AMD R9 290 Graphics Card Review

Today we get to see the last of the R9 lineup – the AMD R9 290. To the surprise of many, AMD is positioning this card against the recently-price-reduced GTX 780. Can it hold up against NVIDIA’s cut down GK110 GPU? The R9 290X sure did, so we have high hopes for the R9 290. Let’s see what it has to offer, shall we?

Wrapping Up The R9 Launch – AMD’s R9 290

AMD is finishing up the launch of its new lineup with the R9 290. It’s not the Big Dog; that was launched last week. This one is the big dog’s slightly littler brother. The R9 290 has 2,560 shaders as compared to the R9 290Xs 2,816 and its top clock is reduced to 947 MHz (from 1,000 MHz).

Aside from those two changes, you’re looking at pretty much identical GPUs. Both share the updated GCN core, both have TrueAudio capabilities and both will benefit from Mantle when it comes out (as will all GCN-equipped cards). Also present is the impressively wide 512-bit memory bus and 4 GB GDDR5 frame buffer.

Testing The AMD R9 290’s Boost Clocks

As mentioned in the R9 290 series launch article, boost frequencies on the R9 290 series are now similar to how NVIDIA handles boost. The GPU is made to stay within its power target and within its temperature target. With the R9 290, there isn’t a problem keeping the GPU at stock in its power range. Temperature is where you may run into some throttling.

Where it gets interesting for the R9 290 is that AMD was originally positioning this card against the GTX 770, which it does a very good job of trouncing. With NVIDIA dropping the bottom out of the GTX 780’s price to $499 from $649, that changed things a little bit. AMD decided they wanted to position the R9 290 against the GTX 780 instead. This is where it gets good. You see, AMD’s stock BIOS sets the card’s top fan speed at 40%, which is nice and quiet. However, it leaves a good bit of performance on the table because the GPU will reach its temperature target (95 °C) and have to throttle back to keep temps under the target.

After NVIDIA dropped their prices, AMD came out with a new driver and had reviewer’s re-test their GPU with the new driver. The fun part is that the only thing the new driver changed was the stock fan speed cap. It went from 40% to 47%. This allowed the GPU to use more fan speed to keep itself away from its temperature target, thus keeping the GPU running at a higher frequency for longer. That is what is borne out in the graph below. The red line is with the R9 290’s max fan speed at 40% and the blue line is with it at 47%.

Interestingly, it is able to kick the fan in and bring that frequency right back up to 947 MHz. it then drops the fan speed until it is needed again. Personally, I think they should have done this from the start. Why on earth you would leave performance on the table over slightly quieter operation. This is a blower cooler, so I understand wanting to keep it a bit quieter, but 7% just seems silly. That difference, just at the tail end of the Heaven benchmark, accounted for about a 120 point difference in the score. It also accounts for FPS in games. Not many FPS in our testing (typically 1-3 FPS), but that really depends on who was doing the testing and where.

This is a weird tangent, but worth mentioning. It’s fall here in North Carolina, and temperatures are blissfully low. When I was testing this card the first time, ambient temperature in our office was ~20-21 °C, so even with 40% max fan, the GPU stayed away from its temp target pretty well. So, when they raised the max fan speed to 47%, it made a difference, but not huge. I would expect other review sites that are in warmer climes (or that turn their heat on earlier than we do in the season) probably saw a bit more of a performance differential between the two drivers.

The best part about all of this is that when you see board partners come out with better cooling solutions than these reference blower coolers, you’re virtually guaranteed operation at the full 947 MHz boost all the time, and with near silence. Of course, you can operate at 947 MHz all the time on the reference cooler too; just raise the max fan speed up a little more (55-60% or so) and you’ll always keep the GPU away from the temp target, thus ensuring that no throttling will occur.

AMD’s Internal Testing

Of course, AMD ran their own performance testing as they always do, so we’ll show you what they came up with before moving on to our own testing. As mentioned, this card is now positioned to compete with the GTX 780 and AMD shows it doing a respectable job against the competition.

