[WCCF] AMD Radeon R9 390X Pictured

[antihelten]

Fixed.

Of course, but under no circumstances does NV marketing want any gamer to compare overall system efficiency in games in terms of perf/watt. That's why they create artificially made up TDPs of 145W/165W for GM204 and why sites like TPU are exclusively focusing in videocard perf/watt, ironic considering no videogame in the world can be played on just the videocard without a motherboard, VRAM, SSD/HDD and CPU present.

NV wants to spread marketing lies how a 970 uses 145W of power because that sounds a heck of a lot better against an R9 290X card with 300W TDP than seeing 40-50W of power usage differences in an i5/i7 gaming rig! Have you not noticed how perf/watt proponents keep using theoretical TDP measurements and refuse to compare overall system efficiency for gaming, like ever?!

Not really sure why you felt the need to "fix" my post, when I quite clearly pointed out the system power versus card power issue in the second paragraph.

Any way I took a look at a wide range of web sites, and there appears to be some rather huge discrepancies in the power gap observed between the 980 and the 290X. Roughly half the sites report approximately a 100W difference, whereas the half reports a 50W difference. No idea what's going on there.

 

[DarkKnightDude]

Tweaktown have spotted a powercolor 390x at computex with hybrid cooler. Looks interesting

http://www.tweaktown.com/news/45669/hands-one-powercolors-next-gen-radeon-r9-390x-video-cards/index.html

Wait, AMD doesn't officially show off the cards yet, but PowerColor is allowed to?

 

[Skurge]

So it's 600mm2+ MINUS the memory controller which eats up a alot of die space. So it could have an effective 700-800mm2?

A man can dream.

Also 2GB is more than enough

 

[3DVagabond]

Tweaktown have spotted a powercolor 390x at computex with hybrid cooler. Looks interesting

http://www.tweaktown.com/news/45669/hands-one-powercolors-next-gen-radeon-r9-390x-video-cards/index.html

Looks like an updated Hawaii PCB. Current card has "LF R29F" this one is "LF R29FA". So either rebrand or updated Hawaii is very likely.

 

[3DVagabond]

Not really sure why you felt the need to "fix" my post, when I quite clearly pointed out the system power versus card power issue in the second paragraph.

Any way I took a look at a wide range of web sites, and there appears to be some rather huge discrepancies in the power gap observed between the 980 and the 290X. Roughly half the sites report approximately a 100W difference, whereas the half reports a 50W difference. No idea what's going on there.

Look at what they used to test with. Without seeing the reviews in question I'll guess that the 100w reviews used Furmark and the 50w ones used a game(s). nVidia more aggressively throttles Furmark than AMD does.

Wait, AMD doesn't officially show off the cards yet, but PowerColor is allowed to?

Tweaktown is known for being irreverent. They were probably sneaked that view. They are based in Taiwan and likely have better sources than the Western press.

 

[guskline]

I sure hope Fiji XT doesn't turn out to be like the Bulldozer release!

 

[stuff_me_good]

Last attempt for real this time, I swear.

Like .vodka said in the other thread, it is really easy to be just a half a mm off and it'll change everything drastically. Also, the method I am using works best when the object is completely perpendicular to the camera, which of course these shots are not.

Oh well, I gave it my best. Do what you want with my measurements...

The only conclusion I am drawing from this is that it is over 500mm².

This one is about perfect. Well defined edges, finally! And it is clear here the shiny parts are the interposer. This is because we know the HBM chips are rectangular (7.29*5.48mm) while the shiny parts are more like squares. No exposed silicon means these HBM chips have the mold (packaging) as described in Hynix's presentation, so these should be the actual measurements for the HBM chips.

This time the plane can be set more clearly and is less extreme, and the clearest base measure is 5.48mm on one side of the HBM chips. Much less guessing now, means more accurate results. And as always photoshop's vanishing point takes into account perspective. I've sharpened up the picture to make the edges even clearer. Let's see:



GPU: 24.01*27.72mm = 665.55mm²
Interposer: 41.48*29.83mm = 1231.38mm²

Rounding as margin of error:

GPU: 24*28mm = 672mm²
Interposer: 41*30mm = 1230mm²


My measurements on WCCFtech's angled picture threw similar estimates. We can all be pretty sure this thing is at least over 600mm² accounting for some big margin of error (a mm or 2 changes everything).

According to these pictures, the most plausible numbers go between 620-670mm^2.

And why I say so? Because hawaii is 438mm^2 and has 2816 shading units. So far we believe that fiji will have 4096 shading units which is 45,45% more than hawaii. So 438 * 1.4545 = 637mm^2 with rough translation.



1. Now here is the thing, according to rumor, GF process should make the chip denser than TSMC process. How much more denser are the GF libraries compared to TSMC? How much die space in mm would be spared on the same die on GF process compared to TSMC?

