[VC]AMD Radeon R9 390X WCE Speculation Thread

[RussianSensation]

I worry about that too. AMD might come out with a very competitive card but if everyone is expecting 290X +65%, 8GB HBM, 250W TDP, etc. there will be disappointment if it doesn't achieve that. I hope the 390X is a monster but I don't want to set my expectations too high.

That never even made sense for several reasons:

1) 2560x1440 and below tends to have a lot of CPU limited benchmarks in reviews. Therefore, it's basically impossible to achieve 65% increase at lower resolutions, which makes the "65% faster on average" claim pretty baseless without specifying the resolution/games.

2) Specs don't match. The rumours have 390X at 4096 SPs @ 1.05Ghz vs. 2816 @ 1.0Ghz for 290X. With perfect scaling, we would only get 52.7% increase. The card would need to have a massive 23.3% increase in IPC per Stream Processor. Since 390X is likely going to be based on many architectural enhancements AMD used in Tonga, but Tonga hardly improved IPC by 23.3% over Tahiti, where would the extra 23.3% come from? It's different for Maxwell vs. Kepler because Maxwell was 3-4 years in development (i.e., a brand new architecture) vs. 390X which is still just an Enhanced version of GCN but not a post-GCN architecture.

3) In some benchmarks the card might be 65% faster if 290X has a specific bottleneck but a review includes a wide variety of games which all have different bottlenecks.

Whoever is making up these rumours of 390X being 65% faster on average than a 290X on average is not using logic imo. First of all, Titan X is only about 31-36% faster than a 980 and 36-47% at 1440P-4K faster than a 290X depending on the review. That means 390X doesn't even need to be 65% faster to be impressive. If 390X is $650-700 and is 47% faster than a 290X, it's already a better card for 1440P/4K.

TPU has Titan X 43% faster than 290X at 1440p/4K. Therefore, why would 390X need to be 65% faster than a 290X to be impressive? Sounds like either fabricated fantasy or someone is purposely trying to create hype that's unrealistic.

NV still left about 12-15% headroom for a performance increase in an after-market cooled consumers 1080GTX/980Ti because if they take the best binned GM200 chips and get the benefit of ~30W lower power usage with 6GB of VRAM, there is a chance we'll see 1180-1220mhz GM200 card.

I think this round pricing for a rumoured 8GB 390X, specs & pricing for the consumer GM200 6GB card, and OC vs. OC performance will be very critical. I think that's one of the major reasons AMD waited to see Titan X benchmarks and pricing. Right now the Titan X is priced way too high at $999 and the stock cooler is basically junk as it runs too hot, with some sites reporting thermal throttling, while OC on the blower results in noise levels similar to a reference 290X.

"Another shortcoming of the cooler, which looks fantastic with its black powder coat, by the way, is fan noise. While not terribly noisy, it definitely emits more noise than NVIDIA's other recent releases and roughly matches the Radeon R9 290X noise." Source

NV left the door wide open for AMD to take advantage of better price/performance and superior cooling and noise levels. Hopefully AMD executes.

 

[Cookie Monster]

could someone tell me why 290x performs better than 980 in crysis 3??? is every other game made for nv gpus or is crysis 3 made for amd gpus?

AMD GPUs tend to run better on that engine (and all games based on it like ryse I believe - correct me if Im wrong) hence another reason why they seem to hold up "better" in todays games than say Kepler based cards.

 

[Cookie Monster]

It does. Was it an article here or somewhere else? I don't remember but GDDR5 memory + controller easily take 30% or more of the total power consumption of the card. It's a lot. HBM reduces that significantly which can be used for more processing power.

IMHO for AMD this helps more with power consumption that bandwidth limits.

Im sure it has power consumption benefits no doubt about that. Would be very interesting to see the kind of benefits HBM brings onto the table in terms of power consumption which is one of the things that AMD needs to bring down.

However Id also be interested to see the kind of benefit/cost ratio between the two at this time.

Some good articles
http://semiengineering.com/time-to-revisit-2-5d-and-3d/
http://chipdesignmag.com/display.php?articleId=5279

If they have a winner on their hands, Id like for them to price them as high as they can. AMD needs a big win.

 

[boozzer]

Who remembers the last time Anandtech used a factory OC'd aftermarket card in their review?

