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

[mohit9206]

So the 390X is supposed to beat Titan X and 980Ti is that correct? Am guessing price would be between $500-600 to lay the SmackDown on Nvidia.$499 would be a sweet price for being the fastest single gpu card. Two of those for under $1k would be a beast but would probably need a 1000 Watt PSU to power the whole system.

No more mention of nvidia
-Subyman

 

[Headfoot]

Try looking back at the history. Power/Heat may actually have been just as valid back then too

Hint: Look at AMD`s market share prior 2012 and what happened when Kepler arrived.
Hint 2: Fermi was a hot mess. Nvidia removed certain features to make Kepler more efficient.

It was a strawman back then too, just with the other side doing the misleading and dissembling. Aftermarket 480s were quite nice temperature and noise wise. 580 even more so.

All in all though, I think it's a good call on AMD's behalf to include a stock watercooled edition in addition to an air cooled edition so they don't get into a snafu where reviewers are quoting low performance figures for 1.5 years because of a poor reference model nobody buys.

 

[RussianSensation]

Two of those for under $1k would be a beast but would probably need a 1000 Watt PSU to power the whole system.

You don't need a 1000W PSU to power 2x250-300W TDP GPUs because in CF/SLI, the power usage is never 100% of a 250-300W GPU x 2 due to 80-90% CF/SLI scaling. You need to reduce the power usage of both GPUs accordingly.

in practice, at the PSU level with a Intel i7 4930K @ 4.7GHz.

http://www.hardwarecanucks.com/foru...md-radeon-r9-295x2-performance-review-18.html

Drop that CPU to an i5/i7 4690K/4770K/4790K @ 4.5Ghz, and you are under 600W of power usage at the PSU level.

I think a lot of gamers don't realize how much PSUs have dropped in price in the last 3-4 years. You can readily purchase a good 850-1000W PSU for $90-100 and a 750W for $50.

Will power any single GPU card in the world.
Antec 750W Gold w/ 5 year warranty = $50

For dual 250W cards max OCed + i7 4790K max OCed:
EVGA 850W Bronze = $90
Corsair 850W Gold Modular = $97

For dual 250W cards max OCed + i7 5960X max OCed:
Rosewill 1000W Bronze Modular = $90
Rosewill 1000W Bronze Semi-Modular = $80

For a bit more you can get 1050W Gold rated!

Rosewill Photon 1050W Gold = $110

With so many choices for high quality $90-110 850-100W PSU and $50-75 600-750W PSUs, modern PC gamers who keep complaining about a requirement for a 750-1000W PSU for a new build or a PSU upgrade that will last 7-10 years are just making excuses for not doing proper research. When I got my SeaSonic Platinum 1000W 3 years ago it was way harder to find a new high quality 1000-1050W unit for $110, if not impossible. Also, unless you are running your GPUs and CPU maxed out 24/7 for distributed computing/rendering, etc. you do not need a Platinum 850-1000W PSU as you would be paying a large premium for little benefit over a Silver/Gold one.

 

[boozzer]

AMD's market woes are almost entirely due to the mining craze. During the most important and influential period of 200 series life, it was either not available or ridiculously priced, destroying any chance for it to establish itself on the US market (and by the end it started to seep into the EU market as well).

The burst of NV market share that people seem to equate with 970/980 is simply a by-product of that.

smart guy and the used cards from mining poached alot of customers from buying new amd cards also

if only amd had enough stock to satisfied both mining and gamers. they would have made bank. too bad :\

 

[the unknown]

Is the r9 370 still on track for April? I'm just starting to look at video cards for my new build to go with my 4790k

 

[RussianSensation]

Is the r9 370 still on track for April? I'm just starting to look at video cards for my new build to go with my 4790k

There was never any credible rumour of R9 370 series launching in April. If you need to buy in a week or so, an after-market R9 290 or a GTX970 with the Witcher 3 coupon are the best cards for the $. Unfortunately if you go down below that, you are seriously compromising.

 

[Kippa]

Pricing wise do you think you might be able to get two 390X's close ish to the cost of one of the expensive Titan Xs? Not necessarily exact but a "bit" close. I'd be interested in seeing how the 390X performs and its price.

 

[3DVagabond]

Pricing wise do you think you might be able to get two 390X's close ish to the cost of one of the expensive Titan Xs? Not necessarily exact but a "bit" close. I'd be interested in seeing how the 390X performs and its price.

Just my thoughts. AMD is going to give us an 8gig water cooled monster that is going to be in the $700-$800 range. They are going to give us a 390X 4gig in the $550-$650 range. They are going to give us a 4gig cut down chip (390 non X) in the $400-$500 range. I think the top model will be faster than Titan, but will be so at the cost of efficiency. I think even the 390 will be faster than the 980 and should o/c to ~Titan X stock performance.

