Test setup & performance

Test Setup

The detailed specifications of our new graphics card testbed system is as follows:-

• Intel Core i7-8086K (4.0GHz, 12MB L3 cache)
• ASUS ROG Strix Maximus X Hero (Intel Z370)
• 4 x 8GB G.Skill Ripjaws V DDR4-3000 (Auto timings: CAS 15-15-15-35)
• Samsung 860 EVO 500GB SSD
• Windows 10 Pro 64-bit
• ASUS PB287Q, 4K monitor

The full line-up of graphics cards and their driver versions are listed below:

• ASUS ROG Strix GeForce RTX 2080 Ti Gaming OC (ForceWare 417.71)
• GALAX GeForce RTX 2080 Ti Hall of Fame (ForceWare 417.71)
• MSI GeForce RTX 2080 Ti Lightning Z (ForceWare 417.71)
• Palit GeForce RTX 2080 Ti GamingPro OC (ForceWare 417.71)
• Zotac Gaming GeForce RTX 2080 Ti AMP Extreme Core (ForceWare 417.71)

Test cards compared

	ASUS ROG Strix GeForce RTX 2080 Ti Gaming OC	GALAX GeForce RTX 2080 Ti Hall of Fame	MSI GeForce RTX 2080 Ti Lightning Z	Palit GeForce RTX 2080 Ti GamingPro OC	Zotac GeForce RTX 2080 Ti AMP Extreme Core
Launch SRP	From S$2359	From S$2299	From S$2388	From S$1799	From S$2069
Core Code	TU102	TU102	TU102	TU102	TU102
GPU Transistor Count	18.9 billion	18.9 billion	18.9 billion	18.9 billion	18.9 billion
Manufacturing Process	12nm FinFET	12nm FinFET	12nm FinFET	12nm FinFET	12nm FinFET
Core Clock	Core: 1,365MHz, Boost: 1,665MHz	Core: 1,350MHz, Boost: 1,635MHz	Core: 1,350MHz, Boost: 1,770MHz	Core: 1,350MHz, Boost: 1,650MHz	Core: 1,350MHz, Boost: 1,755MHz
Stream Processors	4,352	4,352	4,352	4,352	4,352
Texture Mapping Units (TMUs)	272	272	272	272	272
Raster Operator units (ROP)	88	88	88	88	88
Memory Clock (DDR)	11GB GDDR6 14,000MHz	11GB GDDR6 14,000MHz	11GB GDDR6 14,000MHz	11GB GDDR6 14,000MHz	11GB GDDR6 14,400MHz
Memory Bus width	352-bit	352-bit	352-bit	352-bit	352-bit
Memory Bandwidth	616GB/s	616GB/s	616GB/s	616GB/s	633.6GB/s
PCI Express Interface	3.0	3.0	3.0	3.0	3.0
Power Connectors	2x 8-pin	3x 8-pin	3x 8-pin	2x 8-pin	2x 8-pin
Multi GPU Technology	Yes (2-way SLI)	Yes (2-way SLI)	Yes (2-way SLI)	Yes (2-way SLI)	Yes (2-way SLI)
DVI Outputs	No	No	No	No	No
HDMI Outputs	2x	1x	1x	1x	1x
DisplayPort Outputs	2x	3x	3x	3x	3x
HDCP Output Support	Yes	Yes	Yes	Yes	Yes

Benchmarks

Next up, here's a list of all the benchmarks used. We've already run through a wider range of tests in our review of the GeForce RTX 2080 Ti Founders Edition, so we'll just be going through a more limited selection here:

• 3DMark
• Shadow of the Tomb Raider

We used the Fire Strike Extreme benchmark and stress test in 3DMark for our temperature and power consumption tests respectively.

3DMark

The synthetic 3DMark benchmark tests graphics and computational performance at different resolutions, starting at 1080p and going all the way up to 4K. A series of two graphics test, one physics test, and then a combined test stresses your hardware in turn to assess its performance. Compared to the Founders Edition, these custom cards are a hair faster.

The MSI card took the lead in most of the 3DMark benchmarks, probably owing to it having the highest boost clock speeds. However, the difference wasn't that large, and it was just under 4 per cent ahead of the ASUS card in the 4K Fire Strike Ultra benchmark. Still, it's a commendable figure for a high-end card where usually small clock speed differences have little performance impact.

Shadow of the Tomb Raider

NVIDIA has mentioned Shadow of the Tomb Raider a lot lately, largely because it will be one of the first games to support its RTX technology through a post-release patch. At the time of writing, however, the patch hasn't dropped yet, so the results still deal with conventional performance metrics. Once more, the custom cards here are just a smidge faster than the Founders Edition.

However, they once again performed very similarly to each other, and you'd really be wasting time attempting to parse the small differences between them. 

Temperature and power consumption

I ran 40 loops of 3DMark's Fire Strike Extreme stress test and took the peak temperature readings during the run. For power consumption, I measured the total system power draw during the Fire Strike Extreme benchmark.

The custom cards tested here generally ran cooler than the Founders Edition, which posted a peak temperature of  75°C. However, the ASUS card was the standout here, posting temperatures that were a good 10°C less than the competition! This is made even more impressive by the fact that it is also the thinnest card in this shootout. In other words, it doesn't always matter how large your heatsink is, and good design and engineering trumps all. 

When it came to power consumption, the GALAX and MSI drew the most power from the wall socket, perhaps owing to the fact that they both had three 8-pin connectors instead of just two. 

Overclocking

I used the NVIDIA Scanner tool built into EVGA's Precision X1 software to overclock the cards. NVIDIA Scanner is an API that developers can bake into their overclocking software. In a nutshell, it leverages NVIDIA's knowledge of how its GPUs perform to help you achieve the highest possible overclock, with the goal of eliminating the tedious, trial-and-error nature of conventional overclocking. 

After you hit the Scan button, the software will work its way through your card's voltage-frequency curve, running arithmetic tests to see whether the card is stable or not. If it crashes, Scanner will automatically ramp up the voltage and try again. The entire process takes about 20 minutes, after which you'll get a custom voltage-frequency curve and overclocking profile that you can apply. 

Here's a table summarizing the clock speeds each card was able to achieve with NVIDIA Scanner (memory clocks remain unchanged):

Overall, the MSI GeForce RTX 2080 Ti Lightning Z was able to achieve the highest boost clock speeds, which translated into the best performance in 3DMark, albeit by just by a small degree. For instance, it was just over 3 per cent ahead of the ASUS card in Fire Strike Ultra.

The graphs below also showcase one other trait - the 3DMark results are much more even at overclocked settings. What this means is that while MSI is leading the pack by raw performance numbers, the difference is much less pronounced and most cards can pack a similar level of performance in real word games - if you're not splitting hairs.