Complaints through the roof.

http://www.anandtech.com/show/3988/the-use-of-evgas-geforce-gtx-460-ftw-in-last-nights-review

that is cause nv wanted after market oc cards vs amd reference. just look at the recent reviews, you get 980 reference, and g1(one of the best after markets) versions and yet only reference 290/x were used in the same reviews why don't the reviews include both amd reference and best customs? like they did for the nv gpus? your link actually helped my argument

 

[96Firebird]

that is cause nv wanted after market oc cards vs amd reference. just look at the recent reviews, you get 980 reference, and g1(one of the best after markets) versions and yet only reference 290/x were used in the same reviews why don't the reviews include both amd reference and best customs? like they did for the nv gpus? your link actually helped my argument

Who is using the G1?

And TechSpot is indeed using the WF3 290X, but at reference clocks.

The Titan X would cool down to just 30 degrees with an ambient room temp of 21 degrees, though the card did hit 84 degrees when under load in Crysis 3, or slightly hotter than the Gigabyte WindForce 3X R9 290X.

Source

 

[garagisti]

It would be fun if someone included the Vapor-X in the tests, or any other 290x 8gb versions checking for 4k performance, and in CF too.
http://www.tomshardware.com/reviews/sapphire-vapor-x-r9-290x-8gb,3977-3.html

edit: Found a CF test of 8gb cards... not that many games many will play.
http://www.kitguru.net/components/graphic-cards/zardon/sapphire-r9-290x-vapor-x-8gb-cf-review/13/

 

[Elfear]

Whoever is making up these rumours of 390X being 65% faster on average than a 290X on average is not using logic imo.

I agree. I think 390X will be closer to 145-150% of the 290X. That would make it very competitive with the Titan X and hopefully power consumption is close too. Although if it's significantly cheaper than the Titan X, I wouldn't really care about higher power consumption.

 

[boozzer]

Who is using the G1?

And TechSpot is indeed using the WF3 290X, but at reference clocks.

Source

I was talking about the 980/970 reviews where review sites bothered to test both reference and AIBs but not for the amd gpus. Why not include both reference and AIBs for the amd gpus also? why do review sites treat nv and amd differently? agenda? money? both? more reasons? I only found 1 review site to be fair and it is a french one, even AT was doing the same thing. kinda sad.

Forgot to ask: anyone got any reviews that includes 1080p performance? so far, I have only found 1, techpowerup that includes 1080p performance. it also got a nice performance per $, that is actually really helpful.

 

[96Firebird]

I was talking about the 980/970 reviews where review sites bothered to test both reference and AIBs but not for the amd gpus. Why not include both reference and AIBs for the amd gpus also? why do review sites treat nv and amd differently? agenda? money? both? more reasons? I only found 1 review site to be fair and it is a french one, even AT was doing the same thing. kinda sad.

Let me give you a little insight on how I believe it all works...

When sites tested the 980, they used the reference model. When they tested the 970, since there wasn't a reference model (at launch at least), they were sent aftermarket cards to test. Since they were already testing the card, they clocked at both "reference" as well as the OC that came with the card. This is what both Anandtech and TPU did, other sites may have their own methods for testing non-reference launch cards.

Want to know what other card this happened with? The 285. I know what you're thinking, an AMD card?! There was no reference model, so review sites benched the card they received, which was aftermarket. They also included results for the lower-clocked "reference" specs.

 

[raghu78]

That never even made sense for several reasons:

1) 2560x1440 and below tends to have a lot of CPU limited benchmarks in reviews. Therefore, it's basically impossible to achieve 65% increase at lower resolutions, which makes the "65% faster on average" claim pretty baseless without specifying the resolution/games.

2) Specs don't match. The rumours have 390X at 4096 SPs @ 1.05Ghz vs. 2816 @ 1.0Ghz for 290X. With perfect scaling, we would only get 52.7% increase. The card would need to have a massive 23.3% increase in IPC per Stream Processor. Since 390X is likely going to be based on many architectural enhancements AMD used in Tonga, but Tonga hardly improved IPC by 23.3% over Tahiti, where would the extra 23.3% come from? It's different for Maxwell vs. Kepler because Maxwell was 3-4 years in development (i.e., a brand new architecture) vs. 390X which is still just an Enhanced version of GCN but not a post-GCN architecture.

3) In some benchmarks the card might be 65% faster if 290X has a specific bottleneck but a review includes a wide variety of games which all have different bottlenecks.

Whoever is making up these rumours of 390X being 65% faster on average than a 290X on average is not using logic imo. First of all, Titan X is only about 31-36% faster than a 980 and 36-47% at 1440P-4K faster than a 290X depending on the review. That means 390X doesn't even need to be 65% faster to be impressive. If 390X is $650-700 and is 47% faster than a 290X, it's already a better card for 1440P/4K.

