Nvidia RTX 2080 Super hands-on: The result when AMD is out of striking distance

[Karlston]

Nvidia RTX 2080 Super hands-on: The result when AMD is out of striking distance

Last year's $799 card is this year's slightly upgraded $699 card, but is it "Super"?

Earlier this month, Nvidia kicked a stool out from under AMD's feet, just as the graphics-card sector began heating up anew. AMD was set to land a serious blow with new RX 5700 cards in the "pricey but reasonable" range—a range that Nvidia had failed to capture with its "entry-level" RTX cards, the 2060 and 2070. Nvidia responded to AMD's news by unveiling and launching a surprise pair of solid "Super" cards. AMD responded with its own price cut (and a claim that this price-war dance was its plan all along).

As these similarly specced cards jostled for the "$400ish" crown, the winner was ultimately consumers. At every price point, new GPU buyers can expect a solid bang-for-buck quotient between the $349 AMD Radeon RX 5700 and the $499 Nvidia RTX 2070 Super.

Weeks later, we have Nvidia's third Super-branded launch, the RTX 2080 Super. And it's a good reminder of what happens when AMD is not in striking distance of a particular price sector.

Say hello to the “2080 sorta Super”

	RTX 2080 Ti Founders Edition	RTX 2080 Super	RTX 2080 Founders Edition	RTX 2070 Super	RTX 2070	RTX 2060 Super	RTX 2060	GTX 1080 Ti
CUDA Cores	4,352	3,072	2,944	2,560	2,304	2,176	1,920	3,584
Texture Units	272	192	184	184	144	136	120	224
ROPs	88	64	64	64	64	64	48	88
Tensor cores	544	384	368	320	288	272	240	n/a
RT cores	68	48	46	40	36	34	30	n/a
Core Clock	1,350MHz	1,650MHz	1,515MHz	1,605MHz	1,410MHz	1,470MHz	1,365MHz	1,480MHz
Boost Clock	1,635MHz	1,815MHz	1,800MHz	1,770MHz	1,620MHz	1,650MHz	1,680MHz	1,582MHz
Memory Bus Width	352-bit	256-bit	256-bit	256-bit	256-bit	256-bit	192-bit	352-bit
Memory Speed	14GHz	15.5GHz	14GHz	14GHz	14GHz	14GHz	14GHz	11GHz
Memory Bandwidth	616GB/s	496GB/s	448GB/s	448GB/s	448GB/s	448GB/s	336GB/s	484GB/s
Memory Size	11GB GDDR6	8GB GDDR6	8GB GDDR6	8GB GDDR6	8GB GDDR6	8GB GDDR6	6GB GDDR6	11GB GDDR5X
TDP	260W	250W	250W	215W	175W	175W	160W	250W
MSRP	$1,199	$699	$799 (discontinued)	$499	$499 (discontinued)	$399	$349	$699 (discontinued)

 

The RTX 2080 Super, launching at retail today, is a $699 replacement for last year's RTX 2080 (which also had a $699 MSRP). The best thing I can say about this card is that it is technically a $699 replacement for last year's slightly more powerful $799 RTX 2080 Founder's Edition—which was identical to the normal RTX 2080, save an overclock. (This week's 2080 Super goes a slight bit above that FE overclock's boost speed, at 1,815MHz versus the FE's 1,800MHz.) Both the 2080 and its FE variant have effectively been replaced by this new Super.

 

But "Super" branding is arguably the worst thing Nvidia could have saddled this card with. In terms of sheer percentages, the RTX 2080 Super doesn't offer the same percentage boosts found with its other Super launches.

 

The above tale-of-the-tape is telling. Look at the RTX 2080 Super and its non-Super sibling, and you'll see a milder uptick in compute units ("CUDA cores" in Nvidia parlance) and texture units compared to other Super upgrades. Its bump in memory bandwidth also pales compared to the same bump on the 2060 Super.

 

And the RTX-specific cores only get a teensy uptick in the RTX 2080 Super: 384 "tensor cores," up from 368 in the non-Super card, and 48 "RT cores," up from 46. That's a 4.16% jump in both categories, as opposed to the RTX 2070 Super getting a 10% boost and the RTX 2060 Super jumping 11.7%.

 

Let's do the numbers:

 

 

Every test above includes the RTX 2080 and RTX 2080 Super in direct comparison, along with a few other options: this year's RTX 2070 Super, last year's $1,199 RTX 2080 Ti Founder's Edition, and the enduring GTX 1080 Ti from Nvidia's Pascal generation. (That card launched with a $699 MSRP, but the price is a bit moot since the 1080 Ti has all but vanished at retail.) Each card was first tested at factory clocks, then given a mild "safe" overclock as determined by EVGA's Precision X1 software.

 

Whether you prefer to overclock your GPU to the edge of its voltage limit or feel like overclocking is a fool's game, we think this split-the-diff number is a good reminder that Nvidia leaves some power to be squeezed from its official models—which commonly trickles down into other vendors' cards.

 

Additionally, these tests all focus on 4K rendering, with each game's or test's settings largely cranked to max, as opposed to 1080p or 1440p. Lower resolutions tend to be CPU-bound, so I prefer to leave no doubt how each GPU handles a higher-resolution burden (especially on older software that arguably goes easier on modern GPUs). But I also try to disable SMAA once I'm up to 4K, with the exception of the older Hitman benchmark.

 

These tests do not include equivalent AMD cards, owing to the fact that I have a pile of Nvidia cards at my home office without AMD options to match. I'll summarize what we know about AMD's latest Navi cards in a bit.

Welcome to Earth

The first big takeaway from these tests is that Nvidia doesn't want anybody second-guessing the value of its $1,199 RTX 2080 Ti. If you had hopes that the RTX 2080 Super would get anywhere near the 2080 Ti's brute-force approach to 4K resolutions, come back to Earth, please. (Weighing in at 11GB GDDR6 VRAM, compared to the other cards' 8GB, doesn't hurt the RTX 2080 Ti one bit.)

 

There's also the curious fact that every card beneath the RTX 2080 Ti has terribly similar RTX test results. As of press time, no RTX test is more intensive than Quake II RTX, owing to its insane number of light and reflection calls; hence, we run the test in 1080p resolution just to get a reasonable frame rate. (Turn off the RTX effects, and the frame rate jumps to over 1,000fps.) My Quake II RTX tests all revolve around focusing on a single, intense moment in terms of global illumination rendering, perhaps owing to a mix of open sky and a large body of reflective water.

 

In emails to Ars Technica, Nvidia's marketing team pointed to its own Quake 2 RTX tests, which run through an entire level with a pre-defined script. The company claimed a 3.7% performance increase for the RTX 2080 Super over the RTX 2080 FE—and that's in line with my own test results of a focused, intense scene. The thing is, if you want to do the RTX dance, the RTX 2070 Super is in reaching distance of those Quake 2 RTX tests... at $200 less.

 

Above all, these tests are a good reminder that Nvidia already knocked GPU power out of the park with one of its stronger Pascal cards: 2017's GTX 1080 Ti, a card that fell below "Titan" status (and pricing) but continues to endure in terms of 4K viability. Had the market not succumbed to crypto-mining fever in 2017, perhaps this exact model would have been Nvidia's focus in terms of strength and efficiency.

 

Instead, Nvidia pivoted with a focus on proprietary ray-tracing systems. The fact that AMD has made its own announcements about ray tracing is, for better or for worse, a sign that Nvidia was bound to make such a shift anyway. But we're still waiting on more reasons to buy into the RTX revolution before insisting that any GTX 1080 Ti owner should throw their powerful GPU into the trash. (Nvidia uses this opportunity to remind shoppers that the RTX 2080 Super comes with free copies of two RTX-compatible games, July's Wolfenstein Youngblood and August's Control, but we've yet to comprehensively test RTX-enabled modes in either game.)

A thought on AMD’s “prescription”

Via third-party reports, we know that AMD's RX 5700 and RX 5700 XT cards, at $349 and $399 respectively, slot perfectly into the pricey-but-reasonable GPU category. With some exceptions, you'll get more brute-force power from a $499 RTX 2070 Super, and that means you're maybe spending $50 more on power and $50 more on an RTX lottery ticket. Or you can skip the RTX lottery altogether and get something near RTX 2060 Super performance for $50 less on the RX 5700 front.

 

What's more, if you're not interested in 4K resolutions, then you're arguably fine sticking with anything beneath the RTX 2080 Super—especially since RTX games for the next year or so will likely require resolutions in the 1080p and 1440p range, based on what we've seen in games like Battlefield V. And, again, the RTX 2070 Super isn't all that far behind the RTX 2080 Super in RTX-specific performance.

 

The $200 you're spending on a jump to the RTX 2080 Super, then, gets you close to a no-fuss, no-muss 4K gaming experience. But as the above charts show, the RTX 2080 Super falls short when games are set to a mix of "high" and "ultra" settings in 4K resolution. That'll mean doing the usual settings dance (Do I turn down shadow resolution? Reflection quality? Resolution scaling?) when opting for a full 3840×2160 gaming experience.

 

But, again, the worst thing I can say about the RTX 2080 Super is that it suffers in the shadow of the impressive, competitively priced RTX 2070 Super. There's no other "medium-high 4K" card like it on the market, and every boost compared to the non-Super variant is welcome. But we're likely going to have to wait for AMD to light another fire under Nvidia's butt before we see this card's price or performance shift more significantly.

 

Listing image by Nvidia

 

 

 

Source: Nvidia RTX 2080 Super hands-on: The result when AMD is out of striking distance (Ars Technica)

 

(To view the article's image galleries, please visit the above link)