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

[AtenRa]

Really?

What do you think the whole R2xx series with the exception of the 285/290/290X is then?:whiste:

One think, those rebrands came with Kepler. Today rebranding current 2xx series when Maxwell is in the game it is non-productive 😉

 

[Rvenger]

If any rebranding is going on they need to be rebranding Tonga into different flavors.

I wouldn't even mind a lower clocked Hawaii, the power consumption massively increases once the voltage is increased. If you undervolt Hawaii, it massively cuts down on the power consumption. I can knock 50 watts off by going from 1.25v to 1.20v and 50mhz reduced clocks. AMD puts unnecessary voltage to their GPUs much like their CPUs IMO. Maybe the voltage was needed early on release, but certainly not these recent hawaii chips. That recent [H] review of the 290x PCS+ certainly sheds some light on it too.

Same will probably be done with the 390x, pumping too much voltage to the GPU to cover for high leakage chips.

 

[Cloudfire777]

If any rebranding is going on they need to be rebranding Tonga into different flavors.

They will do that. Not sure if its 1 or two cards though

Antigua GPU will simply be rebranded Tonga.

http://vr-zone.com/articles/amd-fij...dad-tobago-gpus-set-debut-computex/89325.html

 

[Enigmoid]

One think, those rebrands came with Kepler. Today rebranding current 2xx series when Maxwell is in the game it is non-productive 😉

Because needless to say, 14/16 nm FF is going to obselete all current 28 nm GPUs. Which means that any new designs on 28nm have a very short lifetime.

New 28nm GPUs cost masks, validation, distribution, etc. Probably in the tune of $25M+ for an entire lineup change. Furthermore, this is additional money that must be made off future sales, over what the old products would have earned (ie for the same profit the old products would earn more to make up for the $25M sunk costs). If that cannot be recouped then the venture is rather pointless.

 

[mohit9206]

I think the biggest question is whether the R9 390X will support the SmartShader technology or not? Nvidia already supports this new technology via the nfiniteFX engine atleast in their current hi end cards.

 

[Silverforce11]

They will do that. Not sure if its 1 or two cards though


http://vr-zone.com/articles/amd-fij...dad-tobago-gpus-set-debut-computex/89325.html

Don't put too much faith on vrzone, they are regurgitating Chiphell performance leaks:

In the link you posted, they are talking about performance numbers looking good, 20% above 980 for the 390X.. then they link this:

http://vr-zone.com/articles/benchma...-x-amd-radeon-r9-390x-gpus-surface/88776.html

So those guys don't know maths. It's not 20%. lol

 

[Shehriazad]

Either way...I would like it if AMDs top chip had the ability to take over the enthusiast market for a while...sure, it's not gonna mean the world at first...but they kind of need anything they can get. xD

 

[SimianR]

If any rebranding is going on they need to be rebranding Tonga into different flavors.

I wouldn't even mind a lower clocked Hawaii, the power consumption massively increases once the voltage is increased. If you undervolt Hawaii, it massively cuts down on the power consumption. I can knock 50 watts off by going from 1.25v to 1.20v and 50mhz reduced clocks. AMD puts unnecessary voltage to their GPUs much like their CPUs IMO. Maybe the voltage was needed early on release, but certainly not these recent hawaii chips. That recent [H] review of the 290x PCS+ certainly sheds some light on it too.

Same will probably be done with the 390x, pumping too much voltage to the GPU to cover for high leakage chips.

I agree. I think if anything though it shows that the 290/290X were probably clocked aggressively above what AMD had originally designed, probably to be more competitive with the 780 and Titan. A 290 clocked at 900 mhz (just 50 below default clocks) and undervolted slightly saves somewhere around 50-60 watts, as you said. It sounds like AMD is going to be very aggressive with the 390X, but if the 390 (non-x) is on air, it might be in their best interest to be more reasonable with the clocks to try and repair the brand image of being hot/noisy/power hungry.

 

[RussianSensation]

Really?