I agree 65% avg is not going to happen. But 55% avg is not out of reach. There is mention that the R9 3xx incorporates tiled GCN for better workload management. I would not be surprised if perf/sp (IPC) gets a meaningful bump. AMD has already laid the foundation for R9 390X perf scaling without bottlenecks. ROP and tesselation bottlenecks are already removed.

http://techreport.com/review/26997/amd-radeon-r9-285-graphics-card-reviewed/2

http://www.hardocp.com/article/2015...cs_features_performance_review/2#.VQkbmY6UeSo

Enhanced Godrays works without any perf loss on R9 285.

Bandwidth is going to increase in a huge way with HBM. Even with a 8GB dual link interposer design, which I am guessing is some kind of memory interleaved design, the effective bandwidth per sp and per compute unit will increase significantly due to color compression. HBM has lower latency than GDDR5 and GCN is basically a low latency architecture. I am confident R9 390X will be more than 50% faster than R9 290X on avg. I think you are making your estimates based on no IPC improvements whereas I expect that to improve meaningfully. My guess is R9 390X is going to be 50 - 55% avg faster than R9 290X , so 5 - 8% faster than Titan-X. My bet is this is where R9 390X will land.

TPU has Titan X 43% faster than 290X at 1440p/4K. Therefore, why would 390X need to be 65% faster than a 290X to be impressive? Sounds like either fabricated fantasy or someone is purposely trying to create hype that's unrealistic.

NV still left about 12-15% headroom for a performance increase in an after-market cooled consumers 1080GTX/980Ti because if they take the best binned GM200 chips and get the benefit of ~30W lower power usage with 6GB of VRAM, there is a chance we'll see 1180-1220mhz GM200 card.

I think this round pricing for a rumoured 8GB 390X, specs & pricing for the consumer GM200 6GB card, and OC vs. OC performance will be very critical. I think that's one of the major reasons AMD waited to see Titan X benchmarks and pricing. Right now the Titan X is priced way too high at $999 and the stock cooler is basically junk as it runs too hot, with some sites reporting thermal throttling, while OC on the blower results in noise levels similar to a reference 290X.

"Another shortcoming of the cooler, which looks fantastic with its black powder coat, by the way, is fan noise. While not terribly noisy, it definitely emits more noise than NVIDIA's other recent releases and roughly matches the Radeon R9 290X noise." Source

NV left the door wide open for AMD to take advantage of better price/performance and superior cooling and noise levels. Hopefully AMD executes.

I think AMD's decision to go AIO CLC is a very well thought out one. AMD gets to keep temps and noise much lower than Titan-X which will result in favourable comparison. There will be no core throttling when overclocking due to heat / temps. OC headroom should be good. I have high expectations from GF 28SHP. :thumbsup:

 

[Majcric]

I hope you're right.

 

[boozzer]

Let me give you a little insight on how I believe it all works...

When sites tested the 980, they used the reference model. When they tested the 970, since there wasn't a reference model (at launch at least), they were sent aftermarket cards to test. Since they were already testing the card, they clocked at both "reference" as well as the OC that came with the card. This is what both Anandtech and TPU did, other sites may have their own methods for testing non-reference launch cards.

Want to know what other card this happened with? The 285. I know what you're thinking, an AMD card?! There was no reference model, so review sites benched the card they received, which was aftermarket. They also included results for the lower-clocked "reference" specs.

I am glad you included the you believe how it works part :thumbsup:

if that is how it was, with no reference models available for testing. why not forgo the reference tests? just do best of the best AIBs from both sides, voila. This was exactly what I wanted when I was shopping around for a gpu back in nov/2014. why used a custom AIB and down clock it? from the power usage charts, they weren't down clocked AIBs

 

[blastingcap]

I think AMD's decision to go AIO CLC is a very well desperate move.

Fixed that for you, if their reference is WCE. However if air cooled is available on day 1 or shortly thereafter, then yes it would be a good move.

Videocardz's latest post says 50-60% faster than 290x at cherrypicked games and resolutions and settings. So I'm thinking 45-50% faster in real life games at 1600p with realistic settings. http://videocardz.com/55146/amd-radeon-r9-390x-possible-specifications-and-performance-leaked

 

[beginner99]

2) Specs don't match. The rumours have 390X at 4096 SPs @ 1.05Ghz vs. 2816 @ 1.0Ghz for 290X. With perfect scaling, we would only get 52.7% increase. The card would need to have a massive 23.3% increase in IPC per Stream Processor.

