[Tom's Hardware] R9 290X 4GB vs. 8GB at 4K - why we need 6-8GB GPUs

[RussianSensation]

While the cost increase over 4GB versions is probably not going to be worthwhile, there seems to be a definite benefit to 8GB gaming at 4K. Keep in mind since Tom's didn't even test dual or triple 290s in CF with SSAA/MSAA at 4K, the benefit could be even greater if you have enough GPU power to crank the AA settings. I hope AMD and NV release 390X/GM200 with 6-8GB of VRAM at minimum next year.

Side-note: Sapphire's Vapor-X 290X actually runs cooler and quieter than EVGA's GTX970 SSC. Shows how weak the ACX 2.0 cooler is given that it has to cool the lower power consuming GM204.

Doubling the amount of VRAM adds a decent amount of power consumption.

Full Review

 

[sm625]

But doesnt 4 GB of GDDR5 cost jsut as much as stepping up to the next higher gpu compute core configuration?

 

[96Firebird]

Maybe I missed it in the review, did they ensure the reference R9 290X wasn't throttling during their testing?

 

[ShintaiDK]

Reference blower 290X to compare with an aftermarket cooler? Not exactly the best card to use as a 4GB counterpart. 95C and throttle.

I would like to see a proper cooled 4GB 290X before making any conclusions.

 

[Eric1987]

Reference blower 290X to compare with an aftermarket cooler? Not exactly the best card to use as a 4GB counterpart. 95C and throttle.

I would like to see a proper cooled 4GB 290X before making any conclusions.

I think thats pretty pointless. That 290x with 8gb is just as fast if not faster than the 980 at 4k. And thats better than a 290x reference.

 

[Insomniator]

Dell is releasing a 5K monitor soon, do we need 10GB cards?

High resolution as 'next gen' is such BS. They just want to sell us new monitors and new cards while using the same graphics from 5 years ago.

 

[Termie]

Maybe I missed it in the review, did they ensure the reference R9 290X wasn't throttling during their testing?

No they did not. In fact, this review is completely flawed, as they state on the second page that the blower cooler might be limiting performance by throttling:

We're interested to see what the extra RAM provides for the Radeon R9 290X when it comes to game performance. Sapphire's Vapor-X is factory overclocked, though, so in order to isolate the difference we need to set our reference Radeon R9 290X to the same 1030 MHz core and 1375 MHz memory clocks. Because of this, any performance delta between the two should be a result of the extra memory (although Sapphire's improved cooler will probably affect the results, too, by allowing the GPU to stay in a boosted state for longer periods).

When whoever was editing this article saw this caveat, they should have pulled the article. I'm frankly shocked Tom's bothered to publish this. It took a lot of time to write, I'm sure, but simply pushing the fans to 100% on that reference 290X would have made the article worth reading.

And Russian - as for your sidenote on heat and noise, you should know quite well that the Vapor-X is a 12"-long triple-fan cooler, and the fact that it runs cooler and quieter than the 9.5"-long dual-fan SC cooler is not much of an accomplishment. The negative reaction to EVGA's ACX 2.0 has been completely overdone. It's an incredibly-compact cooler, and the fastest card to ever be released at under 10" long.

 

[Eric1987]

Dell is releasing a 5K monitor soon, do we need 10GB cards?

High resolution as 'next gen' is such BS. They just want to sell us new monitors and new cards while using the same graphics from 5 years ago.

You're joking right?

 

[Alatar]

Tom's reference 290X must be terrible even when overclocked. It's being clearly beaten at 4K by a 970.

I hope AMD and NV release 390X/GM200 with 6-8GB of VRAM at minimum next year.

GM200 is most likely 384bit. Which would most (almost 100%) likely translate to 6GB for the normal GTX cards and 12GB for the inevitable Titan.

390X is more interesting. As far as I know in order to get over 4GB before the 2nd generation of HBM gets here you need to run GDDR5. And 2nd generation of HBM should not get here before some time in 2016 which is when NV will likely launch Pascal. So GDDR5 for the 390X? Or HBM with less capacity?

 

[el etro]

I don't know if is really good going for 8GB cards to play at any resolution today, AFAIK the only two games that needs this crazy amounts of Vram is both Watch Dogs and Shadow of Mordor(and only these two).
The question of the 290x 4Gb throttling pointed by other guys on topic is right IMO, maybe the VaporX card is only having better framerates because of clock speed inconsistency of the reference 290x.


Someone that have contact with HardOcp guys should ask to them to do the same test.


LOD gone up so much since crysis1 back in 2007, but is clear that the Vram madness is going up so much these times.
Vram needs are growing up more than the games level of detail last years, and it is not a good thing. I prefer HBM coming soon to the GPUs rather than this...

 

[Cerb]

Tom's reference 290X must be terrible even when overclocked. It's being clearly beaten at 4K by a 970.

Most reference 290s and 290Xs have been like that, outside of really good binned chips. Due to the throttling ability, and to save money, AMD specified a pretty poor cooler.

http://www.tomshardware.com/reviews/sapphire-vapor-x-r9-290x-8gb,3977-5.html
Just being blower style is not a miracle property. The proprietary coolers have more to do with non-ref cards performing well than their factory OCs.

