[BitsAndChips]390X ready for launch - AMD ironing out drivers - Computex launch

[twjr]

I'm going to add another couple of cents to the 4gb vs 8gb discussion. I think too many people here are arguing from rational objective viewpoints. I don't doubt that a 4gb 390X card could be successful in all areas bar having more than 4gb of ram. But if nvidia releases a card between the Titan X and 980 with more than 4gb, that is going to be the only metric that matters. It won't matter if the 390x has the best perf/$ or outright performance or even perf/watt so long as nvidia still has the best marketing. As we have seen since the release of the 680 so long as you have the best marketing and something to leverage you will sell cards. As objective buyers some of us can see the vale in AMD's offerings but we are not who makes money for companies.

 

[Serandur]

3GB vs. 4GB for 780 vs. 290 wasn't even the crux of the purchase here. R9 290 OC was faster than 780 OC for $100 less. That made 780 a worse buy regardless of VRAM. In a situation where a card that's as fast or faster had more VRAM and cost less, you should have never purchased the 780 to start with. It's disappointing that some people actually gave you such horrible advice. On our forum, most people would have recommended you to get an after-market 290 when 780 was $500+. When gamers with those cards compared them stock vs. stock or OC vs. OC, 780 couldn't win which made its $100 premium largely a marketing/brand premium tax.

I made the purchase during the whole mining craze when 290 prices were still somewhat inflated and the model selection limited. Back then, Nvidia was the only choice for reliable downsampling and I got a great B1 GK110 780GHz for $510 (only $60 more than a Tri-X 290 with a plastic and aesthetically unappealing design imo and no backplate relative to the 780 GHz).

My 780 boosted to 1150 MHz right out of the box and I genuinely loved the card as a whole package; in benchmarks, it was doing great against even 290Xs at the time at its stock speeds. I originally intended on getting a 290, but market conditions made it quite a tough choice at the time.

This is a very important point you bring up against future-proofing with any current gen. Whether it's AMD or NV, even if they optimize the last gen architecture as best as possible, their last gen cards have inherent architectural weaknesses (tessellation in 5850, 6950, 7970 or R9 290 or compute in Fermi, Kepler). That means, once newer gen games start using some of those graphical features more extensively, no amount of optimization will save R9 280X/7970Ghz from tanking when tessellated God Rays are enabled in FC4. That's more reason to not keep any gen beyond 2-3 years. Just dump it and get something faster.

See, the real problem I had was that it wasn't even a year that I had the 780 before I started getting the first glimpse of the limitations to come. Watch Dogs shook my confidence in the card rather soon after. I recall horrible GCN performance in the game at the time. Gameworks and Nvidia/AMD's asynchronous development schedules really muddy up the whole ability to make proper comparisons. At the time, I was given little reason to doubt the 780; overclocked, it was doing really well in the benchmarks and matched the Titan Black at stock. Who knows how dreadfully Maxwell will age, so much of its efficiency comes from what seems like trickery.

All great points. If you know for sure you play a lot of modded titles and games where more VRAM is 100% going to benefit you, get the card with more VRAM since you will enjoy that benefit right away, not in 2-3 years. Also, if you can easily afford a $700-800 card over the cheaper $400-500 mid-range cards, enjoy it, just don't keep it too long as its resale value will bomb (just look at GTX480/580/780TI or HD7970Ghz/R9 290X)! :thumbsup:

Resale value hits are generally okay by me. I sold my 780 early enough that people were still giving me good offers for it even after Maxwell launched. It might be different this time around with a new process node and all, but a couple hundred dollars lost per year seems like a fair deal to have the card I really want at the time that I want it. I flat-out draw the line above something in the ~$700s though and even then only buy there if there's a really nice AIB model I like.

There is no hypocrisy if you read the posts more carefully. Since you are new to our forums, you should get used to some people putting words in other people's mouths. There is good info on our forums but you just gotta learn how to compare diffeerent viewpoints and make your own decision. Of course if R9 390X is 4GB costs roughly the same as a GM200 6GB and they perform similar, pick the 6GB card. Pretty much no one in this thread disagrees with that comparison.

