Wat?!?
If you mean that it would have similar performance to a 295x2...yes, possible.
If you mean that it's just a 295x2 with HBM...no.
If you mean that it's a crossfired dual GPU....unlikely.
DeathReborn
Platinum Member
- Oct 11, 2005
- 2,786
- 789
- 136
I suspect a single GPU 390X with 8GB was planned but HBM shortages have killed it off.
Doesn't make sense. Dual-GPU cards are very niche and low volume. If anything, they need a 8GB 390X much more.
Would amd spend their thin resources to make something entirely new based on prototype memory tech only to deliver basically the same product as 295x2?
It doesn't make sense.
^
^
These.
The only way that a multi-GPU setup would make sense is if AMD has found a new way to transparently bridge two or more GPUs - in such a way as to appear as a single GPU with a single large pool of VRAM with no negative side effects. Otherwise it makes no sense.
It might be technically possible for AMD to launch a dual gpu based on the high end Bermuda XT 550 sq mm chip but I still believe AMD will not launch a dual GPU at the same time as they are unlikely to be able to meet the demand for the HBM based R9 390X flagship. The R9 390X is a massive 550 sq mm GPU with 8 GB HBM built on a massive silicon interposer. Yield challenges are likely to be there and AMD will do its best to maximize yields by having more than 1 salvage SKU. So even though there will surely be a R9 390 with 3840 sp there could also be a R9 385 with 3584 sp. Why would AMD build a dual GPU based on R9 390X when a single one itself is going to be mighty powerful and they are not going to be able to meet all of the demand. btw I expect the R9 390X to be avg 55 - 60% faster than R9 290X and the gap will be even more at high MSAA or SSAA.
yeah I agree the R9 395x2 could launch in Q4 2015 just in time for the peak holiday season. btw the smallest gap between AMD's fastest single gpu and the corresponding dual-gpu based on the same chip is roughly 50 days between HD 4870 and HD 4870x2. the gap between HD 5870 and HD 5970 was also similar (just below 2 months). The longest gap was R9 290X to R9 295x2 at 5.5 months. I think we will see a similar gap of 6 months between R9 390X and R9 395x2.
http://www.anandtech.com/show/2556
http://www.anandtech.com/show/2584
This rebrand theory is a totally rubbish one. AMD needs strong and efficient next gen GPUs at all price points and thats exactly what they are readying. I have been telling about 3 chips or ASIC (minimum) and a max of 4.Side-note: With AMD now announcing that they only expect next gen products by 2H of 2015, I think this strongly calls into question the re-branding theory. If all you are doing is re-branding R9 200 series, and AMD reduced shipments to OEMs starting as early as Q3 2014, why would AMD need another quarter+ to launch "re-brands"? Launching re-brands is the easiest thing to do and yet AMD has nothing to replace R9 270/270X/280/280X/285/290/290X.
Bermuda - 4096 sp (4 x 1024), 8 GB HBM, 4 shader engines, 4 raster engines, 8 ACE, 64 or 128 ROPs with support for 1/2 rate fp64 operations for Firepro and 1/8 fp64 rate for Radeon. Die size - 550 sq mm.
Fiji - 3072 sp (4 x 768 sp), 4 GB HBM, 4 shader engines, 4 raster engines, 8 ACE, 64 ROPs, 1/16 fp64 rate . Optimized for cost and power. Die size - 400 sq mm
Trinidad - 1536 sp( 2 x 768 sp), 2 GB / 4GB GDDR5, 2 shader engines, 2 raster engines, 2 ACE, 32 ROPs, 1/16 fp64 rate, Die size - 250 sq mm.
I do not see AMD selling any card above USD 500 with 4 GB VRAM and I don't see the R9 390X and R9 390 selling below USD 500. So we will only see R9 390X and R9 390 with 8 GB HBM. I don't think these will come at all with 4 GB models. These SKUs are just too powerful to handicap with 4GB VRAM.
HBM does not makes sense for a product sold below USD 300. But it definitely makes sense in the USD 300+ segment. btw HBM also simplifies the PCB as the memory is on the interposer and with just 4 stacks you get 4 GB HBM. AMD can optimize board layout and come out with very nice and compact designs for R9 380X / R9 380 with 3072 / 2816 sp and 4 GB HBM. but R9 370X and lower perf chips will remain on GDDR5. One reason for me to believe that AMD will have a R9 380X with 4GB HBM is such a chip can be fit in notebooks with lower clocks and voltage/leakage binning whereas any GDDR5 based notebook GPU cannot have a larger than 256 bit memory bus width.
