AMD Radeon R9 390X Arrives In 1H 2015 – May Feature “Hydra” Liquid Cooling

old gamer · 28 Sep 2014 at 17:29

humbug said:
Thats interesting.

That would put the 390X for April, the 380X for February.
With them separately specifying the 380X as 20nm in February, are we to assume the 390X would be 28nm in April?

Thats a bit odd.

Unless the 390x will also be 20nm

Never assume as it makes an ass out of u and me or so ive been told at work

Orangey · 28 Sep 2014 at 17:30

And of course wccf have regurgitated both articles already :rolleyes:

http://wccftech.com/amd-promises-competition-gtx-980-gtx-970-12-points-choose-radeon/
http://wccftech.com/gtx-980-lose-performance-crown-amds-r9-380x-febuary-2015/

win8.1 · 28 Sep 2014 at 17:32

Ayahuasca said:
Yes but aren't the same chip, you've already been told the 980/970 was built for 256bit, they aren't cut down chips as you seem to think.

I don't accept this different chip argument as they are still Maxwell. If the different chip but still Maxwell 970 equivalent sells at launch for £300 approx with 5% less performance than the full fat new chip Maxwell then at least Nvidia customers will be given the same deal as Amd does with theirs but I cannot see that happening unfortunately.

humbug · 28 Sep 2014 at 17:34

old gamer said:
Unless the 390x will also be 20nm

Never assume as it makes an ass out of u and me or so ive been told at work

I assume nothing, i speculate

Orangey said:
And of course wccf have regurgitated both articles already

http://wccftech.com/amd-promises-competition-gtx-980-gtx-970-12-points-choose-radeon/
http://wccftech.com/gtx-980-lose-performance-crown-amds-r9-380x-febuary-2015/

their on the Ball, got'a hand that to them.

Retron · 28 Sep 2014 at 17:35

win8.1 said:
I say cut down compared to full fat Maxwell will be or another way of looking at it that full fat Maxwell will have bigger bus, more rop's and shaders. What should have been out now at launch for the same price as the 970.

It's a shame that the new strategy with Kepler paid off so well for them.

Since the "named" series started with Tesla, we've had the following:

PHP:

	             Midrange (x60)	Highend (x80)
Tesla (200)	  GT200 (448-bit)	GT200 (512-bit)
Fermi (400)	  GF104 (256-bit)	GF100 (384-bit)
Fermi (500)	  GF114 (256-bit)	GF110 (384-bit)
Kepler (600)	 GK106 (192-bit)	GK104 (256-bit)
Kepler (700)	 GK104 (256-bit)	GK110 (384-bit)
Maxwell (900)	n/a	            GM204 (256-bit)

The 400, 500 and 700 series all follow the x04 chip being midrange. The 680 blatantly used a midrange chip when it was released. A cynic would say it's because Nvidia made more money selling GK110 to the compute crowd (and latterly as the Titan), then once yields improved and the performance-at-all-cost crowd had been bled dry, *then* it was time for GK110 to be released as a normal high-end card.

As that seemed to work pretty well, it's no surprise to see the same thing happening again with Maxwell. What's the betting that the consumer form of GM210 initially appears in a Titan format? Pretty high, I'd say!

CAT-THE-FIFTH · 28 Sep 2014 at 17:36

Orangey said:
And of course wccf have regurgitated both articles already

http://wccftech.com/amd-promises-competition-gtx-980-gtx-970-12-points-choose-radeon/
http://wccftech.com/gtx-980-lose-performance-crown-amds-r9-380x-febuary-2015/

I actually saw them on their website first,but I wanted to link to the direct sources instead of reading their "speculation"!!

But funnily enough I am more excited about the texture compression being used for the next AMD APUs. That could lead to a significant performance improvement,since they are very bandwidth limited.

Ayahuasca · 28 Sep 2014 at 17:37

win8.1 said:
I don't accept this different chip argument as they are still Maxwell.

Maxwell is the architecture...

The chip is different whether you want to accept that or not is up to you. You should probably get a little more clued up on things before trying to argue otherwise though.

You seem to be spending an awful lot of time worrying about these mid ranged cards.

win8.1 · 28 Sep 2014 at 17:41

humbug said:
Thats still the case today, the 290X is slower than the 980, but not far off, about ~10% and £150 cheaper.

Of course it uses more power. How much more power, or how much power the 980 actually uses is difficult to pin down as the power switching tech in Maxwell is too fast for conventional meters to read, what has been found is that its rated TDP is a lot less than it actually uses.

That's true. The 980 should also really be only sold for £300 at the most with a good custom cooler.

eddyr · 28 Sep 2014 at 17:41

win8.1 said:
It only matters if you want the best bang for buck at launch for approx £300 with only 5% less performance like Amd offered with the 290. As the gm204 cards are selling for that full fat Maxwell including it's 970 equivalent will be much higher.

I see where your coming from, but knowing some details about the arch implementation seems somewhat irrelevant from a gaming perspective. Sure full fat maxwell should perform a chunk higher than GM204, but look at it this way, future perf increases are always on their way and top performance progression is performance progression. As it was 2, 5, 10 years ago. Top spot commands a silly price. Does it really matter to the gaming market that the highest performing cores alternate between strong and weak DP compute. Not to gamers where DP compute is largely irrelevant.

The only gamers who might take full fat/lite maxwell it into consideration are the high res crowd (4k) where a larger bus width is a consistent benefit.

humbug · 28 Sep 2014 at 17:43

CAT-THE-FIFTH said:
I actually saw them on their website first,but I wanted to link to the direct sources instead of reading their "speculation"!!

