videocardzFirst AMD Radeon R9 290X 1080p performance review

[boxleitnerb]

That even worse for nV cause they still don't have a fully functioning card on the desktop. I wouldn't call Titan or GTX780 conservatively clocked seeing as ref models boost to 1ghz. 7950 is clocked low.

Hawaii is a lot more impressive than GK110 IMO, we just need to see power numbers.

There is a difference between fully unlocked cards not being on the market and cards not existing at all. Why would Nvidia sell 15 SMX GK110s rather than 12 SMX when the former are more expensive to make?

Boost clocks are not sustained, though. Base clocks are low, they could be 10% higher if you ask me.

Here we go about the statements of Titan being a year old now.

I'm glad 1 year is 8 months long now. I don't have to wait as long for product refreshes.

GK110 tapeout: March 2012
Hawaii tapeout: about April 2013 (look at sushi's posts over at XS forums, there was mention of samples floating about around that time)

 

[RussianSensation]

You mean a 561mm2 chip that has 960 dedicated DP units that use about 50mm2, that is quite conservatively clocked and has 20% fused off and on top is about a year older? Shocking!

I rather think this comment is embarrasing for you because you didn't do all your research. We don't need flames here at AT forums, and that is exactly what your comment is. Post reported.

Go ahead, report my post all day long. Please point out anything I said that was offensive. If anything I called SirPauly on constantly spilling marketing-style responses and never addressing anything that is being discussed in detail. If you got a problem with that, please please report me all day. I really could care less. Truth of the matter is that you have no subjective defense at all, which is why you go crying to the mods. Typical.

Anyway, talking about 960 dedicated DP units, funny comment after HD4870/4890/5870/6970/7970 laid waste to 280/285/480/580/680 in DP performance. Of course you knew that already but as a last resort you bring out GK110's DP performance as its saving grace, and in doing so assuming that R9 290X is DP-crippled. Ironic after the crippled 280/285/480/580/680s. Ya.

I like to stick to the subject matter based on facts, not pick an choose metrics that suit my brand of choice. You go crying to the mods whenever you have nothing to say as a rebuttal. Congratulations. :thumbsup:

 

[boxleitnerb]

Go ahead, report my post all day long. Please point out anything I said offensive. You have no subjective defense at all, which is why you go crying to the mods. Typical.

Anyway, talking about 960 dedicated DP units, funny comment after HD4870/4890/5870/6970/7970 laid waste to 280/285/480/580/680 in DP performance. Of course you knew that already.
And Ya, I am not going to cry to the mods about your BS because unlike you, I am much more mature. :thumbsup:

So you flame and then get personal if you're called on it...very mature indeed. If someone flames or spams or threadcraps, a post should be reported, that's how forum communities work.
In general, your spiteful attitude is not very helpful for a civilized discussion, especially when you omit facts.

If you don't find a rebuttal of your comment in my post that is based on arguments and examples, look again, there are four points there.

@DP:
Operative word being dedicated. You know, those things that take up space and are not useful for anything else besides DP. Perfectly valid argument in the context of this discussion about perf/mm2. Your mentioning earlier AMD cards is irrelevant here since they did *not* have dedicated units.

 

[Skurge]

So you flame and then get personal if you're called on it...very mature indeed. If someone flames or spams or threadcraps, a post should be reported, that's how forum communities work.
In general, such a spiteful attitude is not very helpful for a civilized discussion, especially when you omit facts.

If you don't find a rebuttal of your comment in my post that is based on arguments and examples, look again, there are four points there.

@DP:
Operative word being dedicated. You know, those things that take up space and are not useful for anything else besides DP. Perfectly valid argument in the context of this discussion. Your mentioning earlier AMD cards is irrelevant here since they did *not* have dedicated units.

I'm not sure I understand what you mean in that last statement. I didn't know dedicated units existed. I figure all the SPs perform equally.

 

[caswow]

So you flame and then get personal if you're called on it...very mature indeed. If someone flames or spams or threadcraps, a post should be reported, that's how forum communities work.
In general, your spiteful attitude is not very helpful for a civilized discussion, especially when you omit facts.

If you don't find a rebuttal of your comment in my post that is based on arguments and examples, look again, there are four points there.

@DP:
Operative word being dedicated. You know, those things that take up space and are not useful for anything else besides DP. Perfectly valid argument in the context of this discussion. Your mentioning earlier AMD cards is irrelevant here since they did *not* have dedicated units.

tahiti gk104 roles reversed aint it

 

[Vesku]

tahiti gk104 roles reversed aint it

It is actually quite an interesting parallel.

 

[boxleitnerb]

I'm not sure I understand what you mean in that last statement. I didn't know dedicated units existed. I figure all the SPs perform equally.

GK110 has 3840 SPs, 960 of which can only do DP calculations and those are separated from the rest (i.e. not tied to TMUs for example). Nvidia did this because apparently (at least with their current implementation) it is better for perf/W. Essentially, they traded perf/mm2 for perf/W. There is an interview with the German IT magazine c't where two NV engineers confirmed this:
http://translate.google.de/translate?sl=auto&tl=en&js=n&prev=_t&hl=de&ie=UTF-8&u=http%3A%2F%2Fwww.heise.de%2Fnewsticker%2Fmeldung%2FGTC-2012-Die-GK110-Schoepfer-ueber-Performance-und-zukuenftige-Herausforderungen-1578605.html&act=url

Googlish:

c't: Sets Nvidia GK110 to dedicated in double-precision units or work for several single-precision cores together? Danksin: There are independent double-precision units.
c't: Take this much space?
Album: At least not enough. This is a GK110-SMX significantly more die area than those of GK104. Another space eater is the implementation of the ECC functions.

Some knowledgeable people over at 3DCenter calculated the space requirements for these units at about 50mm2. Maybe I can find the link, but this discussion was quite prolongued, might be difficult to find.

Iirc, even GT200(b) and Fermi had dedicated DP units, but I could be mistaken.

 

[Imouto]

You should notice already that it's the very same excuse that AMD loyalists sow at Kepler launch.

2012 AMD zealot: "But Tahiti's die is bigger because it has compute capabilities!"
2012 Nv zealot: "OMG what's with that horrible gaming performance per square mm?"

2013 Nv zealot: "But GK110's die is bigger because it has dedicated compute capabilities!"
2013 AMD zealot: "OMG what's with that horrible gaming performance per square mm?"

 

[Haserath]

AMD takes a hit in mm^2 and perf/power to provide DP performance through the SP units, but it must use less area than a dedicated implementation.

AMD's customers aren't big servers/HPCs, so that doesn't really leave anyone to care about perf/power.

 

[BallaTheFeared]

You should notice already that it's the very same excuse that AMD loyalists sow at Kepler launch.

2012 AMD zealot: "But Tahiti's die is bigger because it has compute capabilities!"
2012 Nv zealot: "OMG what's with that horrible gaming performance per square mm?"

2013 Nv zealot: "But GK110's die is bigger because it has dedicated compute capabilities!"
2013 AMD zealot: "OMG what's with that horrible gaming performance per square mm?"

:thumbsup:

Around and around we go.

 

[blackened23]

You should notice already that it's the very same excuse that AMD loyalists sow at Kepler launch.

2012 AMD zealot: "But Tahiti's die is bigger because it has compute capabilities!"
2012 Nv zealot: "OMG what's with that horrible gaming performance per square mm?"

2013 Nv zealot: "But GK110's die is bigger because it has dedicated compute capabilities!"
2013 AMD zealot: "OMG what's with that horrible gaming performance per square mm?"

The circular arguments have come....full circle

But all this talk of DP and my thoughts are...who the hell cares. All I want to know is games performance and OC ability. Mostly the latter since we already have a rough idea that it is trading with Titan and ~10% faster than the 780. Oh. And price. It better be sub 600$.

 

[Erenhardt]

The circular arguments have come....full circle

But all this talk of DP and my thoughts are...who the hell cares. All I want to know is games performance and OC ability. Mostly the latter since we already have a rough idea that it is trading with Titan and ~10% faster than the 780. Oh. And price. It better be sub 600$.

Logic:
It is trading with $1000 card and is faster than $650 by quite a bit... it better be cheaper then the slower card...
Man, you almost got me!

If just for a second we forget about brands here. Every card there is, is done by one manufacturer. He sells 2 cards:
A Titan for $1000
B 780 for $650
He wants to release a new graphics card that performs the same as card A, but has integrated sound co-processor. How should he price the new product?

 

[SirPauly]

General statement that has nothing to do with what's being discussed in detail wrt to R9 290X vs. 780's power usage.

Personally allow the moderators to decide not vocal posters!

 

[sushiwarrior]

GK110 tapeout: March 2012
Hawaii tapeout: about April 2013 (look at sushi's posts over at XS forums, there was mention of samples floating about around that time)

That wasn't tapeout, that was samples returning from TSMC.

As far as the performance/mm^2 thing, considering Tahiti was already almost at Titan levels for FP64, Hawaii is adding SP's and getting 1:2 DP performance. It's going to demolish it there. However compute will never be that comparable, since a lot of it depends on architectural features that are better for certain tasks. Hawaii will dominate the FLOPS contest though, but that is no real world indicator of performance.

 

[Rvenger]

:thumbsup:

Around and around we go.

Yep :thumbsup: Its like you said. Silly Season!

 

[Rvenger]

The circular arguments have come....full circle

But all this talk of DP and my thoughts are...who the hell cares. All I want to know is games performance and OC ability. Mostly the latter since we already have a rough idea that it is trading with Titan and ~10% faster than the 780. Oh. And price. It better be sub 600$.

If it's not performance then it will shift to power consumption. Then if it's not power consumption it will shift to 'Smoothness'. If its not smoothness then it will shift to performance per watt. If its not performance per watt it will shift to performance per dollar. If its not performance per dollar it will shift to performance per sq mm. My head is about to explode.