Last edited:
We are just developing an interposer for memory stacking. Stacking whole GPU would require much more advanced interface to seamlessly stack GPUs. It may be possible, but I don't think it will happen anytime soon.
Many people here, including me, had reasons to believe 390X was gonna be a little slower than Titan X. There is a thread made today on Chiphell where they said 390X would be 10-20% slower than Titan X.
If GTX 980Ti is faster than Titan X, and cost $700, AMD will have huge problems. Especially since the recent rumor said 980Ti is launching in May before AMD gets to introduce the 390X at Computex.
And to top it all up, they have problems with yields with HBM meaning they won`t be able to ship many units to buyers.
Perhaps Nvidia knew this, thats why they wanted to wait til 2016 before using HBM on their cards. To let the production have a chance to mature and yield rates become better?
395X2 will be one interesting graphic card, thats for sure. Can`t wait to see the TDP on that beast
Good theory. Better to make the HBM available for the 395X2 instead to let performance charts push the sales rather than using HBM on 390X which may not do well in a comparison against GTX 980 Ti.
Last edited:
Assuming the leaked 8GB slides are real they already contradict what you're saying.
http://www.overclock3d.net/gfx/articles/2015/03/16050323113l.png
http://www.kitguru.net/wp-content/uploads/2015/03/AMD-Radeon-R9-390X-Specifications-1024x578.jpg
"up to" 8GB repeated over and over again. And that's marked as the 390X WCE, says nothing about 390.
If there was no 4GB 390X part there would no reason to state that it's "up to 8GB".
Last edited:
There are some truly clairvoyant posters on Chiphell, namely those who works in the industry or at a AIB.
Having said that, don't trust everything you read on Chiphell's forum.
The OP who made that thread also made a thread on March 24th claiming Fuji will only be 20% faster than the 980. His source was wccftech.
:hmm:
20% slower than Titan X would be about 20% faster than 980 (total of 40% diff), so that actually makes him sound more credible, not less.
Until the card is released and tested, I will continue to believe that 390X:
1. Will be faster than Titan X
2. Have 8GB HBM
3. Single GPU
4. $699
No matter how delusional it may be.
20% slower than Titan X would be about 20% faster than 980 (total of 40% diff), so that actually makes him sound more credible, not less.
Until the card is released and tested, I will continue to believe that 390X:
1. Will be faster than Titan X
2. Have 8GB HBM
3. Single GPU
4. $699
No matter how delusional it may be.
I believe that makes me delusional as well.
I worry though, than 390 will be available first, and 390x launching later/paper launch. According to whispers on chiphell, AMD was caught off guard of Titan X's glorious 12 GB, and would need some time to work on the 8GB version of 390x.
If AMD would have released the card 5 months ago they would have sold a few. But AMD likes being late to the party, as always. Everybody drunk and a couple of empty kegs.
AtenRa
Lifer
- Feb 2, 2009
- 14,003
- 3,362
- 136
AMD first on 55nm with HD4870 and GDDR5. Nvidia late at 7 months with GTX285 GDDR-3
AMD first on 40nm with HD4770, April 2009
AMD first on 40nm with HD5870. Nvidia late at 6 months with GTX480
AMD first on 28nm with HD7970. Nvidia late at 3 months with GTX680
If the R9 390X is only 20% faster than GTX 980 (and thus slower than the Titan X), I don't see how it can possibly justify the use of expensive HBM. The only way this makes sense is if the card isn't primarily meant to gain sales, but to give AMD's engineers experience with HBM for next year's APUs. But that's a risky move for a company so strapped for cash.
Think about the price points. If R9 390X is 20% faster than GTX 980 and has a reasonable TDP (220W or less to get similar perf/watt) then AMD is not going to be able to sell it for more than $599, and that's if they are lucky. $499 is more likely. Even when all else is equal, people will just pay more for Nvidia cards than for a comparable AMD card, whether or not that's fair.
If the TDP weighs in at the rumored 300W+, that's even worse. Now AMD not only has to pay for expensive HBM, but a water cooler as well - and if it can't beat Titan X, they'll be lucky to get even $399 for it. This is a money-losing proposition.
AMD needs the performance crown back, but more than that, they also need decent chips at the midrange and not just a bunch of rebrands. Without something to go up against not only GM204 but also GM206 and GM107, they're dead in the water.