That would be a refreshing change of pace from other forums then where even a brand-agnostic enthusiast can't speak his mind without feeling ostracized by Team Red or Team Green depending on what's being criticized. I've seen a lot, the type of devotion people have to the brands they buy or even the assumptions it seems are being made depending on what's being criticized leaves me wary. Case in point, getting painted, even implicitly, as someone trashing a product because of brand loyalties when it's simply me speaking my mind on what I want as a consumer is what I'm addressing. It's the same anonymous, snarky, drive-by "but oh, when and so and so does it, now it's a problem" assumptions I see elsewhere when the truth is I think it's a problem exactly because I always expected company A to maintain a certain advantage over company B because I want company A's product, not because I'm trying to hype up company B and trash company A.

We really need to see where R9 390 nonX, 390X and various GM200 versions of cards end up in price/performance though before making an assessment on what's actually worth $500-800. Right now everyone on this forum is guessing at where R9 390 series will end up in performance as there is no data from AMD on R9 300 series and no one is confident to say that any one leak is credible enough.

I do not disagree, but my personal current goal is to get something simply as overpowered for 1440p as my 780 felt for 1080p with enough VRAM to carry it (980 and 290X kinda struggle a little with the latest games, something ~40-50% more powerful per clock would be just perfect for me). Normally, I would try and hold off until the new process node and stuff with knowledge of 28nm's impending obsolescence, however I just want the absolute fastest of what I feel are these long-overdue monsters for all the awesome games coming out this year and early next year before Pascal and Arctic Islands.

The Witcher 3 is 8 days from now, I'm a little sad the big daddy chips aren't quite here yet. Arkham Knight is next month, Project Cars just released, Just Cause 3 is coming, Total War: Attila looks like a GPU killer, the new AC will be a nice tech demo I'm sure, and I'm doing my big Skylake overhaul this year; all kinds of great and upcoming stuff. I have a good feeling about Fallout 4 upcoming relatively soon as well, sometimes in early-ish 2016. The timing factor is sometimes a personally good excuse to splurge a little more than usual I feel, but though they're all inevitably doomed regardless, I was really looking forward to AMD beating out Nvidia this year outright; more VRAM like always and a balls-to-the-wall Fiji XT monster that does HBM's first appearance justice. I mean, in the end, $500 is itself quite a bit of money too. I maintain that despite video card pricing's massive hyperinflation, I should be completely happy with something I'm buying in that price range and am willing to spend a couple hundred more from there to get it. That's the reason I'm looking at $500+ cards in the first place, I remember when that price made them the best of the best and I still kind of expect it.

 

[boozzer]

I'm going to add another couple of cents to the 4gb vs 8gb discussion. I think too many people here are arguing from rational objective viewpoints. I don't doubt that a 4gb 390X card could be successful in all areas bar having more than 4gb of ram. But if nvidia releases a card between the Titan X and 980 with more than 4gb, that is going to be the only metric that matters. It won't matter if the 390x has the best perf/$ or outright performance or even perf/watt so long as nvidia still has the best marketing. As we have seen since the release of the 680 so long as you have the best marketing and something to leverage you will sell cards. As objective buyers some of us can see the vale in AMD's offerings but we are not who makes money for companies.

marketing has already started, it has already latched onto the 4gb ram

 

[2is]

I'm going to add another couple of cents to the 4gb vs 8gb discussion. I think too many people here are arguing from rational objective viewpoints. I don't doubt that a 4gb 390X card could be successful in all areas bar having more than 4gb of ram. But if nvidia releases a card between the Titan X and 980 with more than 4gb, that is going to be the only metric that matters. It won't matter if the 390x has the best perf/$ or outright performance or even perf/watt so long as nvidia still has the best marketing. As we have seen since the release of the 680 so long as you have the best marketing and something to leverage you will sell cards. As objective buyers some of us can see the vale in AMD's offerings but we are not who makes money for companies.

I'm SO glad you mentioned this...

The argument you're making now is the same one that could have been made for the 680/7970. When it (680) was released it was faster and cheaper than the 7970, but with less RAM. But it didn't maintain that performance lead, especially now with games using more than 2GB of VRAM even at 1080p

That's basically what you're saying about the 390x... It might be faster and cheaper but with less RAM than the 980Ti

Yet for some strange reason, the AMD fans are all in agreement that anyone who bought a 2GB 680 at the time probably made the wrong choice, however, that EXACT same scenario playing out now, only in reverse, and they don't see a problem.

Oh how the goal posts shift...

 

[twjr]

I'm SO glad you mentioned this...