 

[Kippa]

Normally I don't wait around when I want to invest in hardware, this time though I am going to wait and my gut feeling is that it will be worth it this time.

 

[therealnickdanger]

So are we still looking at a June (Computex) release? I can't wait much longer...

 

[n0x1ous]

So are we still looking at a June (Computex) release? I can't wait much longer...

That's the best guess right now

 

[Cloudfire777]

From the mobile side, the R9 M385 is a confirmed Tonga rebrand.

 

[DiogoDX]

From the mobile side, the R9 M385 is a confirmed Tonga rebrand.

Rebrand from what? There are not tonga in the mobile besides apple only M295.

 

[Cloudfire777]

Rebrand from what? There are not tonga in the mobile besides apple only M295.

Wrong.

Again, Dell use M295X in their Alienware notebooks.

 

[ocre]

Just my thoughts. AMD is going to give us an 8gig water cooled monster that is going to be in the $700-$800 range. They are going to give us a 390X 4gig in the $550-$650 range. They are going to give us a 4gig cut down chip (390 non X) in the $400-$500 range. I think the top model will be faster than Titan, but will be so at the cost of efficiency. I think even the 390 will be faster than the 980 and should o/c to ~Titan X stock performance.

That would be great.

Do you think Fiji 8gb version will launch right off the bat? I was thinking the 4gb will launch first and then the 8gb soon afterwards

 

[RussianSensation]

Pricing wise do you think you might be able to get two 390X's close ish to the cost of one of the expensive Titan Xs? Not necessarily exact but a "bit" close. I'd be interested in seeing how the 390X performs and its price.

If the specs and leaked AMD slides are to be believed for the WCE R9 390X, it should be at least as fast as the Titan X. If so, unless NV launches a GM200 6GB soon, AMD could jack up the price to $700-800 for this card. I think if there is an 8GB option for the R9 390 non-X, it's the one to go for. Historically AMD's 2nd best card is the best value on the high end (5850/6950/7950/R9 290). The one time it was better to get the fastest was HD4870 imo. For $100 more, 4870 was 30%+ faster than the 4850.

AMD's dilemma is that NV has made a lot of $ and high margins on GTX970/980 cards. The cost to manufacture a 980 isn't much greater than the 970 which means by June 24, 2015 when 970/970 turn 9 months old, NV could easily drop 970 to $259-269 and GTX980 to $399-449. Alternatively, NV can cook up faster clocked GTX970Ti and GTX985 by late summer. That puts AMD in a very tough position because even if they beat 970/980 on price/performance at launch it has to be substantial enough so that even if NV drops prices, they new R9 300 series are still worth buying.

That's why to me the re-branding theory for everything but 2 cards (390/390X) doesn't make much sense because NV's re-priced 970 at $259 and 980 at $399 would make R9 370X (Tonga)/380/380X (Hawaii) pointless. I think AMD needs to try and improve R9 290/290X by at least 10% respectively to account for NV dropping prices at will.

Rebrand from what? There are not tonga in the mobile besides apple only M295.

Actually Dell has Alienware laptops with an R9 M295X as a $200 upgrade option over the 970M. I haven't found benches of R9 295X vs. 970M yet but this pricing seems whacked. I can't imagine R9 295X2 being that much faster (if at all) to warrant a $200 premium. Either way, if all AMD has for laptops is an R9 285, they are screwed because NV should refresh GTX900M sometime this year and usually when they do, 10-20% performance increases are not out of the question.

I think even the 390 will be faster than the 980 and should o/c to ~Titan X stock performance.

That's not even debatable. The question is how much faster. Even a 3328 SP R9 390 card would already beat a 980 at high resolutions. 980's lead over the 290X today is only 11-12% at 1440P and just 6-8% at 4K. A 1Ghz 3328 SP R9 390 would be 16-17% faster than an R9 290X.

P.S. Looks like R9 290 series may be close to EOL as prices are rising on Newegg and availability is extremely scarce in other stores.

NCIXUS which used to have 10+ models of R9 290X has just 5 air cooled models, with only 1 going for $345, with the rest going for $400+. When you start seeing major price increases for cards that used to sell for $300-350, it's a sure sign the card is close to EOL and soon will disappear from the store's inventory since clearly no new cards are being re-ordered.

Canada Computers - another huge chain of PC parts has only 5 R9 290+290X cards, of which only 2 models are R9 290X. Let's dive into the details - Of these 2 R9 290X cards, only 9 PowerColor are left at $781 CDN. The other one is a Reference R9 290X for $659 CDN, but when you click it, it's no longer in stock.