Think about the price points. If R9 390X is 20% faster than GTX 980 and has a reasonable TDP (220W or less to get similar perf/watt) then AMD is not going to be able to sell it for more than $599, and that's if they are lucky. $499 is more likely. Even when all else is equal, people will just pay more for Nvidia cards than for a comparable AMD card, whether or not that's fair.
If the TDP weighs in at the rumored 300W+, that's even worse. Now AMD not only has to pay for expensive HBM, but a water cooler as well - and if it can't beat Titan X, they'll be lucky to get even $399 for it. This is a money-losing proposition.
AMD needs the performance crown back, but more than that, they also need decent chips at the midrange and not just a bunch of rebrands. Without something to go up against not only GM204 but also GM206 and GM107, they're dead in the water.
These products take years to develop and it's difficult to tell where the other company will be when your product comes out. Neither Nvidia nor AMD have a crystal ball.
New memory tech with a massive 550mm2 die and its 30% faster than R290X (10% slower than 980!)?
Have some common sense people.
Have some common sense people.
RussianSensation
Elite Member
- Sep 5, 2003
- 19,458
- 765
- 126
This is one of the funniest posts I've read on R9 390X. So let me get this straight, GTX980 that costs $500-550 but only beats a $280-300 R9 290X by 8-15% is selling like hot cakes, but R9 390X that would be 91% on the TPU chart based on your own estimate should sell for $399-499 at best?
Nah, AMD won't survive unless it sells a card with 91% of Titan X's performance unless it's $299. Let's bring back HD4870 days. If you told me right now that GM200 6GB is $699-799 and Titan X is $1000, but R9 390X only delivers 90-91% of Titan X's performance, there is NO way I would pay more than $499 for the R9 390X. Why? Cuz NV is the bestest brand in the world!! ^_^
By the time R9 390X threads are done, some people won't be happy until it's 20% faster than the Titan X, has 8GB HBM and costs $499, surely. :biggrin: I also forgot, if AMD doesn't beat $1,000 Titan X, they lost no matter their price/performance because 99% of entire market only buys $700-1000 GPUs. I need to write this down.
Do you honestly believe in what you just typed?
I am guilty of this myself sometimes, but it often feels like some people on this forum have no idea about worldwide prices of GPUs.
Here is pricing in my country after tax:
1. Asus Titan X = $1,547 CDN ~ $1258 USD
2. Asus Strix 980 = $812 CDN ~ $660 USD
Are some people here actually saying with a straight face that a card 20-30% faster than a 980 is a failure if it's priced at $550-700 USD and doesn't beat a Titan X?
The entire R9 300 series is 100% re-brands. Rumours from China point to AMD having 10 million excess R9 200 cards still in warehouses all over the world, which is why AMD delayed printing boxes and stickers for R9 300 series to H2 2015. It would cost AMD too much money to reprint the boxes and put R9 300 stickers on the shrouds of existing 10 million R9 200 inventory. This strategy is brilliant though. Sell flagship R9 290/290X cards for $250-300 for another 3 months while trying to lose more than $180 million in Q2 2015, to try to set a record for how much $ the GPU division can lose in 1 quarter. Then starting August 1, partner up with the world's top marketing firms, pay millions of dollars in perks to every review site in the world, and re-badge the entire R9 200 series while raising the price on R9 290 = 390 to $499 and R9 290X = 390X to $699. Shareholders will be ecstatic! Not even the legendary Jack Welch could have come up with such a winning business strategy.
Last edited:
This is one of the funniest posts I've read on R9 390X. So let me get this straight, GTX980 that costs $500-550 but only beats a $280-300 R9 290X by 8-15% is selling like hot cakes, but R9 390X that would be 91% on the TPU chart based on your own estimate should sell for $399-499 at best?
Yes. I stand by my statement. Consider the fact that the R9 290X, which has roughly 90% of the performance of the GTX 980, is barely selling at $329. AMD has to charge 60% of the price for 90% of the performance, and still has trouble moving inventory. You may argue that this is irrational, and that the R9 290/290X represent good values. And quite possibly they do - assuming you aren't playing mostly GameWorks titles, and don't care about energy efficiency (especially with multiple monitors, where AMD cards guzzle power even on idle), and don't use CUDA or PhysX (if you use OpenCL instead, AMD will probably do better). And there are, indeed, quite a few gamers in this category who are well-served by these cards. But whether or not you or I agree, Nvidia cards are currently considered premium products compared to AMD cards, and thus command a premium price.