The argument you're making now is the same one that could have been made for the 680/7970. When it (680) was released it was faster and cheaper than the 7970, but with less RAM. But it didn't maintain that performance lead, especially now with games using more than 2GB of VRAM even at 1080p

That's basically what you're saying about the 390x... It might be faster and cheaper but with less RAM than the 980Ti

Yet for some strange reason, the AMD fans are all in agreement that anyone who bought a 2GB 680 at the time probably made the wrong choice, however, that EXACT same scenario playing out now, only in reverse, and they don't see a problem.

Oh how the goal posts shift...

I did buy a 670 then because it wasn't going to have any problems with only 2gb. It didn't go the distance though and I regretted the purchase over a cheaper 7950 until I got a 290. To be honest though the biggest problem I had was with heavily modded Skyrim and it could still cause my 290 to struggle. Even a couple of years ago I wouldn't have minded more than 4gb.

I do think a lot of the performance decline in the 670/680 is to do with drivers no longer being optimised for Kepler in addition to having less ram.

Also don't mistake my argument for what I would personally purchase if I were in the position to be building a new computer (hard to take a desktop with you when living out of a backpack). Right now I'd really need to see the premium for an 390x 8gb over 4gb, especially over a 390.

 

[raghu78]

I made the purchase during the whole mining craze when 290 prices were still somewhat inflated and the model selection limited. Back then, Nvidia was the only choice for reliable downsampling and I got a great B1 GK110 780GHz for $510 (only $60 more than a Tri-X 290 with a plastic and aesthetically unappealing design imo and no backplate relative to the 780 GHz).

My 780 boosted to 1150 MHz right out of the box and I genuinely loved the card as a whole package; in benchmarks, it was doing great against even 290Xs at the time at its stock speeds. I originally intended on getting a 290, but market conditions made it quite a tough choice at the time.

very true. the resellers like newegg , amazon were making a killing on the mining craze and gamers were the ones who lost out on good price/perf.

See, the real problem I had was that it wasn't even a year that I had the 780 before I started getting the first glimpse of the limitations to come. Watch Dogs shook my confidence in the card rather soon after. I recall horrible GCN performance in the game at the time. Gameworks and Nvidia/AMD's asynchronous development schedules really muddy up the whole ability to make proper comparisons. At the time, I was given little reason to doubt the 780; overclocked, it was doing really well in the benchmarks and matched the Titan Black at stock. Who knows how dreadfully Maxwell will age, so much of its efficiency comes from what seems like trickery.

GTX 980 looks a good card now just like GTX 680 looked a good card way back in 2012. Both are efficient and do a good job. But we saw the Kepler based GTX 780 and GTX 770 get clearly exposed in some of the newer titles like FC4.

http://www.hardocp.com/article/2015/01/07/far_cry_4_video_card_performance_review/7#.VVF_Gfmqqko

I feel Nvidia's architectures employ some clever design ideas or tricks which can hide some of the differences in raw hardware and architectural capability but sooner or later those ideas/tricks cannot hide the differences in architectural/hardware capability as more demanding games are released. I feel the same situation will repeat even more emphatically in the case of R9 390X vs Titan-X with the architectural and raw hardware capabilities (high bandwidth) of R9 390X making a significant difference as DX12 launches.

I do not disagree, but my personal current goal is to get something simply as overpowered for 1440p as my 780 felt for 1080p with enough VRAM to carry it (980 and 290X kinda struggle a little with the latest games, something ~40-50% more powerful per clock would be just perfect for me). Normally, I would try and hold off until the new process node and stuff with knowledge of 28nm's impending obsolescence, however I just want the absolute fastest of what I feel are these long-overdue monsters for all the awesome games coming out this year and early next year before Pascal and Arctic Islands.

The Witcher 3 is 8 days from now, I'm a little sad the big daddy chips aren't quite here yet. Arkham Knight is next month, Project Cars just released, Just Cause 3 is coming, Total War: Attila looks like a GPU killer, the new AC will be a nice tech demo I'm sure, and I'm doing my big Skylake overhaul this year; all kinds of great and upcoming stuff. I have a good feeling about Fallout 4 upcoming relatively soon as well, sometimes in early-ish 2016. The timing factor is sometimes a personally good excuse to splurge a little more than usual I feel, but though they're all inevitably doomed regardless, I was really looking forward to AMD beating out Nvidia this year outright; more VRAM like always and a balls-to-the-wall Fiji XT monster that does HBM's first appearance justice. I mean, in the end, $500 is itself quite a bit of money too. I maintain that despite video card pricing's massive hyperinflation, I should be completely happy with something I'm buying in that price range and am willing to spend a couple hundred more from there to get it. That's the reason I'm looking at $500+ cards in the first place, I remember when that price made them the best of the best and I still kind of expect it.

