It is 1.55 pm, March 5, 2025. I am pacing the confines of the house workplace/picket shack on the backside of my backyard I have been inhabiting pretty solidly for the previous two months as I benchmark two complete new generations of GPUs from each Nvidia and AMD, and it is 5 minutes earlier than the assessment embargo drops for AMD’s Radeon RX 9070 XT graphics card. I am on the cellphone with somebody I’ve recognized at AMD for years. Once more. And no less than they’re apologising for the very fact the corporate has been unable to offer me any sort of dependable pricing concerning the totally different variations of its GPU I have been testing.
There have been no reference playing cards for AMD’s new RDNA 4 structure, and so I’ve obtained a choice of totally different choices—all with totally different value tags—in entrance of me to base my successfully definitive assessment of what’s really the perfect AMD GPU in lots of a protracted yr. And doubtlessly none of them are on the ultra-aggressive $599 value level that ought to make this spectacular card an outright winner.
After which eight minutes after the embargo lifts, XFX messages to say, “change of plan, sorry” and that now the premium XFX Mercury RX 9070 XT goes to be a reference-priced card on launch the next day. Although really, that by no means involves move in any manner that is seen on-line, and the cardboard really finally ends up retailing for over $900 for the following few months as a substitute of the $599 I am promised.
What a shit technique to spend a birthday. A birthday the place I am not even technically working.
I am beginning with AMD right here solely as a result of the RX 9000-series launch was the end result of a irritating interval of late nights and lengthy days of graphics card benchmarking, however the early-year Nvidia launches of the RTX Blackwell technology have been no higher.
Positive, we had Founders Editions for many of them, giving Nvidia no less than the chance to pay a little bit greater than lip service to its recommended retail costs. However as reviewers, we nonetheless needed to wrestle with late and typically flaky drivers and ultra-short assessment instances; then as customers, we needed to take care of black display screen points, fixed worries about that damned cable, low provide, and completely brutal value gouging from each producers and retailers mixed.
To not point out considerations that a lot of the vaunted generation-on-generation efficiency will increase have been largely right down to the introduction of Multi-Body Era (MFG) in just a few key video games and never widespread precise rendering efficiency will increase from the silicon itself.
The relative chip sizes obtained smaller, arguably much less advanced inside a GPU tier (RTX 5070 vs RTX 4070, I am taking a look at you) and meant you have been left with the sensation that Nvidia was huffing its personal excretions when it got here to the notion that graphics playing cards must be judged purely on efficiency over silicon.
AMD ought to have been in a position to utterly run rampant, what with its mixture of cheaper price and virtually equal efficiency in contrast with the competing RTX 5070 Ti—itself arguably the perfect RTX 50-series GPU. However AMD had a bizarre begin to the yr, the place the RX 9070 XT and RX 9070 playing cards have been introduced to the press previous to the brand new yr however have been notably absent from its huge CES press convention.
I used to be rapidly ushered behind the scenes out in Vegas after the press convention, the place the notably downbeat David McAfee and Frank Azor tried to speak up the RDNA 4 structure regardless of the pink crew presumably having been spooked by Nvidia’s saying its RTX Blackwell playing cards later that day. And, when you remembner, that was an announcement that offered the $549 RTX 5070 as set to ship RTX 4090 gaming efficiency.
As the mandatory inclusion of MFG in these efficiency calculations turned clear, nevertheless, AMD solidified its place. Although the demand and lack of provide inevitably made the RX 9000-series launch a blood bathtub. With no manner for AMD, and to a barely lesser extent Nvidia to manage costs, and the information that voracious GPU consumers would pay by the nostril for playing cards—as they did in the course of the pandemic years—producers and retailers each sniffed a chance to gouge players on pricing, and so the bloodletting started.
It was frankly offensive the way in which that every was decided to extract their very own pound of flesh from the availability/demand iniquity. Although it was clearly comprehensible, as each had watched scalpers make financial institution in the course of the earlier COVID and mining-related GPU provide crises, and now needed to ensure they might fill their respective coffers from players’ want for brand new silicon.
We will speak about dodgy drivers, deceptive efficiency guarantees, and GPU tiering we do not agree with, however the pricing aspect was arguably the largest subject round this calamitous technology of graphics playing cards, and that, no less than, wasn’t both AMD or Nvidia’s fault. Simply blame capitalism, I assume.
The RX 9070 XT and the RTX 5070 Ti, nominally $599 and $749 playing cards respectively, each ended up being priced at $900+ for no less than the primary few months of their lives, and solely in the direction of the tip of the yr have costs and inventory begun to stabilise and settle right down to across the unique launch costs of the playing cards.
It was the worst collective generational GPU launch I can bear in mind in my 20 years as a PC gaming tech journalist, and I used to be there for the nightmarish AMD R600 launch and carried a super-heated GTX 480 all the way in which again from Paris. I, for one, am very glad it is handed. At the least as 2025 fades into reminiscence, we are able to sleep protected within the information that GPU costs have come down. Nonetheless quickly.
Welcome to a different pricing disaster because the RAMpocalypse surges and suggests GDDR pricing for graphics playing cards is ready to rise alongside rumours that Nvidia has stopped bundling VRAM with its chips to card producers. In any case, something with reminiscence goes to be getting ever extra pricey for the foreseeable future, because of damned ‘AI’ demand.
So yeah, sleep protected, and haven’t got nightmares.

Greatest graphics card 2025