The only chance AMD has of being able to charge premium prices for their GPUs is if they claim the performance crown outright. If the R9 390X decisively beats the Titan X, then they can probably get a good price for it, and the card will enhance AMD's brand image. But if they come in 10% below the Titan X as rumored by the leaks, it will only reinforce AMD's image (again, fairly or not) as a Johnny-come-lately cheap knockoff. And the price they can get for the card will reflect that.
What concerns me about this release is the water-cooling rumors. AMD looks like they might be falling into the same trap they did with the FX Centurion (9000-series) CPUs: try in vain to beat a better, more efficient architecture by throwing more power and more cooling at it. A majority of buyers aren't going to go for this.
Ultimately, I'm not going to be buying any of these cards. I do hope that AMD comes out with a replacement chip for Pitcairn (the rumored Trinidad) - one that offers a sizable performance bump over R9 270X with low power (125W TDP or less), modern GCN architecture with all the new features, and hopefully hardware H.265 decoding.
The R290X vs 980 isn't rational.
It's really the R290/X vs 970, because the 970 offers great bang for buck. The 980 was never a rational purchase to start with due to the existence of the 970 at $330 on release.
390X will be judged harshly if it does not beat Titan X. It's late, it's huge, high TDP, its using a new memory tech, it HAS to be fast or its a failure.
It's really the R290/X vs 970, because the 970 offers great bang for buck. The 980 was never a rational purchase to start with due to the existence of the 970 at $330 on release.
390X will be judged harshly if it does not beat Titan X. It's late, it's huge, high TDP, its using a new memory tech, it HAS to be fast or its a failure.
RussianSensation
Elite Member
- Sep 5, 2003
- 19,458
- 765
- 126
I disagree; it'll only be judged a failure by fanboys of a certain brand and AMD haters or Titan X users who will justify why they spent $1000-4000 on 1-4 Titan Xs for bragging rights. When the original Titan came out at $1000, it was hyped, but the minute $650 780 Ghz dropped and then $399 290 and $549 290X, no unbiased gamer with 3 brain cells would justify buying a Titan for gaming.
If R9 390X delivers 90-91% of the performance of Titan X and sells for $599-649, it will be a winner as long as it's competitive with the consumer GM200 6GB card.
With Maxwell, it seems this forum has completely lost its mind because people now completely ignore price/performance, ignore VRAM limits (960's 2GB is basically dismissed as an issue since well you can just lower IQ settings!!! Brilliant idea!). More or less every NV feature under the sun is hyped like the best thing ever, while every single AMD feature (like XDMA smoothness, Eyefinity, etc.) is ignored. 3.5GB stuttering on 970 SLI - nah who cares. SLI stuttering with FCAT - nah, who cares. AMD is even being blamed for CF failures in GW. :hmm: GPU reviewers are afraid to criticize NV because NV now has most of the bargaining power for ads/review samples. They have become like puppets (HardOCP's 960 reviews).
You know what - at this point AMD might as well go bankrupt/close their GPU division because if the market wants/is OK with $550 mid-range 980, $800 980Ti and $1000 Titan X, what's even the point of AMD's GPU division existing? I mean if we are going down the path that AMD fails automatically unless it beats a $1000 card, what's the point of even having 2 competing firms if NV already satisfies 80%+ of the market's needs? Sounds like a total waste to have a choice! According to people in this thread, including you, unless you have the performance crown, you might as well not release anything. Really now?
So you are saying you'd rather buy a $999 R9 390X that's 5% faster than the Titan X over a $549 R9 390X with 90% of the performance of the Titan X? There is nowhere on the price/performance spectrum that R9 390/390X would fall that would make it a better buy than the Titan X? I guess 99% of people on this forum are making 6 figures where $1000 and $2000 for a pair of high-end cards is peanuts. :sneaky:
In your post you already say that 980 was an irrational purchase against 970/290X but you suddenly think R9 390X must be as fast or faster than the Titan X to make sense because it's late? It's amazing how a small minority of Titan X owners on the forums and the constant NV hype and AMD bashing has gotten to some people's heads. This is the first generation of GPUs where people are blatantly defending the failure that is the 960 and are actually justifying why it's better than the R9 290. If 960 was an AMD card, and R9 290 was an NV card, under no circumstances would the 960 be recommended. Sometimes I think of just quitting these forums because the bias in favour of NV is so deep now, it's impossible to have logical discussions.