I am sure you will love what AMD has done with this monster GPU. I expect the R9 390X and R9 390 to only ship in 8 GB versions and the R9 380X and R9 380 to ship in 4 GB versions. There are two ways this could play out. AMD has only one high end chip (520 - 550 sq mm) which has a lot of salvage SKUs to power both R9 390 and R9 380 series or they have two chips with HBM - the big chip closer to 550 sq mm for R9 390 series and the smaller one around 400 sq mm for R9 380 series. In roughly over a month we will finally get to see the cards.

 

[RussianSensation]

I'm not going to quote your entire novel. My mistake was certainly not 680's over 670's, as I'd have the same regret if I owned 670's. Cost wasn't a factor, i'm not regretting that I spend more for 680's over 670's. I regret not waiting until 4GB 680's became available. Your attempt at mind reading has failed you here. Back then the mistake was getting a high end card with 2GB of ram. In 2015 the mistake is going to be getting a high end card with 4GB of ram. It's not going to happen with this consumers money.

I think you still missed my point though. Let's say 680s cost you $1K and you say cost wasn't a factor. It would have been better to buy $800 670s, sell those and get 970s or 980s, etc. If cost isn't a factor, why aren't you upgrading every generation (680s -> 980s -> 980Tis, etc.)? I am guessing you paid extra for the 680s to future-proof? Doesn't look like that strategy worked out at all, while the 6-10% performance differences between those cards in SLI can only be measured in benchmarks not in a blind gaming test. See now you are pushing the same idea that flagship cards barely faster than 2nd tier (i.e., 15% faster) are going to be well worth their high premiums because they'll be more future-proof. They won't be. If VRAM bottlenecks won't make them obsolete, then GPU processing power will. GTX580's performance today is just 8% faster than an HD6970 but HD6970's 2GB of VRAM doesn't help it much either over GTX580's 1.5GB. Does anyone discuss today if 580 was more future-proof than a 6970? No, because I can buy a $240 R9 290 card 2X faster than the 580, so who cares how future-proof 580 was.

GTX470/570 OR 670/680 are also obsolete at the same time. Future-proofing with flagship cards over 2nd tier cards doesn't work because next gen games will mop the floor with them from pixel, shader, geometry, or VRAM side.

680 2GB never provided any real tangible playability over 670 2GBs when both setups were OCed. You can go right ahead and spend $1400 on dual GM200 6GB/R9 390X cards and keep them for 3-4+ years like you kept your 680s, while someone else will use this gen as a stop-gap, and upgrade again at 14nm. Guess what will happen? Over the next 1.5 years, your GM200 6GB SLI cards will be hardly faster than a card with 85% of their performance because software takes 1.5-2 years to catch up (just look at HardOCP's testing of Titan X SLI vs. 980 SLI at 1440P = completely CPU limited). When the software catches up, many people will have sold this gen's stop-gap cards and moved on to something way faster, more efficient and with more features. In the next 1.5-2 years of the useful life following the first 1.5-2 years, those GM200 6GB/R9 390X CF/SLI cards will be beaten by a single $500 card.

In that same time-frame, $1400 GM200 6GBs/R9 390X 8GB CF will lose 40-50%+ of their value (i.e, by July-August 2017 we will probably have a card for $700 as fast as both of those in SLI/CF). While 6GB of VRAM might save you for a bit longer, that Maxwell/R9 390X architecture will start bombing in newer gen games as the architectural weaknesses of those GPUs will become more and more exposed.

History will repeat itself, like it always does and your $1400 worth of $700 flagships "future-proofing theory" will come crumbling down like a deck of cards. How do I know, because it happened for the last 20 years like clock work. Pick any GPU generation you want. Also, once NV releases Pascal and stops optimizing Maxwell drivers, see ya!

If you need a history lesson in how this works:

$1400 GTX780Ti SLI November 2013
$660 GTX970 SLI with better performance Sept 2014. That only took 10 months.

This idea to "future-proof" with $700 flagships for games out in 3-4 years is not going to work. In 4.5 years we now have $89 R7 260X that is about as fast as a $500 GTX580.