Maybe they changed the Stream processor. Remove FP64 in favor of FP32 just like the titan did.

 

[raghu78]

Fixed that for you, if their reference is WCE. However if air cooled is available on day 1 or shortly thereafter, then yes it would be a good move.

Videocardz's latest post says 50-60% faster than 290x at cherrypicked games and resolutions and settings. So I'm thinking 45-50% faster in real life games at 1600p with realistic settings. http://videocardz.com/55146/amd-radeon-r9-390x-possible-specifications-and-performance-leaked

you can call it desperate. but i and many others would disagree. firstly the titan-x is throttling according to few reviewers as it hits the 84c limit easily. Once you overclock the Titan-X , the cooler noise gets quite loud. After spending USD 1000 thats a bad situation. AMD is going to solve both issues with the AIO CLC and they are going to price the card lower to boot. :thumbsup:

 

[AtenRa]

You know what, even if x390 will not be faster than Titan-x or GTX990 (780Ti successor) on average, it is more than 5 years that ATI/AMD has a real High-End GPU that can look NVIDIAs behemoths directly in the eye.
Im very exited about this release because we will have a very competitive second source for High-End GPUs after all those years.

About the performance, at 1600p/4k with Evolve/Mantle games, 390x could be more than 50% faster than 290X and faster than Titan-X.

 

[therealnickdanger]

I watched the PCPer's live coverage of the Titan X with NVIDIA's Tom Petersen yesterday and he took a couple tiny digs at water cooling being a negative solution that "consumers don't want". He never mentioned anyone by name, just that it's something that they "would never consider" for a premium product like the Titan X.

If you can't beat 'em, nitpick them to death.

 

[3DVagabond]

I watched the PCPer's live coverage of the Titan X with NVIDIA's Tom Petersen yesterday and he took a couple tiny digs at water cooling being a negative solution that "consumers don't want". He never mentioned anyone by name, just that it's something that they "would never consider" for a premium product like the Titan X.

If you can't beat 'em, nitpick them to death.

That's looking like their only nit to pick. It's OK though, it's not like Huddy doesn't do the same.

 

[Sunaiac]

it is more than 5 years that ATI/AMD has a real High-End GPU that can look NVIDIAs behemoths directly in the eye.

you mean, unlike the 290X that looks at the 780Ti from above ?

 

[therealnickdanger]

it is more than 5 years that ATI/AMD has a real High-End GPU that can look NVIDIAs behemoths directly in the eye.

You'll have to define what "looking NVIDIA in the eye" means. At 4K, The 290X is faster than its generational equivalent - the 780Ti - in nearly all scenarios and within ~10% of the 980. The Titan X is the first significant lead NVIDIA has had on AMD since the 290X was released... and that may be very short-lived if 390X estimates are accurate (they were very accurate for Titan X).

 

[AtenRa]

Although R9 290X is doing a good job at 1600p/4K today, it was released 8 months after NVIDIAs original Titan. And NVIDIA quickly responded with the GTX780Ti that took the high-end performance title less than a month after 290X release day.

390X is going to be released within the next 2-3 months after Titan-X release and it is going to bring higher performance from day one in many titles with lower noise and better thermals.

 

[DarkKnightDude]

If there's benchmarks for the 390X avaliable, wouldn't they have a card ready then? or close?

 

[Noctifer616]

If there's benchmarks for the 390X avaliable, wouldn't they have a card ready then? or close?

Cards are already in the hands of some game developers, AMD even used the 300 series at GDC to run their VR demo. The 300 series might be ready to launch, we just can't know for sure. AMD must have some reason to release their cards in summer and not not, we just don't know for certain.

 

[Paul98]

What I want to hear about is the 380x, I want to see similar number of SP as the 290x, along with 4GB HBM, and architecture improvements.

 

[therealnickdanger]

Although R9 290X is doing a good job at 1600p/4K today, it was released 8 months after NVIDIAs original Titan. And NVIDIA quickly responded with the GTX780Ti that took the high-end performance title less than a month after 290X release day.

Maybe it was the moment it was launched before drivers were tuned properly, but the narrative for the last year from most publications is that the 290X is the better card for "high end performance".

It would be nice if we could fast-forward to 6 months from now and compare Titan X and 390X with refined drivers, but I am not a Time Lord.