Because of the wide memory bus and extra memory (plus the R9 290’s processing power of course), this card should be a beast at high resolutions. In this test of Battlefield 4, AMD pits the R9 290 at 4K versus the GTX 780 running surround 5760×1080. If we do the math, that’s 33% more pixels – and the R9 290 is darn near keeping up with the GTX 780.

When you go and run the GTX 780 also at 4K resolution, the R9 290 runs away with it. I have issues with the 4K push, as outlined previously, mostly because the vast majority of gamers won’t have a 4K resolution monitor before this GPU is long off the market. However, it paints the R9 290 in a very positive light, which I’m quite sure is why AMD likes to show off its UltraHD performance differential.

Crossfire scaling looks great from AMD’s testing. Across a wide range of games, scaling goes from about 1.8x to very near full 2.0x scaling in a couple titles.

Enough of AMD’s numbers though. Let’s meet the R9 290 in the flesh.

Meet The AMD R9 290

Aesthetically, you won’t be able to tell a reference R9 290 from a reference R9 290X; you need to look at the sticker on the back (or, you know, the box yours comes in).

I actually like the reference cooler’s looks. I don’t really like any blower-style cooler’s noise (other than the one NVIDIA came out with for the TITAN / 780 / 770, which is amazing for a blower cooler), but at least it looks nice.

There are three small intakes for the fan to pull air from the rear as well as from underneath the fan. These seem more for the aforementioned aesthetics than functionality, but hey, every little bit helps.

The R9 290 has two power connectors, in an 8-pin + 6-pin PCIe configuration.

The R9 290 follows in its big brother’s footsteps and eschews the old Crossfire bridge in favor of DMA Crossfire, which shares its data through the PCIe bus.

Also present is the dual BIOS switch, however I couldn’t discern any difference between the two states on this card. On the R9 290X, it functioned to put the card in “Uber” mode, which apparently just changed the max fan speed from 40% to 55%. With the new PowerTune technology, that enabled greater performance (by keeping the card cooler) at the expense of added noise.

When I flipped the switch, nothing happened on this card. Clocks stayed the same and, at least in CCC, the max fan speed stayed at 47%. As there was no discernible change, I left it on the “Uber” side (closest to the video outputs) and benched there for stock numbers.

Speaking of video outputs, this card has the same new configuration as well, with two DVI, one HDMI and one full size DisplayPort outputs. You can use any of these connectors for any three monitors. You can even use up to six monitors if you use a DisplayPort splitter. On the HD 7970, I could never use both DVI outputs, having instead to use one DVI and two mini-DisplayPort outputs. I’m happy to report that wasn’t the case here and you really can use any of the three outputs you feel like using.

Time to pull that cooler off.

Under the Hood

We’ll start by pulling off that plastic shroud. Once removed, you can see the card uses a vapor chamber cooler on the GPU itself, with tight fin spacing, which is part of the reason this cooler is so loud.

I’m also very happy to report that the cooler makes contact with every RAM chip and every MOSFET on the card.

Contact on all those chips was good and they should all be adequately cooled.

Now we meet the monster itself, the AMD R9 290 in the flesh. It’s a great looking PCB, especially for a reference model. I just love seeing all those RAM chips on there. They had to come up with some interesting placement to fit four gigs around that GPU.

The R9 290 GPU itself is definitely larger than its predecessor.

The GDDR5 that comes on the R9 290 is the same SKhynix memory that appeared on the R9 290X.

Also the same is the R9 290’s power section. There appear to be five power phases for the GPU, one for the vRAM and a PLL phase toward the video outputs. This is the only less-than-great spot on an otherwise solid PCB; it could use a slightly more robust power section.

Here’s the BIOS switch and crossfire connectors (well, lack thereof) without the cooler in the way.

Well, that’s the AMD R9 290. It definitely has more in common with the R9 290X than it doesn’t.