2. According to rumors also the HBM memory controller is quite a bit smaller than GDDR5 memory controller, ow much die space HBM memory controller saves from die compared to GRRD5?

3. So if the die is 620-650mm^2 and GF process is more denser and the memory controller saves space, shouldn't this mean that the chip could actually have more than 4096 shading units? Maybe the leaks has shown as only the cut down version of fiji aka fiji pro?

 

[Grooveriding]

I sure hope Fiji XT doesn't turn out to be like the Bulldozer release!

I agree, I want to see them deliver. The chip is huge so that looks positive. It needs to smash Titan X to be a success imo. At least 10% faster, preferably 20% in unbiased non Gameworks titles.

If it is slower it's a fail imo. Looks possible to do well with the huge die, HBM memory, 2x8 pin power delivery. I expect it will use more power than a TX, but that is fine so long as it's faster.

 

[happy medium]

I think you guys are setting yourselves up for a letdown. I mean really, trying to say Fiji will be faster because of a picture? Now I've seen it all.

 

[Noctifer616]

I think you guys are setting yourselves up for a letdown. I mean really, trying to say Fiji will be faster because of a picture? Now I've seen it all.

Bigger die means more transistors, more transistors means more performance. Obviously real world performance can't be judged by the die size, but it can be estimated.

 

[3DVagabond]

I sure hope Fiji XT doesn't turn out to be like the Bulldozer release!

Give it a break, man.

 

[guskline]

I agree, I want to see them deliver. The chip is huge so that looks positive. It needs to smash Titan X to be a success imo. At least 10% faster, preferably 20% in unbiased non Gameworks titles.

If it is slower it's a fail imo. Looks possible to do well with the huge die, HBM memory, 2x8 pin power delivery. I expect it will use more power than a TX, but that is fine so long as it's faster.

I agree. I really was let down with the Bulldozer release. No matter how much power you threw at it, the Intel cpus were beating it.

This time around, AMD has been EXTREMELY tight lipped. The new CEO is all serious business so I have hope that this new chip might be competitive.

I give Nvidia credit, they are firing on all cylinders (GTX970/980; Titan X; now GTX980 TI) while AMD stays the course with its dated Hawaii chips being sold off at bargain basement prices. It's amazing that the R9 290X can stay close to the GTX970 considering the age difference in chip design.

Perhaps an improved Hawaii design, that's a bit more efficient will help fight off the dominance of the GTX970 - the real battle ground.

Like most other posters, it will be very interesting to see the results of AMD's work on Fiji XT on 6-16-15.

I'll "soldier on" with my 2 R9 290s! in CF

 

[antihelten]

Look at what they used to test with. Without seeing the reviews in question I'll guess that the 100w reviews used Furmark and the 50w ones used a game(s). nVidia more aggressively throttles Furmark than AMD does.

Nope, all the sites I looked at used games (some of them also used Furmark of course, but I didn't look at those numbers).

50ish Watt group:
Techspot (42W delta - Metro Last Light)
THG (57W delta - says it's a game, but not which one)
computerbase.de (43W delta - Ryse)


100ish Watt group:
Anandtech (107W delta - Crysis 3)
Techreport (96W delta - Crysis 3)
Hardware Canucks (90W delta - Hitman)
TPU (102W - Metro Last Light)

I don't know, maybe I just looked at a peculiar selection of sites.

 

[3DVagabond]

Nope, all the sites I looked at used games (some of them also used Furmark of course, but I didn't look at those numbers).

50ish Watt group:
Techspot (42W delta - Metro Last Light)
THG (57W delta - says it's a game, but not which one)
computerbase.de (43W delta - Ryse)


100ish Watt group:
Anandtech (107W delta - Crysis 3)
Techreport (96W delta - Crysis 3)
Hardware Canucks (90W delta - Hitman)
TPU (102W - Metro Last Light)

I don't know, maybe I just looked at a peculiar selection of sites.

It could be lots of reasons. Possibly different games or different spots in games. Could be bottlenecks. Stock power settings between the companies could limit more severely. I know the MSI gaming 970/980 use more power than the others because the bios allows for higher draw.

 

[OatisCampbell]

Bigger die means more transistors, more transistors means more performance. Obviously real world performance can't be judged by the die size, but it can be estimated.

I don't know that this has ever been true, here's why:

http://www.anandtech.com/show/8427/amd-fx-8370e-cpu-review-vishera-95w

An AMD Vishera and an intel Ivy Bridge CPU both have 1.2B transistors. Would you say they have equal performance?

http://www.anandtech.com/show/8526/nvidia-geforce-gtx-980-review/7

http://www.anandtech.com/show/7457/the-radeon-r9-290x-review

290X has 6.2B transistors, GTX980 has 5.2B transistors, yet GTX980 significantly outperforms 290X.

We need real benchmarks to know if transistor count advantage equals performance advantage.