 

[5150Joker]

I think you still missed my point. Let's say 680s cost you $1K and you say cost wasn't a factor. It would have been better to buy $800 670s, sell those and get 970s or 980s, etc. I am guessing you paid extra for the 680s to future-proof?

680 2GB never provided any real tangible playability over 670 2GBs when both setups were OCed. You can go right ahead and spend $1400 on dual GM200 6GB/R9 390X cards and keep them for 3+ years like you kept your 680s, while someone else will use this gen as a stop-gap, and upgrade again at 14nm. Guess what will happen? In the next 1.5 years, your GM200 6GB SLI cards will be hardly faster as software takes 1.5-2 years to catch up (just look at HardOCP's testing of Titan X SLI vs. 980 SLI at 1440P = completely CPU limited).

When the software catches up, many people will have sold this gen's stop-gap cards and moved on to something way faster, more efficient and with more features. While your $700 GM200 6GB card will be stuck with 336.6 GB/sec memory bandwidth, $350 mid-range 14nm card will have 600-700GB/sec.

In that same time-frame, $1400 GM200 6GBs/R9 390X 8GB CF will lose 40-50%+ of their value. While 6GB of VRAM might save you for a bit longer, that Maxwell/R9 390X architecture will start bombing in newer gen games as the architectural weaknesses of those GPUs will become more and more exposed.

History will repeat itself, like it always does and your $1400 worth of $700 flagships "future-proofing theory" will come crumbling down like a deck of cards. How do I know, because it happened for the last 20 years like clock work. Pick any GPU generation you want. Also, once NV releases Pascal and stops optimizing Maxwell drivers, see ya!

If you need a history lesson in how this works:

$1400 GTX780Ti SLI November 2013
$660 GTX970 SLI with better performance Sept 2014. That only took 10 months.

This idea to "future-proof" with $700 flagships for games out in 3-4 years is not going to work. In 4.5 years we now have $89 R7 260X that is about as fast as a $500 GTX580.

You act like what you're saying is some sort of revelation. Everyone knows technology depreciates, I don't see why you feel the need to write paragraphs on the subject.

 

[Silverforce11]

You act like what you're saying is some sort of revelation. Everyone knows technology depreciates, I don't see why you feel the need to write paragraphs on the subject.

His post is good for people seeking better value because it makes a lot of sense, but useless for people seeking the fastest at the time. You're the latter so that's why you don't see the point.

I'm inbetween, but I'm sure some here are more aligned towards the value segment.

 

[2is]

I think you still missed my point though.

I didn't miss your point I ignored it because you missed mine when you responded to my post and started talking about something completely different than what I was talking about.

But to address your point... I can buy a better card now, with or without selling my 680's, I just don't want to because the card I want does not yet exist. Me buying a 390x now that has 4GB of ram would be making the same mistake again, even if it saves me $200 over a 6GB 980Ti.

I prefer spending more less frequently then spending less more frequently. I've tried it both ways, I prefer it my way.

 

[Majcric]

What the above poster said. But I want to add I feel the top chip are supported better or have less issues in general. That to me is worth a premium.

 

[RussianSensation]

Oh how the goal posts shift...

There is no goal shifting, just you altering historical benchmarks. In the hands of enthusiasts, without exotic overclocking, 680 did not beat a 7970. Similarly, 680 lost to 7970Ghz on day 1. Your statements that 680 had less VRAM but was the faster card are not correct which is why the comparison isn't even valid when you say some people advocate for a slightly faster R9 390X with less VRAM vs. GM200. That is not the case.

With the 7970Ghz, one got a card that was faster, had more VRAM and cost less, plus HD7970Ghz had 2-3 AAA game bundles! Do we seriously need to go back in time to prove this:

(1) HD7970Ghz > even an after-market 680 in frame times on launch date. Even if you put a highly overclocked 680 against a stock 7970Ghz, the 680 lost. That's as good as it gets for NV comparing a reference 7970Ghz against a factory pre-overclocked 680.

The key trade-offs were power usage and lack of NV specific features like PhysX, TXAA etc.

(2) 680 OC vs. 7970 OC.
Battle of the Elite: Asus GeForce GTX 680 DirectCU II TOP vs. MSI R7970 Lightning
[/B]
A 1293 mhz Boosted GTX680 couldn't even beat an 1165mhz HD7970 Lightning, nevermind 1250-1275mhz 7970 card.