What do you think the whole R2xx series with the exception of the 285/290/290X is then?:whiste:

I don't think you read my statement carefully. R9 290 and R9 290X already sell for $249 and $290-300 in the US. Taking those same chips and just re-branding them straight to R9 380/380X at $249/299 does nothing to change AMD's current market situation. AMD will need to make some changes such as higher clocks, or lower power usage or they'll need to lower prices below those levels. Today AMD can't even sell R9 290/290X at $249/299 so re-branding with no changes does nothing at all.

Also, you guys are contradicting yourselves. You state that OEMs/retailers are telling AMD to stop shipping R9 200 series because the demand isn't there but yet you support the theory of re-branding. If OEMs/retailers aren't interested in $200 R9 280X, $250 R9 290 and $300 R9 290X, they wouldn't be interested in overstock of re-branded R9 370X/380/380X cards at those prices because it would mean AMD would be back to having more overstocked inventory again, only now it would be called R9 300 series.....

This straight up re-branding theory also has major flaws. For starters, if R9 390X is rumoured to be priced at $699, R9 390 nonX would be $499-549. What in the world are you going to have between a $299 R9 380X which you think is a re-branded R9 290X and a $499 R9 390 nonX? Makes no sense!

It's also hard to believe that for the last 1.5 years AMD's GPU division worked on only 2 new chips (R9 390/390X). Whaaat?! That means if you guys believe the 100% re-branding theory besides 390 cards, you also believe AMD will have no new mobile dGPUs for another 1.5 years until 14nm GPUs because you can't use R9 290/290X/390/390X in laptops. Are you guys serious?!

Finally, it's kinda hard to believe that in 1.5 years AMD can't even manage to get a 5-10% increase in GPU clocks on an R9 290/290X card at the same power usage. If clocks are raised, that's automatically not a re-brand because it's a new ASIC revision. Since GTX980 is only 8-15% faster than an R9 290X, it would actually be in AMD's best interests to try and get R9 380X to match a GTX980 and price it at $349-399. But I guess when it comes to AMD, AMD can't improve perf/watt, can't raise clocks at similar power usage 1.5 years later and can't at all improve existing design to squeeze 10-15% more performance out of the same architecture (I guess HD4890 didn't exist either).

AMD could reuse HD7950 V2/7970/7970Ghz as rebrands last generation because GTX760/770 weren't slaughtering those cards in perf/watt or performance. Simply re-badging R9 290/290X at $249/299 does nothing to help AMD against GTX970/980. Furthermore, if R9 390 nonX > 980 and is priced at $550 or lower, it will force NV to drop prices on 980. All of a sudden a $299 R9 290X = aka R9 380X would bomb in sales against a $399 price-reduced GTX980. NV also can just drop GTX970 to $269 and with that 1 move wipe out both the R9 380/380X if they are just re-badged R9 290/290X. I am having a hard time believing AMD is that stupid.

 

[Rvenger]

Tahiti and Pitcairn need to be EOL. They were great chips and still continue to reap the performance enhancements from Mantle and driver releases, but they cannot offer freesync, VSR on a hardware level, or XDMA crossfire. They need to go just like Kepler did now. People who still use GCN 1.0 since launch got the best value for their money IMO because that was one strong running architecture once the driver issues were ironed out towards the end of 2012 and continues to be to this day.

AMD should not increase Hawaii clocks, they need to cut down Fiji to a 3rd tier to flagship. Hawaii (Grenada) needs to be in between full blown Tonga, then cut down Tonga 3 tiers as well.

Fiji XT - 390x
Fiji Pro - 390
Fiji LE - 380x or what ever but cut the shaders/ROPs and bus width
Hawaii (Grenada) XT - 380x@ 950mhz/2816 SP
Hawaii (Grenada) Pro - 380 @ 900mhz/2560 SP
Tonga XT - 370x (4gb GDDR5/2048 SP @ ~1100mhz)
Tonga Pro - 370 (4gb GDDR5/1792 SP @ ~1000mhz)
Tonga LE - 360x 2gb GDDR5/1536 or less SP @ ~800mhz)

Anything below they should be safe to use Bonaire.