Test Setup

Our test setup is one you’ve come to know and love since the new Haswell platform came out this summer. It consists of an i7 4770K operating at 4.0 GHz with RAM at DDR3-1966 / 9-9-9-24. There is plenty of competition for you today, from the top of NVIDIA’s line to AMD’s last generation flagship.

It definitely looks nice all nestled cozy on the test bench.

Overclocking

Overclocking the R9 290 series cards is different than previous generations. AMD says you “no longer have to” set the frequency of a card. Instead, what you now set are percentages. You set the percentage of the power limit you want to give to the card (+20% should be more than adequate for anything you can reach on ambient cooling), then in the same graph, you set the percentage of GPU overclock you want to apply. Unfortunately, that doesn’t give you any actual number, so you’ll have to break out your calculator to see where you’re going, or just run GPU-Z real quick to see the actual frequency you set. To me, this is an annoyance. Frankly, I don’t want AMD to say that I no longer “have to” set the frequency. Newsflash AMD – those of us that overclock want to be able to set the frequency! This is not, I repeat – NOT – a helpful thing.

Scrolling on down, you can see that we aren’t allowed to (excuse me, don’t have to) set memory frequency directly either. The only direct value we have access to is the temperature target. I guess they couldn’t figure out how to positively spin putting temperature as a percentage, so they left that one alone. Fan speed as a percentage is normal and expected.

Where we ended up with our sample was a +14% overclock on the GPU and a +12% overclock on the memory, which, for normal people is 1080 MHz on the GPU and 1400 MHz on the memory.

That’s not a bad overclock at all. Setting the fan speed to a louder, but stronger target allowed the card to boost to its max all the time without issue.

Temperature & Power Consumption

Temperatures on the R9 290 aren’t great, which is exactly how it was designed. The card will not harm itself operating at high temperature and it will keep itself quiet thanks to the stock fan speed cap of 47%. If you prefer to keep your card cooler rather than quieter, you can raise the fan speed cap and lower the temperature target, it’s completely up to you. This is just how the stock BIOS is supposed to keep the card. Thus, this is less an indication of cooler performance and more just showing that the card does what it’s programmed to do.

Power consumption on the R9 290 isn’t surprising. It’s a very built-up Graphics Core Next GPU on the same process as last generation. When you do that, you raise power consumption. The R9 290 draws a little bit more power than the GTX 780, but not much. As long as it performs well enough to justify power consumption in that same neighborhood, you won’t hear a complaint about this card’s power pull from us.

Performance Results

Performance is measured according to our GPU testing procedures, as outlined in our article by that name. Long story short: benchmarks are run at their default settings and games are run with every type of eye candy (MSAA, detail, etc) cranked to the max at 1080p, which is where most gamers will be doing their gaming.

Synthetic Benchmarks

Starting off with a bang, the R9 290 comes out strong. In this older DirectX 10 benchmark, the GTX 780 comes out slightly ahead with the R9 290 at stock, then the R9 290 takes the lead when overclocked. The GTX 770 that AMD used to be positioning this GPU against? Yea, it’s way behind, so the re-positioning seems to be on target from the get-go.

When you start benchmarking with DirectX 11, the picture gets much more rosy for AMD. Only the TITAN beats the R9 290 at stock and even it falls behind after overclocking the 290.

The story repeats itself with Fire Strike, beating the GTX 780 at stock and beating the TITAN overclocked.

If three’s a trend, then consider benchmark results trending in AMD’s direction. The win over the GTX 780 here isn’t huge, but still over 100 points. After being overclocked it becomes a big win over the stock TITAN.

Well then. AMD was positioning this card against the GTX 770. As you can see, from a benchmarking perspective it’s neck and neck with the GTX 780 so repositioning it there makes perfect sense. The R9 290 can even consistently beat TITAN when overclocked. When benchmarking these, I was pleasantly surprised and I’m sure you are too, but you’re probably thinking, like I was, that benchmarks are only part of the story. Gaming is where the rubber meets the road.

