be a GCN 1.0 part from 2012. They'll have something new on the desktop other than Fiji thought right? right? 
|
|
#
?
May 23, 2015 16:05
|
|
repiv posted:They'll have something new on the desktop other than Fiji thought right? right? Of course they will, just because every single sign points towards the only truly new card being a $700+ "halo" product doesn't mean that it's true! 
|
|
|
#
?
May 23, 2015 16:19
|
|
|
AMD's new cards were first reviewed in December 2011 http://www.anandtech.com/show/5261/amd-radeon-hd-7970-review
|
|
#
?
May 23, 2015 16:55
|
|
|
Foxhound posted:Computer crashes or driver crashes? My updated driver kept crashing until I turned off hardware acc in chrome (still not sure it's resolved, but no issues for a few days now) as someone suggested earlier in the thread. What does it say in the log? Turns out it was driver crashes, which usually led to the computer locking up. From my events viewer it looks like the usual error was "display driver nvlddmkm stopped responding then recovered", but I managed to get system stability by doing the following: * Uninstalling drivers with DDU in safe mode * Running CCleaner in safe mode * Installing drivers 347.52 In case anyone has the same troubles, that may work. Thanks for the heads up about disabling hardware acceleration too, I'll do that in case I have any more trouble.
|
|
#
?
May 23, 2015 21:07
|
|
|
I've sat through the last 2 driver releases for my 970 because I keep reading about these issues and I've always been wary about updating drivers when the ones I have seem to work perfectly. Every time something breaks it's just like "....god drat it."
|
|
#
?
May 23, 2015 23:17
|
|
|
Mazz posted:I've sat through the last 2 driver releases for my 970 because I keep reading about these issues and I've always been wary about updating drivers when the ones I have seem to work perfectly. Every time something breaks it's just like "....god drat it." It's not all that hard to roll back to older drivers if they gently caress something up, I've always installed new drivers, be it beta or stable, usually if you're playing new releases the benefits can be pretty substantial. Sometimes drivers can be the difference between "oh I spent an hour tweaking settings to get the performance I want" and "I just turned everything on gently caress the police".
|
|
#
?
May 24, 2015 00:13
|
|
|
People in general hate messin' with their PC's these days and telling them to do a uninstall or roll back or mess with beta drivers will go across as well as a middle finger to their face. Especially since the general perception is that new drivers are supposed to be better across the board while also being stable. That goes double with hardware/software that are perceived as being 'premium'. 'God I bought this poo poo and paid extra cuz' this stuff isn't supposed to happen AT ALL. WTF' Not saying I like it or agree with the attitude. That is just the way it is.
|
|
#
?
May 24, 2015 00:47
|
|
|
Most drivers in general are going to be uniquely hosed up for someone in some way that is complete bullshit but also probably impossible to truly account for, but that should never dissuade people from installing new ones. I remember installing every new driver that came out when I had a 560ti SLI setup and every driver release I'd still have an issue where the driver crashed but, thankfully, recovered. It made gaming annoying but not unplayable since it was somewhat infrequent, but one day a new driver came along and it magically went away. Drivers are weird.
|
|
#
?
May 24, 2015 00:58
|
|
|
I think everyone's suspicions of the 980 Ti not being $699-799 might have some merit. Newegg thinks they're doing everyone a favor selling the eVGA Titan X for $1057.29, and everyone else has inflated the original $999 price to ~$1100.
|
|
#
?
May 24, 2015 08:55
|
|
|
BIG HEADLINE posted:I think everyone's suspicions of the 980 Ti not being $699-799 might have some merit. Newegg thinks they're doing everyone a favor selling the eVGA Titan X for $1057.29, and everyone else has inflated the original $999 price to ~$1100. Microcenter has the EVGA Titan X for $999 (I was in the store yesterday looking at it) which is kind of odd because they tend to mark up every card they sell 10% over MSRP.
|
|
#
?
May 24, 2015 17:37
|
|
|
You can also get the Titan X directly from NVIDIA without markup.
|
|
#
?
May 24, 2015 19:47
|
|
|
KakerMix posted:You can also get the Titan X directly from NVIDIA without markup. TD sells the Asus one for 1029 and you can get about $50 cash back with fatwallet and boa cash back.
|
|
#
?
May 24, 2015 20:05
|
|
|
Don Lapre posted:TD sells the Asus one for 1029 and you can get about $50 cash back with fatwallet and boa cash back. somehow I doubt the audience for a $1000 graphics card cares that much about $20
|
|
#
?
May 24, 2015 23:04
|
|
|
Generic Monk posted:somehow I doubt the audience for a $1000 graphics card cares that much about $20 Thats like 4 less pizzas you have to deliver.
|
|
#
?
May 24, 2015 23:09
|
|
|
They have some room to mark it up a bit all things considered but so far the only thing that has increased in price was the 980, and that was by $50. The titan stayed the same despite being a significant jump, the 970 dropped $70 from msrp. I'm going to guess $750
|
|
#
?
May 25, 2015 02:03
|
|
|
|
| # ? May 30, 2024 18:47 |
|
|
Welp, with ebay 10% back in ebay bucks and 1% cashback from fatwallet couldn't resist grabbing a titan x.
|
|
#
?
May 25, 2015 18:51
|
|