But funnily enough I am more excited about the texture compression being used for the next AMD APUs. That could lead to a significant performance improvement,since they are very bandwidth limited.

A 20nm APU could easily have 1024 Stream Processors, as GCN 1.3 (Tonga) with the Texture compression and the Tessellation performance i think we may have 7850 / R7 265 performance in an APU, if they push that bit more perhaps a 7870 / R9 270.

win8.1 · 28 Sep 2014 at 17:47

Ayahuasca said:
Maxwell is the architecture...

The chip is different whether you want to accept that or not is up to you. You should probably get a little more clued up on things before trying to argue otherwise though.

You seem to be spending an awful lot of time worrying about these mid ranged cards.

Oh I'm clued up enough to know what Nvidia are doing with their last two gen's even if you are happy to buy their first Maxwell offering. Second Maxwell is what they should have been offering you at the same price as the 970 but I suppose coming from a lower performance card your going to be happy whatever.

CAT-THE-FIFTH · 28 Sep 2014 at 17:49

humbug said:
A 20nm APU could easily have 1024 Stream Processors, as GCN 1.3 (Tonga) with the Texture compression and the Tessellation performance i think we may have 7850 / R7 265 performance in an APU, if they push that bit more perhaps a 7870 / R9 270.

Sadly I think Carrizo will be 28NM,but with the texture compression technology and improved tessellation hopefully the IGP can get closer to HD7750 GDDR5 or R7 250 GDDR5 level performance. If they could do that in a 65W TDP APU,that wouldn't be that bad at all IMHO.

Orangey · 28 Sep 2014 at 17:57

4770: released 28th April 2009
5000 series: released September 23rd 2009

Isn't that about the amount of time it would take to port/qualify a design to a new process? So if this Russian source is accurate, we would maybe expect R9 490X 6 months after Feb, on 20nm? With the pre-existing 28nm design leaked on AT forums coming in soon as the 390X.

win8.1 · 28 Sep 2014 at 17:59

eddyr said:
I see where your coming from, but knowing some details about the arch implementation seems somewhat irrelevant from a gaming perspective. Sure full fat maxwell should perform a chunk higher than GM204, but look at it this way, future perf increases are always on their way and top performance progression is performance progression. As it was 2, 5, 10 years ago. Top spot commands a silly price. Does it really matter to the gaming market that the highest performing cores alternate between strong and weak DP compute. Not to gamers where DP compute is largely irrelevant.

The only gamers who might take full fat/lite maxwell it into consideration are the high res crowd (4k) where a larger bus width is a consistent benefit.

I don't mind not having the top spot card but I would like Nvidia to offer first the best of Maxwell with 5% less performance than their top spot card at £300 approx. What I don't like is milking the consumer twice with 256bit, less shaders, rop versions for £300 then £500+ cards come out on the same architecture. Even lower than 4K users would see the benefit if they did not do that.

humbug · 28 Sep 2014 at 18:00

CAT-THE-FIFTH said:
Sadly I think Carrizo will be 28NM,but with the texture compression technology and improved tessellation hopefully the IGP can get closer to HD7750 GDDR5 or R7 250 GDDR5 level performance. If they could do that in a 65W TDP APU,that wouldn't be that bad at all IMHO.

Currently Kaveri runs 512 SP's, 64bit
With Tonga tech it probably could run like a GDDR5 7750 (512 SP's, 128bit)

Its incredible that a 4 core CPU + 7750 runs at about 60 Watts in that package.

Ayahuasca · 28 Sep 2014 at 18:07

win8.1 said:
Oh I'm clued up enough to know what Nvidia are doing with their last two gen's even if you are happy to buy their first Maxwell offering. Second Maxwell is what they should have been offering you at the same price as the 970 but I suppose coming from a lower performance card your going to be happy whatever.

Well you're not though are you, even without the 970/980 nvidia would not have released a different, better chip for £260. Just like AMD's new top end card won't be <£300.

ICDP · 28 Sep 2014 at 18:19

If AMDs next card is 28nm card and requires a water cooler then they would be best to cancel it and ride the storm. It would be a total failure at less than anything but 20% + faster than GTX980.

Uncle Petey · 28 Sep 2014 at 18:21

Why would it be a failure if it's not at least 20% faster than a 980?

Final8y · 28 Sep 2014 at 18:26

win8.1 said:
Oh I'm clued up enough to know what Nvidia are doing with their last two gen's even if you are happy to buy their first Maxwell offering. Second Maxwell is what they should have been offering you at the same price as the 970 but I suppose coming from a lower performance card your going to be happy whatever.

I must admit im seeing a repeat of what happened over the last 2 years and i would not pay what they are asking for, no matter whos brand is on the card, and its expected that a new gen mid range to match an old gen high end card but not with the high end price tag.

But because performance jumps have been slowing down for quite some time, so when a mid range matched high end people pay the high end price and the real high end gets priced even higher.

win8.1 · 28 Sep 2014 at 18:26

Ayahuasca said:
Well you're not though are you, even without the 970/980 nvidia would not have released a different, better chip for £260. Just like AMD's new top end card won't be <£300.

Amd's 290 was a top end chip (just not the very top) with 5% less performance than the 290x it was launched for little more than £300. It will be the same with the 390. Of course I know Nvidia won't launch a card 5% less than their very top end card of their second offering of Maxwell at Amd's bang for buck price, we are talking about Nvidia here after all.