(3) High resolution gaming - 7970Ghz also won that.

This is one of the things NV fans never acknowledge. GCN 7970 was faster than 680 not in 15-20 months, but just 3 months from 680's launch.

These wins were never acknowledged though ..... and it took a whopping 2-3 years until people woke up to find out that OH, it is true HD7970Ghz did age better but that's because it was also faster from day 1.

 

[2is]

7970GHz Edition did not exist when the 680 was released. It came after. Nice try, but you are shifting goal posts.

 

[Majcric]

But the cards are so close in performance it's not worth talking about. Excluding Sli and xfire id like to see more charts where that extra 1gb of VRAM helped the 7970. Thus far, I have seen very few and most have upgraded and moved on.

 

[digitaldurandal]

I'm SO glad you mentioned this...

The argument you're making now is the same one that could have been made for the 680/7970. When it (680) was released it was faster and cheaper than the 7970, but with less RAM. But it didn't maintain that performance lead, especially now with games using more than 2GB of VRAM even at 1080p

That's basically what you're saying about the 390x... It might be faster and cheaper but with less RAM than the 980Ti

Yet for some strange reason, the AMD fans are all in agreement that anyone who bought a 2GB 680 at the time probably made the wrong choice, however, that EXACT same scenario playing out now, only in reverse, and they don't see a problem.

Oh how the goal posts shift...

I disagree though. I bought 2x 670s and only recently replaced them. In 95% of the situations where VRAM would matter I was limited by the GPU anyways (running 5760x1200.) You can see this to be the case even with the 680s that had models with double the VRAM. There was no performance difference in nearly any game even at the highest settings.

7950 may have been a better buy, sure, but it was not really due to VRAM.

I do not know what the performance of the 390X will be at this point obviously, but the question I have to ask is: will the VRAM actually limit the card?

We have seen that 980SLI continues to improve even at 4k resolutions with MSAA and has been unable to hit a wall due to VRAM. Even 970SLI has been for the most part good and barely hindered by the 3.5+0.5 GB frankenstein configuration.

This points to there still being a GPU wall before VRAM becomes a question. Which has been what I have argued with people about on these forums for some time.

I think that the mistake of not having 4GB of VRAM will be purely marketing. The same reason low end cards had 4GB even when the higher end cards did not.. because the consumer doesn't know better.

Will games come that use more than 4GB of VRAM surely, will games come that need more than 4GB and do so before coming up against a GPU wall? Unlikely, if two 980s don't hit that VRAM wall before becoming GPU limited 1 390X likely will not either.

Again, I do think AMD is making a mistake if they only release in 4GB, just not necessarily for any reason outside of marketing. We will know soon enough.

 

[2is]

I disagree though. I bought 2x 670s and only recently replaced them. In 95% of the situations where VRAM would matter I was limited by the GPU anyways (running 5760x1200.) You can see this to be the case even with the 680s that had models with double the VRAM. There was no performance difference in nearly any game even at the highest settings.

7950 may have been a better buy, sure, but it was not really due to VRAM.

I do not know what the performance of the 390X will be at this point obviously, but the question I have to ask is: will the VRAM actually limit the card?

We have seen that 980SLI continues to improve even at 4k resolutions with MSAA and has been unable to hit a wall due to VRAM. Even 970SLI has been for the most part good and barely hindered by the 3.5+0.5 GB frankenstein configuration.

This points to there still being a GPU wall before VRAM becomes a question. Which has been what I have argued with people about on these forums for some time.

I think that the mistake of not having 4GB of VRAM will be purely marketing. The same reason low end cards had 4GB even when the higher end cards did not.. because the consumer doesn't know better.

Will games come that use more than 4GB of VRAM surely, will games come that need more than 4GB and do so before coming up against a GPU wall? Unlikely, if two 980s don't hit that VRAM wall before becoming GPU limited 1 390X likely will not either.

Again, I do think AMD is making a mistake if they only release in 4GB, just not necessarily for any reason outside of marketing. We will know soon enough.

Except I'm running SLI. in a single card setup I'd agree that you will run out of GPU grunt on a 670 or 680 before VRAM becomes an issue.

 

[RussianSensation]

But the cards are so close in performance it's not worth talking about. Excluding Sli and xfire id like to see more charts where that extra 1gb of VRAM helped the 7970. Thus far, I have seen very few and most have upgraded and moved on.