Like I said, if they were going the rebrand route, this example above, would be ideal.

 

[digitaldurandal]

NV also can just drop GTX970 to $269 and with that 1 move wipe out both the R9 380/380X if they are just re-badged R9 290/290X.

Even if AMD rebadged the cards and lowered the price even further NV would not respond with a $269 GTX970. I don't believe for a second they will lower the price beyond $299 and IMO that would be in the case of serious sales issues or upcoming EOL for the card.

Nvidia has always been able to charge more for less. They continue to profit doing so and their partners continue to profit doing so. I don't see why they would change now.

I could see a drop in price for the 980 though, right now it is collecting a hefty premium over the 970. Once 980ti drops the two cards will end up over and under the current price point.

 

[AtenRa]

Tahiti and Pitcairn need to be EOL. They were great chips and still continue to reap the performance enhancements from Mantle and driver releases, but they cannot offer freesync, VSR on a hardware level, or XDMA crossfire. They need to go just like Kepler did now. People who still use GCN 1.0 since launch got the best value for their money IMO because that was one strong running architecture once the driver issues were ironed out towards the end of 2012 and continues to be to this day.

AMD should not increase Hawaii clocks, they need to cut down Fiji to a 3rd tier to flagship. Hawaii (Grenada) needs to be in between full blown Tonga, then cut down Tonga 3 tiers as well.

Fiji XT - 390x
Fiji Pro - 390
Fiji LE - 380x or what ever but cut the shaders/ROPs and bus width
Hawaii (Grenada) XT - 380x@ 950mhz/2816 SP
Hawaii (Grenada) Pro - 380 @ 900mhz/2560 SP
Tonga XT - 370x (4gb GDDR5/2048 SP @ ~1100mhz)
Tonga Pro - 370 (4gb GDDR5/1792 SP @ ~1000mhz)
Tonga LE - 360x 2gb GDDR5/1536 or less SP @ ~800mhz)

Anything below they should be safe to use Bonaire.


Like I said, if they were going the rebrand route, this example above, would be ideal.

Are you suggesting to rebrand Hawaii (R9 290X and 290) and at the same time regress performance as well with lower clocks ??

Not going to happen.

 

[AtenRa]

Even if AMD rebadged the cards and lowered the price even further NV would not respond with a $269 GTX970. I don't believe for a second they will lower the price beyond $299 and IMO that would be in the case of serious sales issues or upcoming EOL for the card.

Nvidia has always been able to charge more for less. They continue to profit doing so and their partners continue to profit doing so. I don't see why they would change now.

I could see a drop in price for the 980 though, right now it is collecting a hefty premium over the 970. Once 980ti drops the two cards will end up over and under the current price point.

NV will only lower the price if it has a serious competitor, a rebranded Hawaii is not for the GTX970-980.
If AMD will have a chip with 90-95% the performance of the GTX980 at half or less the price with lower power consumption than the GTX980, or if AMD introduce a 10-15% higher performance part at lower price than the GTX980, it will only then trigger a price cut response from NVIDIA.

 

[looncraz]

I think the biggest question is whether the R9 390X will support the SmartShader technology or not? Nvidia already supports this new technology via the nfiniteFX engine atleast in their current hi end cards.

SmartShader is an AMD technology (acquired when they bought ATI)... it dates to 2001. I'm thinking you mean something else.

Then again, nfiniteFX dates to 2002, it seems, so maybe you're just a decade or so late to the party?

Programmable shaders exist in all current products, AFAICT.

 

[tential]

Are you suggesting to rebrand Hawaii (R9 290X and 290) and at the same time regress performance as well with lower clocks ??

Not going to happen.

I think it goes back to something he was saying previously where he wants to also cut the voltage down a lot on those cards too so they work in a tighter power envelope which is apparently important to consumers now that Maxwell is out.

 

[AtenRa]

I think it goes back to something he was saying previously where he wants to also cut the voltage down a lot on those cards too so they work in a tighter power envelope which is apparently important to consumers now that Maxwell is out.