Whatever. In my country the Titan X would cost me > $1,200 USD. I don't care even if hypothetically 99.9% of the GPU market thinks $550 mid-range and $1000 flagships are good buys. I don't follow the herd and I was taught to think for myself. IMO, what we are seeing now is all the green fans coming out of the woodwork because they feel AMD is basically finished and they are just revealing their true colours. With nearly 80% market share for NV, they feel they can talk more smack now since the majority of the market will back their opinion, no matter how irrational. Also, it's far easier to be a part of the majority than to stand up for your own beliefs against the majority.
If R9 390X comes in just 10% slower than the Titan X and costs say $600-650, they'll just bash it for being late and underpowered. I pretty much 100% expect for this outcome to happen if R9 390X is slower. If R9 390X actually ties or beats the Titan X, they'll say who cares since it's late and they had Titan X performance for X months, blah blah blah. I guess the term enthusiast has changed over the years to mean whoever can afford to spend the most on PC parts.
Honestly, some people on these forums need to ask themselves - what's even the point of AMD if NV delivers 100% of the time every generation and they think without AMD NV won't slow down progress or raise prices? In that case, AMD doesn't need to exist, period.
Last edited:
I'll judge it as a non-bias gamer, thanks. I will say it again: Coming very late with a massive die, high TDP & an advantage on memory tech without beating Titan X is FAILURE. Utter FAILURE.
Why? Because AMD hasn't gone for a massive GPU die ever. They've been competing with smaller dies and coming close to NV's huge die. Now they are about to release their own massive GPU with a huge memory bandwidth & latency advantage.. and they cannot beat GM200? It would be much worse than Hawaii, a 25% smaller die to GK110 that is able to match it and eventually exceed it. They won't operate with a big mm2 deficit, they have the advantage in tech, they absolutely have to pwn GM200, no excuses.
Also, AMD hasn't ever been a "winner" when they didn't have the performance crown. Selling value cards for lower margins than their competitors != winning.
If your idea that AMD is winning by coming second best and undercutting NV or Intel, its not a very good one with terribly diminishing marketshare.
Historically, AMD has never been so close on performance to NV's big-die as Hawaii vs GK110 was/is (compare 5870 vs 480, 6970 vs 580, Tahiti vs GK110). It's a steady progression where AMD is catching up. For them to suddenly go balls to wall with a massive GPU + new vram tech and not exceed NV's big-die is pure fail.
Last edited:
RussianSensation
Elite Member
- Sep 5, 2003
- 19,458
- 765
- 126
Alright, then your viewpoint is no matter the price, if R9 390X matches the Titan X or is even 0.1% slower compared to the Titan X, a it's failure. If I understood this correctly, then, I fully expect you to be buying Titan X(s) at $1000 a pop on the date of 390X's release if the above statements are true, right?
I guess to you R9 290/290X were utter failures too. Not even sure why you bought your 290s then when you could have had dual Titans for $2000 9 months before that? If price doesn't matter to you in any way, one has to ask why in the world are you even waiting for R9 390/390X or care how it performs? People who can easily afford $1000-2000 cards, should not even care because they could always just sell them and get a 10-20% faster card down the line. The type of consumers who freely purchase $1000 cards in pairs do not need to wait to save $400-500 on some future GPU or wait 3-6 months to get 10-15% more performance. They just don't care because they'll buy the next fastest $1000 card(s). Your argument is all over the place and lacks consistency because if all you care about is performance above all else, and timing is so critical for your gaming needs, you would have already had Titan Xs. :hmm:
Last edited:
You are missing the point.
AMD makes smaller GPUs that come close to bigger NV GPU. The performance gap has been historically 15-20%.
The gap closed a great deal with Hawaii vs GK110, with the R290X beating Titan and eventually beating 780Ti.
Analyze it on a tech level. Die size, TDP, architecture, ram advantage. If AMD's latest 550mm2 die cannot beat GM200, it's a failure.
AMD makes smaller GPUs that come close to bigger NV GPU. The performance gap has been historically 15-20%.
The gap closed a great deal with Hawaii vs GK110, with the R290X beating Titan and eventually beating 780Ti.
Analyze it on a tech level. Die size, TDP, architecture, ram advantage. If AMD's latest 550mm2 die cannot beat GM200, it's a failure.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 23K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter