The RTX 4080 fills the gap left between the RTX 3090 and 4090. But at what cost?
The GeForce RTX 4090 launched just one month ago, and set a new standard for GPU performance. But raw performance has been an almost ancillary topic, as much of the discussion has revolved around the ongoing 12VHPWR connector saga. It also goes without saying that the $1599 starting MSRP of the card (and higher street prices) has not been universally praised.
Enter the GeForce RTX 4080. The savior of the RTX 40 Series, right?? Well, forget about $699 price tags, such as the one attached to the GeForce RTX 3080 launch two years ago (though it was nearly impossible to obtain one at list price in those days). This is 2022, and while things aren’t as scarce as they were a couple of years ago, they certainly cost more.
The crypto boom inflated the street prices of every graphics card on the planet from 2020 though the early part of this year, and while the bubble continues to burst other factors seem to have assured us of these new, higher GPU prices going forward. Thus, the RTX 4080 Founders Edition carries a list price of $1199 – a 70% increase over the RTX 3080’s list price at launch. Ouch.
When discussing GPU prices, it’s easy to point the finger at NVIDIA, particularly as AMD uses GeForce prices as part of their overall marketing strategy with Radeon. We recently learned that the upcoming Radeon RX 7900 XTX will launch at $999, and it remains to be seen if there is a $200 disparity in performance, as well.
I try to be neutral and refrain from consumer advocacy, judging products on their own merit and then bringing price into the mix as we look at price/performance ratios. I am not alone in disliking the trend towards $1000 enthusiast graphics cards. If we look at the past, however, these prices are actually not unprecedented (adjusted for inflation, anyway).
Go back 16 years to 2006, and if I walked into my local Best Buy that summer a BFG GeForce 7900 GTX was $599.99, as was an ATI Radeon X1900 XTX.
Ok, so maybe “walked into my local Best Buy” was unrealistic, as these pricey cards were only available online – based on the “store pickup: not available” status from Best Buy’s website in May of 2006. I can’t remember if my store had them on the shelves.
Anyway, these $599.99 street prices, when adjusted for inflation, are equivalent to $882.98 today. That’s still preferable to the $999 – $1199 of today’s enthusiast segment, but it doesn’t seem so crazy now, right? Well, we had all grown accustomed to cards that offered better performance for less money every year, and AMD spoiled us with a $399 flagship card (the RX 5700 XT) just three years ago.
We should move on from this depressing pricing narrative, and on to the new GPU itself.
GeForce RTX 4080 Specs
At the heart of the GeForce RTX 4080 resides the AD103, which at 379 mm^2 is about 38% smaller than the 608 mm^2 of the RTX 4090’s AD102. It has a much smaller TGP than the RTX 4090, as well, at 320W.
|RTX 4090||RTX 4080||RTX 3090 Ti||RTX 3090||RTX 3080 Ti||RTX 3080|
|Architecture||Ada Lovelace||Ada Lovelace||Ampere||Ampere||Ampere||Ampere|
|Base Clock||2235 MHz||2205 MHz||1670 MHz||1395 MHz||1365 MHz||1440 MHz|
|Boost Clock||2520 MHz||2505 MHz||1860 MHz||1695 MHz||1665 MHz||1710 MHz|
|Memory||24GB GDDR6X||16GB GDDR6X||24GB GDDR6X||24GB GDDR6X||12GB GDDR6X||10GB GDDR6X|
|Memory Data Rate||21 Gbps||22.4 Gbps||21 Gbps||19.5 Gbps||19 Gbps||19 Gbps|
|Memory Bandwidth||1 TB/s||716 GB/s||1 TB/s||936 GB/s||912 GB/s||760 GB/s|
|Die Size||608 mm^2||379 mm^2||628 mm^2||628 mm^2||628 mm^2||628 mm^2|
|Process Tech||TSMC 4nm NV Custom||TSMC 4nm NV Custom||Samsung 8nm NV Custom||Samsung 8nm NV Custom||Samsung 8nm NV Custom||Samsung 8nm NV Custom|
If we compare the GPU specs of the RTX 4080 to the RTX 4090, we see that quite a bit of room remains for a later RTX 4080 Ti offering. This new GPU features 76 SMs for a total of 9728 CUDA cores, while the RTX 4090 has 128 SMs for a total of 16384 CUDA cores.
The memory system with the RTX 4080 seems to be a weak point, as it offers “just” 16 GB of GDDR6X on a 256-bit bus. That narrower bus width means that even with its very fast memory data rate of 22.4 Gbps, there’s still only 716 GB/s of memory bandwidth – which is actually less than the 760 GB/s of the RTX 3080 (which had a 320-bit interface).