Rebranding Hawaii R9 290X to R9 380X with lower power but at the same time regressing performance lower than GTX970 will not make you sell the new Graphics Card at higher price or even the same price as you currently sell the R9 290X.

 

[raghu78]

NV will only lower the price if it has a serious competitor, a rebranded Hawaii is not for the GTX970-980. If AMD will have a chip with 90-95% the performance of the GTX980 at half or less the price with lower power consumption than the GTX980, or if AMD introduce a 10-15% higher performance part at lower price than the GTX980, it will only then trigger a price cut response from NVIDIA.

This is exactly what I expect will happen with the R9 380X. btw I am confident AMD has 3 new chips with the latest GCN architecture.

BERMUDA
R9 390X - 4096 sp, 8 GB HBM, 64 or 128 ROPs, 1/8 fp64 rate for Radeon and 1/2 fp64 rate for Firepro. - 500 - 540 sq mm. USD 750 - 800. 10% faster than Titan-X.

R9 390 - 3840 sp, 8 GB HBM, 64 or 128 ROPs, 1/8 fp64 rate for Radeon and 1/2 fp64 rate for Firepro. - 500 - 540 sq mm. USD 600 - 650. equal to Titan-X.

FIJI
R9 380X - 3072 sp, 4 GB HBM, 64 ROPs, 1/16 fp64 rate - 380 - 400 sq mm. USD 400 - 450. 15 - 20% faster than GTX 980.

R9 380 2816 sp, 4 GB HBM, 64 ROPs, 1/16 fp64 rate - 380 - 400 sq mm. USD 300 - 350. 5 - 10% faster than GTX 980.

TRINIDAD
R9 370X - 1536 sp, 2 GB GDDR5, 32 ROPs. 5-10% faster than GTX 960. USD 200

R9 370 - 1408 sp, 2 GB GDDR5, 32 ROPs. on par or 5% slower than GTX 960. USD 150

 

[ocre]

The 290X vs. 290 scaling doesn't work because there is not a linear increase in ROPs/memory bandwidth. R9 390X should have more than double the effective memory bandwidth increase over R9 290X and more than 50% increase in shader, texture and pixel fill-rate. But even if we assume your 45.6% increase, that puts us 2-3% behind Titan X. At $700, that would already make it a better buy than the Titan X. Now imagine R9 390 nonX just 10% slower than an R9 390X --> we would end up with a chip 13-14% slower than the Titan X for $500. Say hello to 2 of those in CF! Good-bye Titan X! :biggrin:

As I said before, all the hype is centered around 390X but if you look back at the history of AMD (5850/6950 unlocked/7950/R9 290 nonX), it's those 2nd tier AMD cards, when overclocked, provide hands down the best value on the AMD high-end. Remember this article?

AMD tends to charge too high of a premium for the small difference for its flagship cards (remember X850XT PE vs. X800XT or 9800XT 256MB over 9800 Pro?) AMD can still entice gamers to step-up to the R9 390X at $700 over the $500 R9 390 nonX if the latter only has 4GB of VRAM+air cooling, but the former has 8GB+AIO CLC. That would provide a huge incentive for high-end gamers to step-up to the 390X.

I just have no idea how that seems so exciting to you. As a person who has talked about selling their 980 to buy Fiji, I would really really be disappointed with that scenario.

Maybe I went the wrong way about it or it is that hard to understand what I am getting at.

500$ for a card 10-15% slower than the Titan X, its not all that amassing at all. Not after all this time, not when overclocked 980s are already that fast and about that price.

I am sorry, RS. I know you have had it out against the 980 for all this time but I really hope you can see that after all these months, a 390 non x offering that price for that performance really isn't something to get super hyped up over. It is just not all that exciting.

And I know you brought up the 1000$ Titan x as justification, but you don't see me out buying the Titan x. If you read any of my post, I suggest 980 owners to run their cards overclocked. I mean, your argument for how great the 390 price would by referencing the Titan x pricing can exactly be used on the 980 right now. The 980s overclock to very decent clocks with such ease, its effortless. All without crazy noise and temps.