 

[caswow]

one must be suicidal buying any ref. cooled gpu. thats why we have good aftermarked cooling solutions.

 

[toyota]

one must be suicidal buying any AMD ref. cooled gpu. thats why we have good aftermarked cooling solutions.

FTFY

 

[96Firebird]

No they did not. In fact, this review is completely flawed, as they state on the second page that the blower cooler might be limiting performance by throttling:

I agree. Looking at the temperature graph, the reference 290X went right up to 94°C and stayed there. I have no doubts it was throttling, especially since it was overclocked. I don't see why they couldn't have used a 290X 4GB Vapor-X, unless they didn't have one on hand. In that case, wait until you get one or use another aftermarket card and match the clocks.

I don't think this article proves the need for over 4GB @ 4K...

Edit - I like the initiative Tom's has been taking in testing off the beaten path, but they really need to get their testing controls in check before publishing their articles.

 

[Rvenger]

I agree. Looking at the temperature graph, the reference 290X went right up to 94°C and stayed there. I have no doubts it was throttling, especially since it was overclocked. I don't see why they couldn't have used a 290X 4GB Vapor-X, unless they didn't have one on hand. In that case, wait until you get one or use another aftermarket card and match the clocks.

I don't think this article proves the need for over 4GB @ 4K...

Edit - I like the initiative Tom's has been taking in testing off the beaten path, but they really need to get their testing controls in check before publishing their articles.

Exactly. They should have removed the fan cap on the reference card. It would have been insanely loud but still wouldn't have throttled. Now this review doesn't prove anything. It just shows that an aftermarket overclocked 290x is faster than a throttled reference 290x.

 

[ViRGE]

Exactly. They should have removed the fan cap on the reference card. It would have been insanely loud but still wouldn't have throttled. Now this review doesn't prove anything. It just shows that an aftermarket overclocked 290x is faster than a throttled reference 290x.

Indeed. The fact that there's even a built-in switch for this (the uber mode switch) makes the whole thing rather silly.

 

[el etro]

Next thing for powertune control: clockspeeds versus fan speed. Adequate a certain level of performance(clock and memory clock) to a certain fan setting. I say this assuming we will never see again a cooler so loud like this.

 

[f1sherman]

I'll keep the fan speed tied to the temperature as it is right now, ty-very-much.
You think this is wrong?

Cooler loudness is not the problem.
W, W/m^2 and cooler effectiveness is

 

[CakeMonster]

Isn't it possible to lock these cards at a certain clock speed?

It's a real shame the results aren't compatible because I'm really curious about this topic.

 

[el etro]

Power usage for 290x is not so bad. Perf/mm² they done well in Hawaii(compared to GK110 of course). They need to time past now in order to launch the next radeons in 20nm.

 

[SolMiester]

The Vapour X cooler doesnt butt up to the PCI bracket leaving huge holes around the shroud so i would imagine most of the hot air is exhausted inside the case as opposed to the EVGA unit which looks a better fit to the shroud and therefore exhaust most of the air outside the case.
I am therefore not surprised they have a big difference in heat measurement at the back of the case. Cooler?, my a**, the Vapour is drawing 80w more, HTF is it supposed to run cooler?

 

[Silverforce11]

http://hexus.net/tech/reviews/graphics/76685-sapphire-radeon-r9-290x-vapor-x-8gb/?page=9

Entirely different results here. Comparing both custom Sapphire models, with the only difference being vram and most likely binning.

"Here's a surprise. You'd think that Sapphire's overclocked Vapor-X, with 8GB of GDDR5 memory in tow, would consumer more power than a reference Radeon R9 290. That doesn't appear to be the case in our benchmarks, suggesting that AMD's press cards are overvolted."

Reference R290/X are indeed overvolted, mine can run with -50mV vcore and save a lot of power.

 

[96Firebird]

http://hexus.net/tech/reviews/graphics/76685-sapphire-radeon-r9-290x-vapor-x-8gb/?page=9

Entirely different results here. Comparing both custom Sapphire models, with the only difference being vram and most likely binning.

8GB model also has 200MHz higher memory clockspeed...

 

[f1sherman]

That's either lucky sample, or binning.
http://hexus.net/tech/reviews/graphics/76685-sapphire-radeon-r9-290x-vapor-x-8gb/?page=10

AMD is not that incompetent to fry their GPU's at +50mV than what's needed.
My 290 can almost run with -50mV (it doesn't pass ALL workloads, and especially not fluctuating ones). So no, ref. cards are not overvolted.

 

[Silverforce11]

AMD when they released the 7970 Ghz or 7950 Boost reference cards, they went with a flat 1.25vcore which anyone who has used the 7900 series knows was ridiculous.

My custom PCS+ 7950 had a default vcore of 1.087v !!

I also got a reference 7970 for mining and promptly set it to 1.1v and it ran fine, saving a lot of power and less fan speed required.

Never underestimate AMD's incompetence.