Comparing the memory bandwidth from the GeForce RTX 3080 Ti, with its 384-bit interface, paints an even less favorable picture of this new 256-bit RTX 4080. The RTX 3080 Ti offered 912 GB/s of bandwidth, just under the 936 GB/s number from the RTX 3090. We know the RTX 3080 Ti and RTX 3090 are pretty close in performance, but how far ahead this RTX 4080 might be remains to be seen.
The deciding factor, which is not listed in the above table, may very well be the L2 cache size, which is significantly higher with this new architecture. The RTX 4080 has 64 MB (65536 KB) of L2 cache, while the RTX 3080 Ti, for instance, has just 6144 KB. We’ve seen what a large cache can do on the AMD side with the Infinity Cache on RX 6000 Series GPUs, and that is closer to an L3 cache. 64 MB of L2 is absolutely massive.
The Founders Edition Card
Upon opening the box I was a bit surprised to discover that the design of this RTX 4080 Founders Edition is identical to that of the RTX 4090 Founders Edition, so if nothing else it should run very cool and quiet considering this is a far less power-hungry GPU.
This also means that case width remains vital to successful use of the card, as plenty of additional space is required between the card and side panel to prevent strain on the connector.
And yes, the proper insertion of the connector is also vital, and there may be other aspects to the adapter issues still under investigation as you read this. Regardless, you’re still going to want at least an inch and a half of space between card and glass with this design.
Some Benchmark Results
We will revisit the tests from the RTX 4090 review to see how this new card fares against its bigger brother, and other recent high-end NVIDIA offerings. Don’t look for flagship AMD cards here, as we still don’t have an RX 6900 XT or RX 6950 XT to test.
|PC Perspective GPU Test Platform|
|Processor||AMD Ryzen 9 7950X (Stock)|
|Motherboard||MSI MEG X670E ACE
BIOS v1.25 Beta
AGESA ComboPI 22.214.171.124 Patch A
Resizable BAR Enabled
|Memory||32GB (16GBx2) G.Skill Trident Z NEO DDR5-6000 CL30|
|Storage||SK Hynix Platinum P41 2TB NVMe SSD|
|Power Supply||be quiet! Dark Power Pro 12 1500W|
|Operating System||Windows 11 Pro, 21H2|
|Drivers||GeForce Game Ready Driver 526.72|
Some quick notes as to test methodology used in the making of this review, which is a topic that I don’t usually get into. For the benchmarks to follow, as always, each result I add to the chart is the average of three separate runs. If there is an unusual result during testing, such outliers (results with variance above 1%-3%) trigger investigation, invariably get scrapped, and then the game is restarted and the test is repeated until consistent results are achieved.
Also, as our RTX 4090 review contained results from Cyberpunk 2077 that clearly had display scaling enabled, even though it was toggled “off” and the game restarted before benchmarking, I was unwilling to use the game for benchmarking again. And then it was updated, so I tried again with the new v1.61, and turning FSR off actually worked! The results are all new for this game, and cards were tested with the RTX 4080 press driver.
A lot of work actually goes into reviews like this, as I am fanatical about accuracy in benchmarking. If you have the same test setup as I used here, with the same OS and driver version, and don’t have any rogue processes or other disruptive Windows nonsense going on, you should get identical results to those presented here.
We begin with 3DMark, and specifically Time Spy Extreme. This is a 4K DX12 test, and we are just looking at the GPU score here:
The RTX 4080 gets off to an impressive start here, I must admit. It’s not quite halfway between the RTX 3090 and RTX 4090, but it’s significantly faster than our factory overclocked RTX 3090 (MSI SUPRIM) in this test. Comparing just the 40 Series cards, the RTX 4080 FE was about 27% slower than the RTX 4090 FE.
Next we’ll take a look at 3DMark Port Royal, which was “the world’s first dedicated real-time ray tracing benchmark for gamers” according to UL Benchmarks. Speed Way is the new kid on the block (a very demanding DX12 Ultimate test), but I didn’t get all of the cards tested in time (that will have to wait for the next review!).
In Port Royal the RTX 4080 FE is about 30% slower than the RTX 4090 FE, which is a slightly bigger gap than we saw in Time Spy Extreme (~27%).
Next we’ll look at a few games, starting with Metro Exodus Enhanced Edition.
We are seeing pretty much the same scaling here as in the 3DMark tests. Here the RTX 4080 FE is 27% slower than the RTX 4090 FE, and 33% faster than the RTX 3090. But how about in a less NVIDIA-optimized title like DiRT 5?
Ok, still a good showing, but the RTX 4080 FE isn’t quite as far ahead of the RTX 3090 in this test – though it is still 27% faster than the RTX 3090.
Next we will look at Cyberpunk 2077, and these results are all new for this review. We tested the new version (v1.61) and this time FSR was really disabled. For real. Absolutely 100% sure on this. RTX 4090 numbers are lower and everything.