 

[blackened23]

If it's not performance then it will shift to power consumption. Then if it's not power consumption it will shift to 'Smoothness'. If its not smoothness then it will shift to performance per watt. If its not performance per watt it will shift to performance per dollar. If its not performance per dollar it will shift to performance per sq mm. My head is about to explode.

Basically! Essentially, those on green cool aid who would never be objective will maintain green cool aid status. Agreed.

Personally this GPU boils down to price to me. If it's sub 600$, I consider that a great GPU. 700$ and I won't. I really don't see anyone can complain if it's sub 600$ while trading with Titan....

 

[Silverforce11]

For AMD that is unlikely if it's priced at $500.

7950 never had much of an impact until it was well below the price of the 670 around $300.

Uhh this is blatantly false. 7950 impacted significantly from launch, snapping the heels of its bigger brother for $100 less, and a massive OCer. About 2 months after the launch, I got mine for $339, at the time, a gtx670 was around $420 and 680 was $550. I was tossing between the 670 and 7950 (early on AMD didn't have their turbo drivers yet, requiring big OC to perform better than "equivalent" NV cards) but decided to get a 7950 and OC it, which is a silicon gamble but I had a good sample.

Recently I grabbed this 670 when it was on sale for $299. At the time, I was tossing whether I shouldn't just grab another 7950, they were still around $300.. in hindsight, I should have, because OC vs OC, the 7950 blow the 670 away.

Now, few would care about the R290X, but since the 290 itself has non-crippled TMU/ROPs and bus, OC it should be a monster for much cheaper.

In summary, its the R290 that is the exciting part. Looking forward to seeing how well it OCs.

 

[BallaTheFeared]

It didn't sell, that was my point.

With GK104 Nvidia took market share, with GK110 Nvidia made record revenue.

28nm has been one heck of a run for Nvidia with AMD trying to play catch up with their driver issues, from less than expected performance, to single card stutter, to broken crossfire, it's just been a mess really.

 

[bystander36]

It didn't sell, that was my point.

With GK104 Nvidia took market share, with GK110 Nvidia made record revenue.

28nm has been one heck of a run for Nvidia with AMD trying to play catch up with their driver issues, from less than expected performance, to single card stutter, to broken crossfire, it's just been a mess really.

Now that AMD has had all their issues exposed, and they finally started addressing these issues, their next generation could be a real winner.

I am not holding my breath on Mantle, I think it's being way over hyped, but with so many of their big issues finally being addressed, they may no longer have the problems that prevent a lot of people from using their stuff.

 

[sushiwarrior]

It didn't sell, that was my point.

With GK104 Nvidia took market share, with GK110 Nvidia made record revenue.

28nm has been one heck of a run for Nvidia with AMD trying to play catch up with their driver issues, from less than expected performance, to single card stutter, to broken crossfire, it's just been a mess really.

I can't remember the exact share numbers, but I am quite sure that AMD gained quite a bit of market share over the 6xxx generation.

 

[Vesku]

It didn't sell, that was my point.

With GK104 Nvidia took market share, with GK110 Nvidia made record revenue.

28nm has been one heck of a run for Nvidia with AMD trying to play catch up with their driver issues, from less than expected performance, to single card stutter, to broken crossfire, it's just been a mess really.

My 7950 had no performance problems from the time I bought it to now. Multi-gpu (lackluster crossfire support) sales are a pretty small part of overall marketshare so, IMO, Nvidia marketing did its job yet again and saved Nvidia from a real price war.

Looks like things might actually change in the consumers favor with this R9-290 series assuming AMD prices it at $599 or less. Getting the feeling from internet posts and PR that Nvidia's marketing team is starting to sweat.

 

[3DVagabond]

Don't forget it actually does scientific HPC not just kiddie HPC with Hyper-Q and Dynamic Parallelism.

Not without ECC RAM it doesn't. Errors would creep in to the simulations and make them worthless.

 

[3DVagabond]

If it's not performance then it will shift to power consumption. Then if it's not power consumption it will shift to 'Smoothness'. If its not smoothness then it will shift to performance per watt. If its not performance per watt it will shift to performance per dollar. If its not performance per dollar it will shift to performance per sq mm. My head is about to explode.

And if all else fails...

It didn't sell, that was my point.

With GK104 Nvidia took market share, with GK110 Nvidia made record revenue.

28nm has been one heck of a run for Nvidia with AMD trying to play catch up with their driver issues, from less than expected performance, to single card stutter, to broken crossfire, it's just been a mess really.

...drivers and market share.

 

[wand3r3r]

If it's not performance then it will shift to power consumption. Then if it's not power consumption it will shift to 'Smoothness'. If its not smoothness then it will shift to performance per watt. If its not performance per watt it will shift to performance per dollar. If its not performance per dollar it will shift to performance per sq mm. My head is about to explode.

Indeed and it's annoying to watch.

On topic, where's the 290? It'd be nice to see how it is performing relative to the 290x as it will likely be the more popular of the two.