There are very few games where 3GB of VRAM allows 7970Ghz to provide superior gameplay over 680 due to VRAM.

Most of the time, the reason 7970Ghz wins is because it's just a faster GPU to begin with.

I think some members on this forum are only looking at 4GB vs. 8GB of VRAM and ignoring the big elephant in the room - R9 390 nonX OCed.

Why does this matter?

Let's look at history:

1. HD5850 OC ~ 97-98% as good as 5870 OC

Proof
"The default Radeon HD 5850 is an average 12.7% slower than the Radeon HD 5870 in the low-quality modes and 14.1% slower in the high-quality modes. When the frequencies of the Radeon HD 5850 are increased to the level of the HD 5870 (850/4800MHz), the gap is smaller at 2% and 2.6%"
http://www.xbitlabs.com/articles/graphics/display/radeon-hd5850_10.html#sect0

2. HD6950 unlocked+OC = 100% as good as 6970 OC (No need for benches as the card is identical).

3. HD7950 OC ~ 95-98% as good as HD7970 OC

Proof:

http://www.legionhardware.com/artic...z_edition_7950_iceq_xsup2_boost_clock,13.html

2nd Proof:

"Our look at the reference Radeon HD 7950 and HD 7970 cards shows there to be a 15 per cent gap in performance in favour of the range-topping GPU. But raise the HD 7950's clocks to the higher speeds of the 7970 and the gap melts away to around five percent, often a little less, depending upon gaming title."
http://hexus.net/tech/reviews/graphics/34761-amd-hd-7950-vs-hd-7970-clocks/?page=10

4. R9 290 OC ~ 95% as fast as R9 290X OC
Proof

Back to what I am saying -- what if we have $500 R9 390 4GB vs. $700 R9 390X 8GB and the latter is only 5-7% faster when both are max OCed? Last 4 consecutive generations of AMD cards, the 2nd tier card from the top was stupid fast when overclocked but cost way less.

Is R9 390 4GB a fail then if it's the next 5850/6950/7950/R9 290?

All this hype around R9 390X and no one except me is excited for R9 390 nonX? Really?

 

[2is]

A 5850 is largely the same as a 5870 with the exception of clock speeds (both RAM and GPU) and stream processors. You're unlikely to find a scenario where one provides poor gameplay while the other doesn't, it's just that the 5870 will have a few more FPS tacked onto it.

You can't have that same level of guarantee between cards that not only have different clock speeds, but one has half the ram as the other. Apples to oranges

Not to mention, no one is complaining about an 8GB 390x... The complaint is whether or not 8GB will be offered. Hype usually surrounds the high end. You have over 16k posts in these forums, you should know this.

 

[Serandur]

I'm just going to leave this little piece of personal evidence about why crossfire R9 390s with 4 GBs of VRAM would be a bad idea (from my point of view). Here are some charts showing VRAM usage, GPU usage, FPS, and frametimes on SLI 970s with FPS capped to 30 FPS (to completely eliminate CPU limitation issues; it's a 3770K, it can handle Unity at well above 30 FPS). I've got Unity set to max settings at 1920x1080 with 8xMSAA. I just went with that configuration just because (as severe as possible while demonstrating the limitations even at 1920x1080), 2560x1440 would have yielded similar results with MSAA though it was much milder than this with just 2xMSAA at 2560x1440 but still present even with a 60 FPS target. 970s may complicate the issue with the 3.5 GBs thing, but unlike most scenarios Unity apparently needed the VRAM enough that the 970s actually used over 3.5 and went to the full 4 GB point.

Notice the almost consistently less than 99% GPU usage, the full 4 GB usage, the FPS only minutely fluctuating (meaning average FPS is not affected too much, a demonstration of why relying on FPS benchmarks to show VRAM issues isn't always that meticulous) while the simultaneous frametime recording graph revealed something far more serious going on (those spikes are way above what their equivalent FPS should dictate). Those are all stutters from a lack of VRAM, just me running around a little in Paris and turning the camera. Not smooth... at all. R9 390s are likely to be in the realm of 60-70% faster, will scale better in multi-GPU too, and the 970s were not overclocked, but that VRAM...

 

[3DVagabond]

Probably not the best card to use when trying to show any kind of typical VRAM results. No offense, but the 970's memory system is an abortion to begin with.

 

[Serandur]

Probably not the best card to use when trying to show any kind of typical VRAM results. No offense, but the 970's memory system is an abortion to begin with.