Looks like a complete drought of R9 290 cards and rising prices - all signs that the launch of the next generation is extremely close as the inventory is running out and not being replenished. What's odd is how quickly the inventory is running out but if the launch is only in June, in a matter of 3-4 weeks AMD will have nothing to sell at reasonable prices.

 

[Head1985]

by June 24, 2015 when 970/970 turn 9 months old, .

This is most scary thing.
I cant imagine ATI ever allow this.
Hawaii was 7 months late and didnt beat kepler.We can only hope Fiji will be better much better.
NV can easily rebrand 980 and sell it for 300 if they want.980 is still only mainstream GPU with HIGH end price because no competition.

 

[RussianSensation]

This is most scary thing.
I cant imagine ATI ever allow this.
Hawaii was 7 months late and didnt beat kepler.We can only hope Fiji will be better much better.
NV can easily rebrand 980 and sell it for 300 if they want.980 is still only mainstream GPU with HIGH end price because no competition.

Maybe some higher ups at AMD made a huge miscalculation. They either thought 20nm node was a go 2 years ago and banked on it and/or they thought HBM 8GB would be available much faster. Maybe a combination of both of these miscalculations resulted in the engineers being thrown for a loop. Since GCN is a compute monster and has DP in bucket loads, AMD's engineers would now have to figure out a way of how to improve perf/watt, keep the high DP they need for FirePro, and still manage a huge improvement in performance all in a somewhat reasonable power usage but now on a 28nm node.

With NV, they basically said screw DP performance and made Maxwell a pure SP + gaming perf/watt architecture. The end result as you can imagine is AMD's engineers being asked an impossible task of making a card that's as fast or faster for compute + gaming but has similar perf/watt. Something had to give and I think that something is going to be perf/watt. I mean if you think about it logically, NV utilized 99% of the 250W TDP in the Titan X for gaming/SP performance. That makes it way harder for AMD's engineers to compete, especially since AMD/ATI never made 550-600mm2 GPU die.

Hopefully even if R9 390 series arrives 9-10 months after 970/980 launched, it makes it worthwhile because let's face it overall 970/980 barely moved the performance benchmark from 10 months old 290/290X/780Ti. I mean today a 980 is just 6-8% faster than a 290X at 4K. That's very unimpressive. If right now you are a gamer sitting on an overclocked 290/290X/780/780Ti, 970/980 are not real worthy upgrades imo and the Titan X is way too expensive. As a result, despite how well 970/980 sold, there is a huge market for untapped upgrades that AMD can fill with a card at least 15-20% faster than a 980.

 

[therealnickdanger]

As a result, despite how well 970/980 sold, there is a huge market for untapped upgrades that AMD can fill with a card at least 15-20% faster than a 980.

I think accessing that untapped market will hinge entirely on the pricing. AMD has to be faster while being a lot cheaper with each SKU to really make this release a victory. Just beating the high-end SKUs won't be enough.

 

[DearLord]

per usual, I agree with much of what Russian has to say. Amd's dilemma is that it's going to be hard to price the 390/x competitively with the 970/80, and anything in the 500+ USD zone will not really be accessible to the mainstream. Maybe gains to the lower chips in the 300 line can make up for this, but I see the 390/x falling in a very awkward sales spot. They might get more people to bite than, say the titan x, but a lot less than a 970 with a price drop to around 250.

 

[ocre]

Maybe some higher ups at AMD made a huge miscalculation. They either thought 20nm node was a go 2 years ago and banked on it and/or they thought HBM 8GB would be available much faster. Maybe a combination of both of these miscalculations resulted in the engineers being thrown for a loop. Since GCN is a compute monster and has DP in bucket loads, AMD's engineers would now have to figure out a way of how to improve perf/watt, keep the high DP they need for FirePro, and still manage a huge improvement in performance all in a somewhat reasonable power usage but now on a 28nm node.

With NV, they basically said screw DP performance and made Maxwell a pure SP + gaming perf/watt architecture. The end result as you can imagine is AMD's engineers being asked an impossible task of making a card that's as fast or faster for compute + gaming but has similar perf/watt. Something had to give and I think that something is going to be perf/watt. I mean if you think about it logically, NV utilized 99% of the 250W TDP in the Titan X for gaming/SP performance. That makes it way harder for AMD's engineers to compete, especially since AMD/ATI never made 550-600mm2 GPU die.

