Dictator93
Member
It could be quite usefully for bandwidth heavy alpha effects and such.
-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
AMD Radeon 300 series (possible) specs
- Thread starter wachie
- Start date
It could be quite usefully for bandwidth heavy alpha effects and such.
To be frank, I'm not too impressed by Tonga's compression algorithms. Either the card is really unbalanced or something else is up.
KojiKnight
Member
Lower power and not eating up half the board space means we might actually see higher end cards in smaller form factors as well. New generation of HTPCs that can actually be good at gaming incoming!
I never said anything about performing worse. Just that its advantage isn't exactly that big of an advantage if its best use cases are for situations where its limited vRAM becomes an issue anyways.
For me, I have to wonder whether that's really worth getting a refund of my 970 now, using my old 670 for possibly up to 6 months or more, and then possibly paying even *more* for a 390 that will probably suck more power and not really give me the memory advantage that I'm waiting for.
I plan on moving to a 1440p monitor at some point and when the Oculus Rift CV1 releases, I basically need something that can handle resolutions higher than 1440p(as it renders content higher than display resolution). I don't want to keep upgrading every year, so I'm hoping to find something that will last me longer. And I'm worried a 390 isn't going to be the answer due to its limited RAM. But I also don't want to be using a GTX670 for a whole nother year or more.
But yes, an 8GB 290X is a potential option, but maybe a bit too power hungry for my PSU.
Dictator93
Member
The card is super unbalnced for its price, AMDs compression tech may also be not the greatest.
KillaKalz347
Banned
The Gaming Evolved app is Raptr, it supports DVR functionality now, never used it. But I'm sure it supports desktop capture.
OminoMichelin
Member
It does, but in my experience it's not very good. The overlay causes some performance problems when recording.
Because in many ways people are telling the truth. We have actual proof in the DX12 thread, you guys just don't like to see it. Why would you ban people for telling it like it is?
http://www.neogaf.com/forum/showthread.php?t=987116&highlight=dx12
Anyway, what's with the 4gb of ram? With all the talk of "future proofing" around here, you'd think there would be a little more concern with the fact that these cards won't last any longer than the last generation of cards. Yeah, they'll be faster, right up until you hit the 4gb mark and they choke just like every other card. I didn't upgrade to a 980 because it's only 4gb, I'm sure as hell not going to spend even more money to do it in the middle of 2015. Either this is a huge mistake on AMD's part or there will be more ram on these cards and we just don't know it.
Because as Serandur said:
I do agree 4gb isn't enough.
Especially as the high end has such a huge bandwidth and gpu power it would be foolish not to downsample many games.
Could you link to said proof? browsed a few pages of that thread and all it's boiled down to is X1 talk at this point.
Where are these news sites getting this 640GB/s number? Hynix is currently only providing chips with maximum of 128GB/s bandwidth and four of those only add up to 512GB/s.
The proof in is how poorly AMD drivers deal with CPU utilization in DX11 in those benchmarks. DX11 is 5 years old. This whole narrative that it's some internet meme and not reality is false. Hell, the Omega drivers, which were a significant improvement, weren't even released until the end of last year. How can people claim the drivers have been good since 2006, 2010, whatever year they want to pull out of their behind, when it was literally 3 months ago that they significantly improved? And even as good as the Omega drivers are, these benchmarks were done after they were released.
Well that's unfortunate that they are stuck with 4gb. All Nvidia would have to do is release 6-8gb cards to turn these into 970's all over again. They show a few benchmarks of games at 4k using >4gb ram and these cards would be destroyed. HBM is sexy as hell, but they probably should refine it a bit more before using it on their high end cards.
mugurumakensei
Member
I'd also expect these chips to not be that power efficient. Clock for clock Nvidia GPUs currently use a lot less power. I doubt first-gen HBM will bring down the power usage that much considering it is AMD at the helm. Like mantle, it will cover up AMD's deficiencies til its competitors have their answer.
Because people are never talking about "poor" CPU optimization when they discuss AMD drivers, they're talking about graphical performances, graphical glitches, "missing features" (supersampling, PhysX, you name it) or game crashes.
Those (outside of PhysX which Nvidia just bought out anyway) are mostly a myth.
PC GPUs don't exist to provide any kind of proof of future performance. The entire premise is flawed. These devices are superseded at their price point 18-24 months following release. Sometimes sooner. There is no future proofing. There is only, 'what I am willing to spend'.
It's simply at total odds with the realities of semi-conductor development and especially that of the GPU, the performance of which scales almost linearly with transistor count.
In the case of the design in question, there has to be a HBM1 design so they can sell some units and make money to cover the R&D costs. There'll be an HBM2 design later on that can support higher densities. The HBM1 process was only very recently finalised I think.
It's not a huge mistake because the product will certainly out-perform their existing design by a considerable margin and not everyone owns a 290x or 970 already.
I wouldn't bother at all. Hunt down some benchmarks and you'll find a 970 performs to within 5% of an 8GB 290X at 3840x2160 and the 970 is on a par or faster at lower resolutions.
Thanks for the answers. By "it does", do you mean it does support desktop shadow recording?
Because I play almost everything in borderless windowed ("desktop"), so anything else is almost useless to me. I have shadowplay running 24/7 on my 2560x1440 desktop and it works very well.
KojiKnight
Member
Currently.
Ricky Ricardo
Member
Artist trolled too much for his own good. Pretty sure he's perma'd
SolidSnakeUS
Member
I hope the 390X owns the GTX 980 not only in performance, but in price. I may actually go for one from my GTX 970.
Shame. Oh well wachie will take his place.
Makes sense. At some point I put him on my ignore list. Whenever I do that I usually see a few hidden posts here and there from those people, and then at some point they just stop.
Hmmm, wachie is on my ignore list now...
Though I'm considering removing him, from what I see when other people quote him he doesn't seem as batshit crazy as artist was.
And the fuckers don't even use the physx technology. It's like that time Creative bought out Aureal and just fucking killed spatial audio technology.
It's not. The new devices use an HBM interface to stacked DRAM modules I believe. DDR refers to Double Data Rate, which just means it can transfer data twice in one clock cycle.
GDDR5 is just another type of DRAM.
BoboBrazil
Member
Only 4GB? That's so disappointing considering I recently bought a 4k monitor for 4k gaming.
adamantypants
Member
Pending price/performance looks like AMD might be the upgrade to my 780ti
WhatYearIsIt.jpg
I would like to upgrade to a 390, but I currently have an HD 7970 with 6 GB of VRAM. I am not sure whether or not losing 2 GB of VRAM would negatively impact my performance, or if it even matters because games don't seem to use that much VRAM at this point.
This is the reason why. For now they're using the first generation of HBM which allows 1GB stacks, generation two is rumored to be ready by next year for their 400 series. In theory they could add more stacks, but the problem would be the interposer (the thing connecting these RAM stacks to the GPU) being too big or too expensive to make. I assume more stacks also means a bigger and more complex GPU because it still needs to communicate with the RAM.
General Lee
Member
I don't think it's explicitly stated anywhere that the amount of stacks will be 4. The slides merely talk about 4 Gb stacks being the limit. Of course 4 GB is probably still the most likely case.
BoboBrazil
Member
This is the reason why. For now they're using the first generation of HBM which allows 1GB stacks, generation two is rumored to be ready by next year for their 400 series. In theory they could add more stacks, but the problem would be the interposer (the thing connecting these RAM stacks to the GPU) being too big or too expensive to make. I assume more stacks also means a bigger and more complex GPU because it still needs to communicate with the RAM.
If this is the case how come the PS4 got 8GB and why couldn't they use that technology for a graphics card? Also isn't there a 6GB graphics card already out by Nvidia?
ashecitism
Member
PS4 doesn't use HBM, neither do the 6GB Titan or the 8GB R9 290 variants. They use GDDR5, this is new technology. Stacked DRAM.
SolidSnakeUS
Member
Isn't that because it shares the memory between the GPU and system RAM?
I think that's Q1.
MaxVarEmreis
Member
No. April, May, June.
I thought you were artist second account. Same style on the avatars but with less trolling.
No sorry. I'm a Nvidiot through and through, he is an AMD fan.
We are definitely not the same person.
PS4 has GCN GPU that has 256bit memory interface, 32bit communication with each of its 8 chip mounting points. What Sony did there was instead of mounting 8 chips, they added 2 on each mounting point [one above and one below motherboard, taking advantage of specific feature of GDDR that 2 chips can share one communication array] and mounted 16 512MB chips [which we though were not manufacturable then in high quantities]. When Cerny announced 8GB, GAF freaked the fuck out with good reason.
Nvidia achieved 6GB by having 386bit memory interface [12 mounting points [no double side used] x 512MB chips = 6GB]. Radeon 290X has 512mbit interface, which enables it to to up to 8GB of vRAM with no doble sided mounting also]. edit - I'm not sure if they have used 512MB chips, maybe they went with 256MB chips and doublesided arrays.
In new GPU series, AMD is moving away from GDDR5 and going straight to the new tech.
OminoMichelin
Member
Graphics cards (and the PS4) use GDDR5 which has been around for quite a few years now. Putting it simply, the reason why the PS4 and high end graphics cards have more memory is because the actual chips have a higher density. The PS4 uses currently 16 512MB chips, so 8GB in total. Recently production of 1GB GDDR5 chips started, so the PS4 will probably get a revision that will only have 8 of those to save costs.
HBM is just another memory standard that will slowly replace GDDR5, just like how GDDR5 replaced GDDR3 in the past. The main reasons why is that has a lot more bandwidth while lowering power consumption, space and price in the long run. HBM2 will increase the density and bandwidth even more oin the future.
Last rumours were pointing to 290X or higher levels of TDP and water cooling. However TDP is never a concern for enthusiast segment.
The linkedin profile mentioned 300w TDP. But as we all know TDP is not actual power consumption.
A 290x at stock didn't consume 290w and neither did some overclocked models :
http://www.techpowerup.com/reviews/Sapphire/R9_290X_Tri-X_OC/22.html
Does the 7970 really benefit from all that memory? Those versions of the card solely exist because it was the previous flagship card, so they had to make allowances for those wanting to use high resolution displays.
No it doesn't. And the Nvidia cards at peak consume more than their specified specs (AMD and Nvidia also measure TDP differently). Notably some of the non-reference cards manage to keep power down at OC because they also cool the gpu better. Running silicon at lower temperatures reduces its leakage power consumption.
The issue here is that many benchmarks do full system power consumption measurements (such as AT). They're not entirely wrong for doing it as Nvidia cards do have lower overheads on the rest of the system, even if they aren't as efficient as stated (or implied) on paper.
I don't really know how Nvidia calculate their TDP but it seems to vary from generation to generation. Maxwell cards have impressive perf/watt still, but their advertised TDP are sometimes exceeded.
Like this custom 970 (MSI) which is 20w above standard 970 consumption :
http://www.guru3d.com/articles_pages/msi_geforce_gtx_970_gaming_review,7.html
Yeah it is impressive, but people often quote the TDP figures as relative comparisons when it isn't all that appropriate at giving an accurate figure. You could measure a realistic figure with full system tests only if you could isolate the AMD overheads.
You could only do real measurements if you can tap into the PCI-E and Mobo power rails (or before the Graphics card VRM).