 

[caswow]

I don't know that this has ever been true, here's why:

http://www.anandtech.com/show/8427/amd-fx-8370e-cpu-review-vishera-95w

An AMD Vishera and an intel Ivy Bridge CPU both have 1.2B transistors. Would you say they have equal performance?

http://www.anandtech.com/show/8526/nvidia-geforce-gtx-980-review/7

http://www.anandtech.com/show/7457/the-radeon-r9-290x-review

290X has 6.2B transistors, GTX980 has 5.2B transistors, yet GTX980 significantly outperforms 290X.

We need real benchmarks to know if transistor count advantage equals performance advantage.

now check dp perf then you will have your answer :thumbsup:

 

[Noctifer616]

We need real benchmarks to know if transistor count advantage equals performance advantage.

Yes, but you don't expect a 300mm^2 chip to outperform a 600mm^2 chip on the same node. Very unlikely. If the two chips are about the same size chances are they could have similar performance or AMD is slightly lower.

Again, just an estimate. Real world performance and power use is very difficult to estimate with HBM on the Fiji due to the impact is has on the die.

 

[3DVagabond]

I don't know that this has ever been true, here's why:

http://www.anandtech.com/show/8427/amd-fx-8370e-cpu-review-vishera-95w

An AMD Vishera and an intel Ivy Bridge CPU both have 1.2B transistors. Would you say they have equal performance?

http://www.anandtech.com/show/8526/nvidia-geforce-gtx-980-review/7

http://www.anandtech.com/show/7457/the-radeon-r9-290x-review

290X has 6.2B transistors, GTX980 has 5.2B transistors, yet GTX980 significantly outperforms 290X.

We need real benchmarks to know if transistor count advantage equals performance advantage.

AMD vs Intel comparisons are dumb when comparing AMD to nVidia. As much as some would like to believe it, nVidia is no Intel. Not even in the same zip code, never mind the same neighborhood.

If you compare Hawaii to GK110, GK110 has ~7.1B transistors. It used to outperform Hawaii in gaming, not in DP, but it doesn't even do that anymore.

Maxwell doesn't have any real DP capability. And it's not just turned off. It's not there at all. There's where you fewer transistors likely went.

 

[Silverforce11]

290X has 6.2B transistors, GTX980 has 5.2B transistors, yet GTX980 significantly outperforms 290X.

While not accurate, its an indication.

438mm2 Hawaii (R290X) = 561mm2 GK110 (780Ti)

NV moves to Maxwell 2.

438mm2 Hawaii is ~10% slower than 398mm2 GM204 (980).

AMD moves to Fiji + HBM...

>600mm2 Fiji vs 601mm2 GM200 = ??

As you well know, AMD has always competed with a smaller die and they come close to matching NV's top GPU for many generations now. The last time they moved first with new vram tech, they stomped on NV hard with a very small GPU that was half the size of NV's.

The performance gap was closed with Hawaii. It shows AMD is able to match NV with a smaller GPU. The next step that is logical, would mean their huge GPU + new vram tech should stomp on NV's huge GPU.

If it doesn't, it's a flop and going backwards in progress.

I mentioned this awhile ago and @RS and some others seem to think its OK for AMD to be slower, as long as they priced it cheaper. NO. Lisa is different, she's fully aware that approach is what has led them to this situation, losing the performance crown and forced to go cheap is NOT a good business strategy. Not for the CPUs and not for GPUs.

It's not OK to go balls to the walls with a massive GPU and new vram tech and be slower.

 

[Creig]

Tweaktown have spotted a powercolor 390x at computex with hybrid cooler. Looks interesting

http://www.tweaktown.com/news/45669/hands-one-powercolors-next-gen-radeon-r9-390x-video-cards/index.html

It looks especially interesting since it only has one 8-pin and one 6-pin power connector.

 

[Skurge]

It looks especially interesting since it only has one 8-pin and one 6-pin power connector.

It Also looks especially hideous.

 

[DarkKnightDude]

That's the rebadged/tweaked Hawaii.

 

[Creig]

It Also looks especially hideous.

I can't disagree there. But at least we know it's staying within the power envelope of current 290X cards.

 

[LTC8K6]

Wait, AMD doesn't officially show off the cards yet, but PowerColor is allowed to?

It's just an R9-290X with a hybrid cooler.

We have seen them before, no reason not to show it.

 

[antihelten]

It could be lots of reasons. Possibly different games or different spots in games. Could be bottlenecks. Stock power settings between the companies could limit more severely. I know the MSI gaming 970/980 use more power than the others because the bios allows for higher draw.

Difference in games seems unlikely, since two sites using the same game (metro last light) sees 42 and 102 Watts respectively. Also all of the cards should be reference (as far as I can tell), so there shouldn't be that big of a difference in bios settings.

Any way I'm not saying there's anything of note here, I just stumbled over it and found it mildly interesting.