I just hope for something a little more compelling for the 390. I think those figures and price just isn't gonna be the splash you think it is. I am hoping for something more than that.

Its not so much the performance but the price for that performance. It wouldn't move us far enough in my opinion. The Titan x failed to move us at all in price vs performance and I was hoping for a really epic comeback from AMD. Fiji for a tko

 

[Erenhardt]

No one is expecting new mid range sweat spot gpu and Spanish Inquisition.

7870 (270x) is still quite potent card. Its very old. It was released 3 years ago. It still holds its own. It is not much slower than recently released 960, doesn't take much more power, the core is actually smaller and has less xtors.

It lacks some new features, and should be redesigned with most priority. This is where most money is made. Strong 150W desktop design can be used as top tier mobile dgpu. 3 years is a lot of time to improve performance.

The new Pitcairn type gpu would be more than welcome.

 

[boozzer]

Are you suggesting to rebrand Hawaii (R9 290X and 290) and at the same time regress performance as well with lower clocks ??

Not going to happen.

good job ignoring everything he posted.D:

 

[mohit9206]

SmartShader is an AMD technology (acquired when they bought ATI)... it dates to 2001. I'm thinking you mean something else.

Then again, nfiniteFX dates to 2002, it seems, so maybe you're just a decade or so late to the party?

Programmable shaders exist in all current products, AFAICT.

Good to know 390X will have programmable pixel and vertex shaders. Its a very important feature after all. Its the next most important feature after hardware T&L in my opinion. But do games even take advantage of these features? Because if you see this Anandtech article, it says there are still no games that take advantage of T&L so whether this programmable pixel and vertex shaders will be put to good use remains to be seen...
" We are just going to have to wait and see what real world advantages SMARTSHADER technology offers over NVIDIA's nfiniteFX technology. There is no question that the product will be impressive; let's just hope that the advantages offered will actually be used."
http://www.anandtech.com/show/800/4

 

[loccothan]

The cards are ready, the software (drivers) are not.

HBM is a brand new technology. I bet you that making drivers for those and making sure its 100% stable, aint an easy task. Probably new ways to move data in and out the memory stacks compared to traditional GDDR. 😉
Like mentioned in the OP, pushing out a fast card but unreliable could hurt them way more than losing 2 months of sales. We all know how everyone justify and spin things, even reviewers

Yep, thats true Bro.
AMD has Brand New/Fresh Tech ! Drivers are very important, and un-easy task. New land to explore for ATI CAT drivers programmers !
I'm waiting for R390X and now im happy with my RIG
Next Step will be AMD Zen 16cores 16/18nm 😀

 

[dacostafilipe]

Good to know 390X will have programmable pixel and vertex shaders. Its a very important feature after all. Its the next most important feature after hardware T&L in my opinion. But do games even take advantage of these features? Because if you see this Anandtech article, it says there are still no games that take advantage of T&L so whether this programmable pixel and vertex shaders will be put to good use remains to be seen...
" We are just going to have to wait and see what real world advantages SMARTSHADER technology offers over NVIDIA's nfiniteFX technology. There is no question that the product will be impressive; let's just hope that the advantages offered will actually be used."
http://www.anandtech.com/show/800/4

😕

*looks at date* ... ah yeah right.

 

[ShintaiDK]

No one is expecting new mid range sweat spot gpu and Spanish Inquisition.

The spanish inquisition actually gave 30 days notice.

 

[AtenRa]

good job ignoring everything he posted.D:

Do you realize that a 900MHz Hawaii Pro will have the same performance as a Tonga XT but the Hawaii chip will have higher power consumption, bigger die size and 512-bit memory that will add more cost to the PCB etc etc ??

Not only that, by doing so AMD will replace Tonga(GCN 1.2) with an older Hawaii GCN 1.1 architecture.

A scenario like this is not going to happen in a million years.