I am aware and do not take offense to it. They're long gone because of it. However, while not conclusive, the graphs and similar conditions I've gotten in other games or settings do suggest a couple things.

One, having owned two 970s myself, I assure you they do not ever cross the ~3.5 GB mark purely for useless allocation, ever. That's how I noticed something was originally up with the cards. So, if they are actually allocating into that last 512 MB portion, it's a fairly sure thing that they, at minimum, actually need 3.5 GBs of VRAM at that given time. Coincidentally, when I put an 8 GB Vapor-X 290X to the test, with no MSAA at 1440p and the settings I ran my 970s at, it didn't cross over the high 3000s in MBs, but the 2xMSAA added on top which caused my 970s to have an inconsistent hitching and to spill over the 3.5 GBs barrier pushed the 290X into the ~4500 MBs range.

Skyrim and Shadow of Mordor were much worse because even the 970s have enough grunt to do 60 FPS with crazy resolutions and mods on the former and max settings (and downsampling) on the latter, but they similarly ran out of VRAM and had inconsistent hitching (like, every time I swung a sword or turned the camera bad). SoM had a much less severe but similar problem on a single 980 which is also more than enough for 60 FPS on the game. 8 GB 290X was literally perfect.

Second point is that, at the very least, it's clear that FPS does not truly reflect VRAM deficiencies unless of the extremely severe variety. I'm pretty much saying a lot of the benchmarks people rely on kind of mask the truth a little bit, it's frametime spikes that are the real metric to test for this except, again, in severe scenarios where you see VRAM-starved cards literally tanking to unplayable FPS.

 

[raghu78]

Back to what I am saying -- what if we have $500 R9 390 4GB vs. $700 R9 390X 8GB and the latter is only 5-7% faster when both are max OCed? Last 4 consecutive generations of AMD cards, the 2nd tier card from the top was stupid fast when overclocked but cost way less.

Is R9 390 4GB a fail then if it's the next 5850/6950/7950/R9 290?

All this hype around R9 390X and no one except me is excited for R9 390 nonX? Really?

I expect both the R9 390X and R9 390 to sport 8 GB HBM. I also expect R9 380X and R9 380 to sport 4 GB HBM. After having pushed higher VRAM for 3 generations (HD 6970 2GB, HD 7970 3GB, R9 290X 4 GB) I do not think AMD will regress on that tradition.

People buying R9 390 and R9 390X will basically be buying it for 1440p and 4K gaming. My expectation is these cards will be as good or even better than Titan-X in raw GPU performance. So 4GB is a handicap especially at 1440p and 4K as we already see games use more than 4GB at 1440p

http://www.hardocp.com/article/2015...gtx_titan_x_video_card_review/14#.VVGuJPmqqko

Watch Dogs, Dying Light, COD AW (with SSAA) are a few examples. Anyway we will see within a month if AMD has stood true to its legacy / heritage.

 

[Vesku]

Skyrim and Shadow of Mordor were much worse because even the 970s have enough grunt to do 60 FPS with crazy resolutions and mods on the former and max settings (and downsampling) on the latter, but they similarly ran out of VRAM and had inconsistent hitching (like, every time I swung a sword or turned the camera bad). SoM had a much less severe but similar problem on a single 980 which is also more than enough for 60 FPS on the game. 8 GB 290X was literally perfect.

But due to the 970s memory bandwidth issues over 3.5GB, bandwidth will drop considerably and that's assuming it is actually using that extra 512MB partition and not just faking it, even 4GB 290s and 290Xs provide a better experience in those currently rare scenarios not just the 8GB versions. It's not as if the 980 or a potential 4GB 390(x) will have many issues with RAM in a year for the vast majority of gamers, 4K was only recently estimated at 1% of market.

 

[flopper]

People buying R9 390 and R9 390X will basically be buying it for 1440p and 4K gaming. My expectation is these cards will be as good or even better than Titan-X in raw GPU performance. So 4GB is a handicap especially at 1440p and 4K as we already see games use more than 4GB at 1440p

.

8gb 390x anything else wont make sense.
it be the better option for 4K gaming for any serious gamer.

 

[3DVagabond]

8gb 390x anything else wont make sense.
it be the better option for 4K gaming for any serious gamer.

No doubt it's the better option, if they have it. Dual link interposer tech might not be viable yet.