In Cyberpunk 2077 – at 3440×1440 and the Ultra (non-RT) preset, the RTX 4080 FE is about 25% faster than the MSI SUPRIM RTX 3090.
Next we will look at a couple of DLSS benchmark results, beginning with another look at Metro Exodus Enhanced Edition. Bear in mind, the two tests to follow are using DLSS 2.0, not 3.0. We will have to follow up with new DLSS 3.0 results from this card.
With DLSS set to “Balanced” the RTX 4080 FE got closer to the RTX 4090 FE than I thought would be possible. It was only 13% below than the RTX 4090 here! I will have to revisit this test at a higher resolution than 3440×1440 to see if this impressive position holds up.
Next up we will look at Bright Memory Infinite, run with the “Very High” RTX preset and DLSS set to “Quality”.
In this test the RTX 4080 FE is pretty much exactly between the RTX 3090 and RTX 4090. It’s about 25% behind the RTX 4090 FE, and the RTX 3090 is about 24% behind the RTX 4080 FE. That was probably a confusing way to expound on that, but in any case it’s in the middle.
Power, Clock Speeds, and Thermals
When compared to the RTX 4090 FE, the RTX 4080 FE has roughly 25% lower performance (yes I know this will vary depending on the test, and I only tested three titles), at a 25% lower price. So, will it also consume 25% less power?
We will look at power and clocks to find out.
GPU clocks are a little higher from the RTX 4080 FE, compared to our RTX 4090 FE sample, and power draw is indeed about 25% lower. We didn’t quite hit the 320W TGP of the card, coming close only with a momentary reading of 318.9 watts – which was an outlier as all other results were 313.9 watts or lower.
Now we will move on to a protracted section about thermals and fan speeds.
During the 10x iterations of the Metro Exodus benchmark (3440×1440, extreme preset) our RTX 4080 Founders Edition sample was a very cool-running card, and the fans were not noticeable over CPU cooler noise. These 40 Series Founders Editions are very quiet cards, and at some point I will have to implement an ultra-quiet (or passive) CPU cooling setup to allow for accurate GPU noise measurements.
The RTX 4080 FE registered a max 60.6 C GPU Temperature, and 70.7 C Hot Spot, as reported by the card and captured via GPU-Z data logging. The room during these tests was just under 19 C (~18.8 C), which is a few degrees cooler than our RTX 4090 FE testing (~21.5 C ambient), so looking at deltas makes more sense.
Comparing delta temps, then, we find that the RTX 4080 FE produced a max of 41.8 C GPU / 51.9 C Hot Spot, while the RTX 4090 FE produced a max 45.5 C GPU / 55.3 C Hot Spot. These slightly cooler temps make sense considering a smaller, less power-hungry GPU is being cooled with a thermal design created for the RTX 4090 FE’s big 608 mm^2 GPU and 450W TGP.
Please note, however, that the RTX 4080 FE’s use of this cooler will differ in that the fan speeds are lower by default, since it has less of a thermal load to contend with. I recorded a max fan 1 speed of 1421 RPM, and max fan 2 speed of 1326 RPM from the RTX 4090 FE under load. This RTX 4080 FE, on the other hand, had a max fan 1 speed of 1275 RPM, and max fan 2 speed of 1223 RPM under the same conditions.
Not a huge difference, but still about 10% lower fan speeds on average and slightly better temps.
I could be really lame here and say that NVIDIA’s GeForce RTX 4080 Founders Edition performs exactly where it should in relation to the RTX 4090, given the price. If we pretend that this conclusion isn’t concerning cards that are $1599 and $1199, it is probably easier to digest.
For $400 less – 25% less money than the list price of an RTX 4090 FE – we are getting a card that performs almost exactly that much slower. I really think that NVIDIA priced this – in relation to the RTX 4090 FE – appropriately. Not that there’s anything appropriate about a $1199 RTX xx80 card.
I guess this just isn’t the right time in history to be a PC gaming enthusiast. Everything costs too damn much, and with things tight financially for a lot of people, the inflated prices compared to previous generations make enthusiast-level cards like the RTX 4080 FE out of reach for most of us.
AMD is promising a lot with the upcoming RX 7900 XTX and RX 7900 XT, but even at $999 and $899, respectively, they won’t come cheap. Rest assured, if the RX 7900 XTX really is an RTX 4080 competitor, and costs $200 less, NVIDIA will have some sort of answer for it.
NVIDIA makes the fastest gaming graphics cards on the planet right now, with the RTX 4080 only surpassed by the RTX 4090. They are going to cost you, and it’s frustrating that prices have gone up this much in two years, but here we are.
For my own system, I think I’m going to continue playing older games that don’t require the latest graphics until prices come down to earth. But I’m one of those weird people who think “PC gaming” and “DOSBox” are synonyms.