Hopefully even if R9 390 series arrives 9-10 months after 970/980 launched, it makes it worthwhile because let's face it overall 970/980 barely moved the performance benchmark from 10 months old 290/290X/780Ti. I mean today a 980 is just 6-8% faster than a 290X at 4K. That's very unimpressive. If right now you are a gamer sitting on an overclocked 290/290X/780/780Ti, 970/980 are not real worthy upgrades imo and the Titan X is way too expensive. As a result, despite how well 970/980 sold, there is a huge market for untapped upgrades that AMD can fill with a card at least 15-20% faster than a 980.

I am not sure 15-20% over a 980 will get many 980 users to upgrade, but i am hoping for at least that much performance for about $500. It might attract some 970 owners though but.........

There are plenty of people out there that have 780s, gk104s, tahiti, hawaii, etc. Plenty that havent made a move yet. These people would be more likely to spend 500 on a card than people with lesser cards as they usually stick with cards in their price bracket.

The real problem is AMD needs to launch yesterday. The longer the wait, the more sales lost. That is happening every single day.
I think you are right about the miscalculation. The lack of 20nm really putting AMD off guard. When you bring up engineers having a harder time with large 550-600mm^2 die, i think it is similar to my thoughts. I think AMD had to do respin(s). That would be a major set back but could have been very rewarding in the end. Making such large die chips is hard. Nvidia has several generations of experience now. But they have had obvious issues along the way.

There is just no way i believe that AMD planned to give nvidia all this time with maxwell. I think that nvidia was looking way ahead when the decided to try first generation maxwell on 28nm. This went so well they started to move the other chips too. We can talk about it like it is easy but nothing about manufacturing chips is simple. One minor mishap and you end up months behind.

I am hoping that AMD will be able to release something really nice after all this time. Hopefully it will be all worth it to everyone that waits.

 

[JDG1980]

Maybe some higher ups at AMD made a huge miscalculation. They either thought 20nm node was a go 2 years ago and banked on it and/or they thought HBM 8GB would be available much faster. Maybe a combination of both of these miscalculations resulted in the engineers being thrown for a loop.

The higher-ups at AMD seem to have been making a lot of miscalculations. I don't know how many more the company can afford. I'm starting to get the feeling that AMD has pretty much nothing of substance to offer until 2016 - assuming they survive long enough to reach the promised land of 16nm FinFET.

With NV, they basically said screw DP performance and made Maxwell a pure SP + gaming perf/watt architecture. The end result as you can imagine is AMD's engineers being asked an impossible task of making a card that's as fast or faster for compute + gaming but has similar perf/watt. Something had to give and I think that something is going to be perf/watt. I mean if you think about it logically, NV utilized 99% of the 250W TDP in the Titan X for gaming/SP performance. That makes it way harder for AMD's engineers to compete, especially since AMD/ATI never made 550-600mm2 GPU die.

That excuse doesn't work for most of AMD's lineup. Of all the GCN chips, only Tahiti and Hawaii had substantial DP performance (1/4 SP for Tahiti, 1/2 SP for Hawaii, though the latter was limited to 1/8 on Radeons). Cape Verde, Pitcairn, Bonaire, and Tonga all have very limited DP performance (1/16 SP). The AMD cards do all work better with OpenCL, but that's a driver optimization issue, not a hardware limitation. Maxwell actually increased compute performance relative to Kepler on some integer workloads (Scrypt mining).

I think something went very wrong with Tonga - it was supposed to compete with Maxwell on perf/watt, but it couldn't. Maybe it was initially designed for GloFo's 28nm SHP process, but that didn't work out for some reason or other and it got ported to TSMC instead. The fact that no fully enabled Tonga has been released to the consumer AIB market is really odd - surely yields on a mature process like 28nm can't be so poor that the high-end Retina iMac is sucking up all of the non-defective chips. Even if AMD wanted to hold off on it on consumer cards because of a large back stock of Tahiti (sunk cost fallacy), that doesn't explain why they used a castrated Tonga for FirePro W7100 instead of the full chip.

 

[ocre]

I thought tonga was glofo 28nm silicon. Is it not?

 

[utahraptor]

At this point is there any evidence to support claims that they had to use two GPUs on the 390x flagship to hit 8GB? I have heard some rumors I dictating that. I think the bulk of the rumors say just a special interposer which hopefully won't compromise bandwidth negating the entire reason for having the special new memory.

The only reason I think the dual GPU = 8 GB version is because of the rumors they are spending a this time making the drivers deliver on the promise of unified memory for crossfire (game seeing full 8Gb instead of two copies of 4.)

 

[96Firebird]

I thought tonga was glofo 28nm silicon. Is it not?

According to Ryan Smith it is TSMC 28nm...

First chart