R9 300 cards listed in new driver - R9 370 is a rebrand

[3DVagabond]

Rumor are suggesting that R9 390X is just 20% faster GTX 980 than let alone GTX Titan X.

One rumor.

 

[tviceman]

R9 285 1080p and 2160p index beats the R9 280X :

http://www.tomshardware.com/charts/...tml?prod[7297]=on&prod[7279]=on&prod[7276]=on

Toms has got to be hands down the worst site to ever quote. For anything.

Techpowerup tests a huge range of games (19), and 280x is ~15% faster at both 1440p and 4k.

Tonga is a dud. Spin it however you want. It has not been proven to be faster overall than Tahiti (except when g0d RAYS!!! OMG) and it has less vram standard. It's the same size as Tahiti, it has more transistors, it did nothing for the market. It's a dud, dud, dud and if Hawaii has the performance of the supposed AMD leaks from their February uber secret metting, then Tonga as at best nearly 2x slower than Fiji at it's best. All of my points remain. Hawaii cannot go away unless Fiji yields like crap or Fiji's uber performance rumors prove false.

 

[RussianSensation]

As much as I'd like to see AMD do well with something they release, what you said is not true at all. In June it will be what? 9-10 months since the 980 came out. If by that time they only manage a 20% improvement, no matter the price, I call it a failure.

Interesting perspective. GTX980 only beats R9 290 by 20-25% or so and R9 290 has been on sale for $399 10 months before 980 came out for $550. 980 sold like hot cakes with these metrics but you suggest that a 20% faster R9 390 at say $500-550 against a 980 is a failure? What if AMD brings out a card with 980's performance for $399, that's a failure too?

I am personally far far more excited now for 14nm given the pricing of 980/Titan X and the delays of R9 390 cards, but people seem to overinflated how great 970/980 were when they dropped. After-market R9 290 like the Sapphire Tri-X were selling for $350-370 before 970/980 came out. Just because most reviewers/gamers didn't pay attention to those prices, doesn't mean it wasn't the reality.

So the range is pretty wild out there in rumor-land. Anything from 20% above 980 to 50% and demolishing Titan X. It's fun to speculate and that's about all it should be.

I don't think so. Titan X is nearly 50% faster than an R9 290X and has 20%+ overclocking headroom. For R9 390X to "demolish" a Titan X, it would need to be 20-25% faster. To me 5-10% faster is not demolishing.

That means if R9 390X is 65% faster than an R9 290X it's barely 10% faster than a stock Titan X. Honestly, not sure how we got to this point but a lot of people on this forum seems to just keep ignoring price altogether since 980 came out. Sure, Titan X overclocked brings awesome performance, but at $1K price, who cares for 99% of the market. If we get consumer GM200 6GB/390 non-X/390X with 87-90% of Titan X's performance at $499-600, the Titan X is all but irrelevant for 99% of gamers like the original Titan became less than 9 months from its launch once R9 290/290X launched.

Alternatively, if I had to choose between SLI/CF for $1100 that offered me 87-90% of Titan X's performance when CF/SLI doesn't work, but 50-60% more performance when SLI/CF works, I'd take the dual card setup all day, every day. Since many gamers are not loaded, for A LOT of people "demolishing" the Titan means owning it in price/performance by miles, rather than outright beating in in performance. Would you say R9 290 at $399 demolished a $650 GTX780 in performance? No, but as a buying choice, it certainly did for brand agnostic PC gamers. Literally overnight a $650 780 became irrelevant. R9 390 nonX/X will not demolish a Titan X in performance, but it's almost a guarantee that they will supersede Titan X as the smarter buying choices (and of course can't discount consumer GM200 6GB cards).

 

[Silverforce11]

I don't think so. Titan X is nearly 50% faster than an R9 290X and has 20%+ overclocking headroom. For R9 390X to "demolish" a Titan X, it would need to be 20-25% faster. To me 5-10% faster is not demolishing.

There's developers & journalist who claim the leaks prior of 50% faster than R290X is a lowball figure and that its a "lot faster".

So as I said, lots of figures being thrown around.

 

[RussianSensation]

There's developers & journalist who claim the leaks prior of 50% faster than R290X is a lowball figure and that its a "lot faster".

So as I said, lots of figures being thrown around.

Right but you have to take all these leaks with skepticism. The hype behind R9 390X is going to set the card up for a failure. Prior to all these leaks in the last 2 months, most people would have been satisfied if R9 390X came in at sub-$650 and matched NV's top card - i.e., the Titan X. By the time R9 390X launches, people are going to expect 50% faster than a 980 for $499. I feel like there is a large double standard about what NV needs to do and what AMD needs to do to impress gamers. 970 barely undercut after-market R9 290s and came out nearly 1 year later and the media went nuts....all because it had trivial HDMI 2.0 for 1% of 4K TV owners and superior perf/watt. All in all, 970/980 hardly changed the absolute performance level from a $350 R9 290. You need to overclock 980 to 1.45Ghz+ to really make it a worthwhile purchase over a stock R9 290 imo for 1440p+, but when 970/980 came out, R9 290 was nearly 1 years old. Yet, the media and gamers went crazy over what I will always consider a pair of some of the most overhyped cards of all time. Even the 960 is hardly getting the deserved flack it deserves for bringing soooo little over the ancient GTX760.

Look, today a $530-550 980 is just 8-15% faster than an $290-300 R9 290X, and you need a max overclocked the 980 to beat a max overclocked R9 290X by 25%, but yet we read how some gamers would find it a total failure if a $500 R9 390 is only 15-20% faster than a 980. I guess everyone on AT now is a 1% where price/performance is irrelevant. Amazing how perf/watt marketing is completely taking over.

Naturally many gamers want to compare a Titan X to R9 390X but I am sorry a $500-650 GPU does not in any way compete with a $1000 one. Who cross-shops a $500-650 GPU with a $1000 one? If we are going to make such a comparison, the ONLY fair comparison is dual $500-650 cards against $1000 Titan X or R9 390X vs. consumer GM200 at $500-700. Think about it, if going dual cards, it'll be R9 390 nonXs for ~$1000 against dual Titan Xs for $2000. Would you call it a failure if R9 390 non-X is 10-15% slower than the Titan X for half the price? Last time I checked someone buying a $25-30K car doesn't cross shop a $50K car.

Gamers who are patient but aren't top 1% income earners/benchmarkers/early adopters aren't going to fall for the early adopter Titan X hype because of that $1K price tag. I am telling you right now once R9 390 nonX/X/consumer GM200 6GB drop, except for Titan X owners/those who need SP CUDA performance, and top 1%s, no one will care about the Titan X. In 1.5 years, we will see Pascal on 14/16nm and Titan X matching/beating performance in a $399-499 card. Those who can easily afford the Titan X(s) do not need to even wait for R9 390X. If someone can afford the latest and greatest all the time, then he/she can easily sell the Titan X(s) and pick up R9 390X(s) whenever. For everyone else, price/performance relative to the Titan X absolutely does matter. In 1.5 years from the original Titan's launch, we could get 3x 970/290s for ~$1K. For that reason, only looking at Titan X as a performance barometer but ignoring its awful price/performance is missing the point of cards like R9 390 nonX/X/consumer GM200.

 

[raghu78]

This is my educated guess. AMD has an all new GPU stack. I say this because I do not see the reason to rebrand older GPUs when AMD already designed a completely new chip called Tonga which was more of a test bed chip. Why did they need to do that when Tahiti or R9 280 and R9 280X were already selling. Its clear that AMD wanted to test out the tesselation, ROP, memory bandwidth improvements before tweaking the core shader architecture for more IPC or perf/sp. AMD is cash strapped and more likely to be very judicious with how they bring improvements to their GPUs. They are not going to spend billions of dollars like Nvidia and make a huge leap in perf and efficiency as we see from Kepler to Maxwell. Thats why GCN is seeing incremental architectural improvements in steps than a single leap.

Further more all the speculation on the web about rebranded products is pure nonsense. Roy Taylor just confirmed that.

http://i.imgur.com/xV8CuIc.png

I see 3 new GPUs or ASICs

R9 390X - 4096 sp, 4 shader engines (4 x 1024 = 4096 sp), 4 tesselation engines, 4 raster engines, 8 GB HBM, 64 or 128 ROPs, designed for 4k gaming. 1/8 rate double precision for Radeon and 1/2 rate double precision performance for flagship Firepro. 500 - 550 sq mm die. TDP 260W.

USD 800 . 5 - 10% faster than Titan-X

R9 380X - 3072 sp, 4 shader engines (4 x 768 = 3072 sp), 4 tesselation engines, 4 raster engines , 4 GB HBM, 64 ROPs, cost and gaming performance optimized die. 1/16 double precision performance. 350 - 380 sq mm die. TDP 200w

USD 400. 10 - 15% faster than GTX 980

R9 370X - 1536 sp, 2 shader engines (768 x 2 = 1536), 2 tesselation engines, 2 raster engines, 2/4 GB GDDR5, 32 ROPs, 1/16 double precision performance, mid range gaming card. TDP 120- 130w. USD 200.

5% faster than GTX 960.

btw I am thinking these ASICs are codenamed Bermuda, Fiji and Trinidad.

I don't think so. Titan X is nearly 50% faster than an R9 290X and has 20%+ overclocking headroom. For R9 390X to "demolish" a Titan X, it would need to be 20-25% faster. To me 5-10% faster is not demolishing.

10% avg faster is no small achievement if especially it happens at only 10% higher power. btw if that happens at 20% lower cost its a homerun for R9 390X. 780 Ti was 8-10% faster than R9 290X at launch and it was USD 150 costlier but due to mining craze in reality there was not that much price diff. This time around a 10% faster product at a 20% lower price would force Nvidia to launch a 6GB fully enabled GM200 WITH MUCH HIGHER CLOCKS for USD 750-800. Remember the Titan-X was already throttling so its not going to be easy to improve perf without increasing temps, heat and noise. This would result in a unfavourable comparison against R9 390X in press reviews.

That means if R9 390X is 65% faster than an R9 290X it's barely 10% faster than a stock Titan X. Honestly, not sure how we got to this point but a lot of people on this forum seems to just keep ignoring price altogether since 980 came out. Sure, Titan X overclocked brings awesome performance, but at $1K price, who cares for 99% of the market. If we get consumer GM200 6GB/390 non-X/390X with 87-90% of Titan X's performance at $499-600, the Titan X is all but irrelevant for 99% of gamers like the original Titan became less than 9 months from its launch once R9 290/290X launched.

Alternatively, if I had to choose between SLI/CF for $1100 that offered me 87-90% of Titan X's performance when CF/SLI doesn't work, but 50-60% more performance when SLI/CF works, I'd take the dual card setup all day, every day. Since many gamers are not loaded, for A LOT of people "demolishing" the Titan means owning it in price/performance by miles, rather than outright beating in in performance. Would you say R9 290 at $399 demolished a $650 GTX780 in performance? No, but as a buying choice, it certainly did for brand agnostic PC gamers. Literally overnight a $650 780 became irrelevant. R9 390 nonX/X will not demolish a Titan X in performance, but it's almost a guarantee that they will supersede Titan X as the smarter buying choices (and of course can't discount consumer GM200 6GB cards).

Titan-X is 45-48% faster than R9 290X. If R9 390X is on avg 55% faster than R9 290X that would mean the R9 390 is fighting the Titan-X and at a price closer to USD 700. How is that not owning in price / perf ? So you need to wait it out for launch reviews before drawing conclusions.

 

[RussianSensation]

Titan-X is 45-48% faster than R9 290X. If R9 390X is on avg 55% faster than R9 290X that would mean the R9 390 is fighting the Titan-X and at a price closer to USD 700. How is that not owning in price / perf ?

I actually agree with you. It's just that this forum is already setting up the R9 390X card for failure if it doesn't match or beat the Titan, regardless of price. That to me is illogical. As I said already, anyone who can easily afford the best, can buy Titan X and then whatever else is the best card. It's expected that the Titan X owner is paying a large premium to own the best card for X number of months and if so, shouldn't at all care if R9 390X beats the Titan X on price/performance or if Titan X's resale value will drop like a rock in 12 months. This type of customer wouldn't even care if in 12 months there is a card for $399 with Titan X's performance. However, for the rest of PC gamers who aren't spending thousands on Titan Xs, R9 300 series as a whole has the potential to provide better price/performance than 960/970/980/Titan X. For that reason, for 99% of PC gamers who aren't NV shareholders/employees/AMD stock shorters, we would be way better off if R9 300 series is a good one.

 

[Vesku]

I wouldn't read too much into Roy's email, he's not specifically denying any particular claim nor should he be expected to.

 

[Head1985]

R9 390X - 4096 sp, 4 shader engines (4 x 1024 = 4096 sp), 4 tesselation engines, 4 raster engines, 8 GB HBM, 64 or 128 ROPs, designed for 4k gaming. 1/8 rate double precision for Radeon and 1/2 rate double precision performance for flagship Firepro. 500 - 550 sq mm die. TDP 260W.

USD 800 . 5 - 10% faster than Titan-X

R9 380X - 3072 sp, 4 shader engines (4 x 768 = 3072 sp), 4 tesselation engines, 4 raster engines , 4 GB HBM, 64 ROPs, cost and gaming performance optimized die. 1/16 double precision performance. 350 - 380 sq mm die. TDP 200w

USD 400. 10 - 15% faster than GTX 980

I think best case scenario:380x is new SKU 390x perform as should
390x-5-10%above TITANX in 4K and on par in 1920x1080/2560x1440.Its cpu limited
390-20% above GTX980 and 10% under TITANX
380x-5% above GTX980
380-5% above GTX970

Worst case scenario:380/380x are rebrand 290/290x
390x-underperforms and only 20% above GTX980 in 1080/1440P on par with TITANX in 4K
390- 5% above GTX980 in 1080/1440 and 15-20% faster in 4K
380x-rebrand 290x.Same/worse performance than GTX970
380-rebrand 290.Same performance as 290

 

[Cloudfire777]

The thing about the chart that was posted along with R9 390X specifications, is that it could be the typical corporate PR slide where they mislead people a little bit. Like Nvidia do with their geforce cards, or what we saw with Mantle slides etc.

Although it seems that R9 390X get 50-65% above R9 290X, they could have used an old driver on the R9 290X and a unreleased new one for R9 390X which would have given 290X also a great boost in the same games.
So in reality the two could be closer than what the graph suggests.


Which would make the +20% above GTX 980 which VR-Zone presented legit.
That AMD wait to release Fiji while GTX Titan X roam free on the market and taking potential market share away from AMD is not good business wise.
BitsAndChips posted that it was because of drivers earlier, which very well could be true. But it may not be because of the new HBM memory and getting a stable driver before pushing out the product. It could be because AMD are polishing the performance of the unreleased drivers to squeeze out as much as they can to make it look good against Titan X.

So because of HBM and the price of it driving up the cost of the R9 390X vs earlier AMD GPU launches with plain GDDR5, the price for 390X today and the performance with current untweaked drivers against Titan X, may not put AMD in a good light like they used to price/performance wise. The rumor said $700+ for R9 390X. Say a 20% slower 390X cost $750. It may not be such a insta "I will totally buy this instead of Titan X" for the folks that are in the lookout for a new GPU. Which have always been AMD`s joker in the GPU market.
So AMD may be holding back to work on performance to make it as appealing as possible against Titan X before they announce it.

 

[gamervivek]

The thing about the chart that was posted along with R9 390X specifications

The words below the chart are blurred, but you can easily make them out to be saying 'based on performance estimates'. So it seems a way off.

 

[Cloudfire777]

The words below the chart are blurred, but you can easily make them out to be saying 'based on performance estimates'. So it seems a way off.

Yeah estimates graphs and those synthetic graphs are usually made to make things better than they really are

 

[jpiniero]

If it's really 2x Full Tonga plus HBM you would expect it to be very close to Titan X.

 

[DearLord]

Personally I believe that a key component in how this all plays out depends on the ram situation with, particularly, the 390x but also the 390 (assuming they're both released). Greedy and lazy game firms have gotten in the habit of dumping the entire ram, or as much as possible, of games into the graphics ram instead of splitting between the vram and the system ram. Be that as it may, unfortunately it has gotten to the point where many consumers cannot play certain aaa titles with even 4 gb of vram no matter how much system ram is available, or how fast it is. This in large part explains the weird 12gb of ram on the titan x, which even running 8 or 9 intensive games/applications simultaneously eats up less than 10gb and the performance is bottlenecked by other limitations of the card and system generally. AMD likely knows that the promise of combined vram in dx12 is yet only a promise, and so, rumor has it, decided to produce an 8gb version of the 390. Unfortunately, I believe that this is going to bump up the launch retail price of such a card, but here's hoping i'm wrong...

 

[exar333]

This news isn't too surprising as HBM is probably pretty expensive, being brand-new. That doesn't matter 'as much' on the higher SKUs, but it could be a $50 increase in cost (just throwing that out there) for a mid-range part, and that's substantial.

HBM on the top-end SKUs should be fantastic and hopefully AMD will replace a 380/380x with maybe a 385/385x with 4GB HBM in 3Q/4Q when costs go down and expensive HBM is more available.

If the 390x is ~ Titan X, and the 390 is between the 980 and Titan X, I don't see a big issue with the 380X (re-brand of current 290x). That performance difference isn't bad, and the current 290/290x pricing would likely remain the same. Sure, the 380/380x might get killed on efficiency vs. 970, but its a great performer.

Where AMD will be hurting is in the lower-mid range area (think 960 competitor). This is where a fatter 285 might be helpful, and more than 2GB RAM options.

Just my $0.02.

Edit: The re-branding also may explain the delay in rolling this out as AMD will want to reduce the redundant 2xx volume as much as possible (or more correctly, their channel partners ) prior to launching the new cards. This also may explain some 'shortages' of volume recently too, since production likely shifted over. It will be interesting to see what tweaks/changes were done on the re-brands, and if 're-brand' is even the right word here...

 

[parvadomus]

R9 390X - 4096 sp, 4 shader engines (4 x 1024 = 4096 sp), 4 tesselation engines, 4 raster engines, 8 GB HBM, 64 or 128 ROPs, designed for 4k gaming. 1/8 rate double precision for Radeon and 1/2 rate double precision performance for flagship Firepro. 500 - 550 sq mm die. TDP 260W.

USD 800 . 5 - 10% faster than Titan-X

If 390X is 4096SP - 128ROPs it will certainly destroy Titan.
That is R7 265 x 4, it will match it @ 800-850Mhz. D:

 

[Gloomy]

I think they'll keep the configuration from Tonga, 512SPs per Shader Engine. 8 Shader Engines. With either 2 or 3 ROP clusters per shader engine (64 or 96 ROPS).

 

[bystander36]

If 390X is 4096SP - 128ROPs it will certainly destroy Titan.
That is R7 265 x 4, it will match it @ 800-850Mhz. D:

I don't know where you come up with that.

The Titan X has exactly 50% more SP's than the 980, those 4096 SP is slightly less than 50% more SP than the 290x. The Titan X also has 50% more ROPs than the 980, while the 390x has slightly less than 50% more than the 290x. Based on that, I'd expect to have a similar increase over the 290x as the Titan X has over the 980. The one area the 390x may show a larger increase in spec's is the clocks.
http://videocardz.com/55146/amd-radeon-r9-390x-possible-specifications-and-performance-leaked

 

[Gloomy]

I don't know where you come up with that.

The Titan X has exactly 50% more SP's than the 980, those 4096 SP is slightly less than 50% more SP than the 290x. The Titan X also has 50% more ROPs than the 980, while the 390x has slightly less than 50% more than the 290x. Based on that, I'd expect to have a similar increase over the 290x as the Titan X has over the 980. The one area the 390x may show a larger increase in spec's is the clocks.
http://videocardz.com/55146/amd-radeon-r9-390x-possible-specifications-and-performance-leaked

How do you know how many ROPs the 390X has?

 

[RussianSensation]

Worst case scenario:380/380x are rebrand 290/290x
390x-underperforms and only 20% above GTX980 in 1080/1440P on par with TITANX in 4K
390- 5% above GTX980 in 1080/1440 and 15-20% faster in 4K
380x-rebrand 290x.Same/worse performance than GTX970
380-rebrand 290.Same performance as 290

I don't think this is going to happen because R9 290X is faster than a 970.
http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_Titan_X/29.html
and
http://www.sweclockers.com/recension/20193-nvidia-geforce-gtx-titan-x/18#pagehead

The Titan X has exactly 50% more SP's than the 980, those 4096 SP is slightly less than 50% more SP than the 290x. The Titan X also has 50% more ROPs than the 980, while the 390x has slightly less than 50% more than the 290x. Based on that, I'd expect to have a similar increase over the 290x as the Titan X has over the 980. The one area the 390x may show a larger increase in spec's is the clocks.
http://videocardz.com/55146/amd-radeon-r9-390x-possible-specifications-and-performance-leaked

In your assumptions you didn't account for a single improvement in GCN architecture (Tonga has 40% more efficient colour fill-rate/memory bandwidth and superior tessellation/geometry performance to R9 290X). Therefore, it's not as simple as just applying a 45% increase to R9 390X based on 4096 SPs/256TMUs.

The Titan X is 45-48% faster than a 290X depending on the review. Based on paper specs alone (1.05Ghz 4096 SP/256 TMU part), the R9 390X has a 53% greater theoretical performance, not accounting for a single improvement in GCN 1.2/1.3 vs. 1.1.

Which would make the +20% above GTX 980 which VR-Zone presented legit.

That estimate doesn't make a lot of sense. Even with only a 50Mhz bump in clocks, the R9 390X is already 53% faster on paper than a 290X. Let's assume non-linear scaling and give it 45-48% faster performance and we are within 0-3% of the Titan X and > 33% faster than a 980! A lot of people overestimate the average performance increase of a 980 over a 290X because certain reviewers fill most of their test suite with GW titles which massively skews things in favour of a 980. Over a larger test suite, the 980 is only 6-9% faster than a 290X at 4K.

If we look at R9 290X vs. 280X, GCN has nearly perfect scaling with increased number of SPs/TMUs. 37.5% maximum theoretical increase <=> 35% real world.

Therefore, how can a 4096 SP/256TMU R9 390X @ 1.05Ghz be only 20% faster than a 980 at higher resolutions? If anything, the R9 390X has the potential to outperform the Titan X at 4K.

 

[bystander36]

I don't think this is going to happen because R9 290X is faster than a 970.
http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_Titan_X/29.html
and
http://www.sweclockers.com/recension/20193-nvidia-geforce-gtx-titan-x/18#pagehead



In your assumptions you didn't account for a single improvement in GCN architecture (Tonga has 40% more efficient colour fill-rate/memory bandwidth and superior tessellation/geometry performance to R9 290X). Therefore, it's not as simple as just applying a 45% increase to R9 390X based on 4096 SPs/256TMUs.

The Titan X is 45-48% faster than a 290X depending on the review. Based on paper specs alone (1.05Ghz 4096 SP/256 TMU part), the R9 390X has a 53% greater theoretical performance, not accounting for a single improvement in GCN 1.2/1.3 vs. 1.1.

It's not like the Titan X has a 50% higher performance than the 980. They never scale perfectly like that.

The point stands though, based on the stats he gave, it probably won't destroy the Titan X. It'll likely see a similar improvement.

 

[digitaldurandal]

many consumers cannot play certain aaa titles with even 4 gb of vram no matter how much system ram is available, or how fast it is.

What game requires more than 4 GB of vram? Are you talking about playing one specific game at 4k with maxed out MSAA? If so that definitely doesn't affect many consumers and likely the GPU would give out before the card could handle the settings regardless.

I think the Titan X has 12 GB of ram for three reasons.

1 is it looks great on the box, the same reason why there were cards with very low performance with 4 GB of ram when Nvidia's top cards had 2.

2 at least originally the Titan line was designed for more than just gaming performance. Although now that the DP is gone I don't know if that is still true or not.

3 I think they put 12gb simply to say they have that over AMDs 8gb card.

 

[escrow4]

So someone give me a figure going from a 780Ti GHz @ 1215MHz in game @ 1200p -> Titan X @ stock for now - how many % faster. If its 30%+ I'll buy one and upgrade. 980 is a sidegrade and Nvidia has murdered Kepler so . . . . . . . and is a 390X even coming before GTA V is out?

 

[Silverforce11]

It's not like the Titan X has a 50% higher performance than the 980. They never scale perfectly like that.

The point stands though, based on the stats he gave, it probably won't destroy the Titan X. It'll likely see a similar improvement.

He just showed you with actual data that shows GCN scales very well with increased shaders, rops, tmus. Almost perfect.

There's reasons to think with HBM's lower latency, they could also improve IPC/SP on top of any other architectural tweaks.

At this junction, you either are a pessimist (20% above 980), realistic (50% above R290X) or optimistic (much greater than 50% above R290X). There's leaks to support any stance.

 

[lilltesaito]

He just showed you with actual data that shows GCN scales very well with increased shaders, rops, tmus. Almost perfect.

There's reasons to think with HBM's lower latency, they could also improve IPC/SP on top of any other architectural tweaks.

At this junction, you either are a pessimist (20% above 980), realistic (50% above R290X) or optimistic (much greater than 50% above R290X). There's leaks to support any stance.

I will take door number 2 please.