Game Testing

Coming out swinging, the stock R9 290 beats even TITAN, simply extending its lead when overclocked.

The R9 290 can’t beat NVIDIA’s pair of GK110 GPUs in Batman: Arkham City, with both of them coming out ahead. Only overclocked does the R9 290 barely edge out the GTX 780.

Battlefield 3 restores what appeared in benchmarks to be the natural order.

GK110 comes out ahead in Civilization V, but the R9 290 does a solid job against the GTX 770.

Dirt 3 goes AMD’s way, with even TITAN unable to compete with the R9 290.

Natural order is restored again for both Metro 2033 and Crysis 3 – GTX 780 < R9 290 < TITAN < overclocked R9 290.

1080p gaming is much like benchmarking. Overall the R9 290 is better than the GTX 780, sometimes by a fair bit, but to be fair, on occasion the two are close. Most interesting was that the R9 290 can consistently beat a stock GTX TITAN when overclocked. That’s huge, especially considering where this card should be priced.

Before moving on to high resolution testing, I wanted to share a few more games with you guys. We’re in the process of modernizing our gaming suite. The database is small but growing for these (and a couple more) games, so comparisons aren’t worth graphing right now. However, since they were run, I thought I’d share what sort of FPS you can expect out of these more recent games with you, both stock and overclocked. The biggest part of this is Battlefield 4, which just launched last week. The R9 290 has excellent FPS using the Ultra preset.

AMD Eyefinity / NVIDIA Surround

AMD really pushed 4K gaming with the R9 290 series launch, and for good reason; the R9 290X performed great there. We don’t have any 4K monitors (should I mention again the only 4K monitor on Newegg costs $3,500?), but we DO have Eyefinity and Surround we can test, to the tune of 5760 x 1080 resolution.

Wow! It’s no wonder AMD pushed 4K so much. At these high resolutions, with its 4 GB frame buffer operating on a 512-bit wide bus, the stock R9 290 not only takes out the GTX 780 handily, it beats the TITAN…in Every. Game. Overclocked just makes matters worse for NVIDIA.

Final Thoughts & Conclusion

Let’s see, hmm…what’s missing from this review? Oh, right, price! I intentionally didn’t tell you the price before so you would have to see how it performed before we got to it. So, what does the AMD R9 290 cost? $399 MSRP. That’s right, there is GTX 780-level performance available for another $100 less than the GTX 780’s current price. Let’s not forget that’s after the original price drop due to the R9 290X out-performing the GTX 780.

The performance of this card was great, impressing me start to finish but when AMD emailed the final pricing, I was blown away. I had expected $449, but $399 re-writes the landscape. Again.

Are there drawbacks? Sure there are. For starters, it isn’t exactly low in power consumption; the GTX 780 beats it there. Additionally, the reference cooler on these things isn’t anything to call home about. It looks good, but it’s noisy. At over about 60-65% fan speed it’s annoyingly so; and you’ll need that cooling ability to get the most out of your card. Hopefully board partners will come out with non-reference cooling solutions sooner rather than later. For you water coolers, EK Water Blocks already has a block out for the R9 290X and the PCB looks so similar they are likely interchangeable (verify with EK first!).

If you’re concerned with noise and you don’t have a low-ambient room to put this card in for gaming, you may want to wait for partner cards. If you can tolerate a bit of extra fan noise – and let’s face it, you know what you’re getting into with any blower cooler other than NVIDIA’s on the high end GTX line – this card will tear through some games at high FPS.

Cooler notwithstanding, the fact remains that the AMD R9 290 can meet or beat a GTX 780 for $100 less. That’s not chump change. This is a great price on a very well performing card. Add solid overclocking as well as a 512-bit memory bus with 4 GB frame buffer, and the R9 290 is sure looking like a winner from AMD. I daresay, at the high end there isn’t a better price-to-performance ratio card on the market right now.

– Jeremy Vaughan (hokiealumnus)