AMD executives react to Nvidia’s RTX Spark — ‘you’re just wrong if you don’t get a Strix Halo notebook’
AMD welcomes the competition, and postures that it has the superior platform.
Regardless of your opinion of Nvidia’s RTX Spark, there’s no doubt that it’s the most consequential consumer announcement to come out of Computex 2026. Unlike Intel, which told Tom’s Hardware it’s handling the launch with “a healthy dose of paranoia,” AMD’s executives are confident that its Strix Halo and upcoming Gorgon Halo products will compete well with the N1X and N1 under the RTX Spark brand.
“I’m really excited that Nvidia has joined the game. You know, we were the only game in town for almost two years now, and the large local memory is becoming super critical in the agentic AI [workloads],” said AMD’s Rahul Tikoo, senior vice president and general manager of AMD’s client business. “I'm actually happy to see Nvidia join the race for these great products.”
AMD, of course, believes that Strix Halo and eventually Gorgon Halo are positioned well against RTX Spark devices. In a separate discussion, AMD’s Andrej Zdravkovic, chief software officer, said, “At this point in time… I mean, you’re just wrong if you don’t get a Strix Halo notebook,” when speaking on the choice of machine for developers. As the software lead at AMD, however, Zdravkovic is distanced from the hardware. Tikoo had more direct comments on the hardware comparison.
“I’m actually curious about what [Nvidia has] done, but when I look at their specs, their specs are 128 gigs of local memory. We’ve done it on Strix Halo. Their specs are a 20-core CPU. We have a 16-core / 32-thread CPU in here,” Tikoo said. “So, if you just compare the specs, I don’t see… now, Gorgon Halo, which is coming out in Q3, is going to be a better product.”
Hardware is only one part of the battle, which has become clear as AMD continues to push its way into AI infrastructure. Tom’s Hardware asked Zdravkovic about the so-called ‘CUDA moat’ that Nvidia has built for itself, and how AMD plans to address that as it rolls out updates for its own ROCm stack.
“If you asked me the same question like three years ago, I would be, yeah, that really matters. I think that matters less at this point,” Zdravkovic told Tom’s Hardware. “Nvidia has created a phenomenal ecosystem around CUDA, and our advantage is that ROCm is, from a developer point of view, extremely easy to use… the shift from one to another is easy, and the only challenge is if your application ends up using some of the specific commands that Nvidia has and we don’t, and the other way around.”
The posturing against Nvidia is expected, both on the hardware and software side, but Tikoo also pointed out that Nvidia’s entrance into the consumer PC market has downstream benefits for AMD.
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
“Nvidia has brought validity into the space… I think it’s also going to help the ecosystem move forward faster, right, because Nvidia and [AMD] are the two big players in this space, and both of us now being in this space not only drives the cloud ecosystem, it drives the AI ecosystem in the PC on Windows, and so we’re excited about that,” Tikoo said.
We’re still a few months away before we can see what material impact RTX Spark has on the broader industry, though we expect the initial rollout to be more muted than Nvidia’s keynote suggests. Although Nvidia plans to sell configurations of RTX Spark down as low as 16 GB of memory, the initial configuration will top out at 128 GB and likely demand several thousand dollars. At least initially, it’s a product that looks like it will appeal to a relatively small (but growing) market of AI developers, not dissimilar to Strix Halo.
AMD’s upcoming Gorgon Halo chips are largely a refresh of Strix Halo, leveraging the same Zen 5 cores for the CPU and RDNA 3.5 cores for the GPU, though with a bump up to 192 GB of unified memory. At least from the memory perspective, which continues to be an important specification for AI workloads, AMD has the edge. But, as we’re all well aware, there’s far more that goes into a platform (especially a consumer platform that costs several thousand dollars) than memory alone.
It will be interesting to see how the dynamic between Nvidia and AMD plays out, as we expect Gorgon Halo and RTX Spark to arrive in the same window; AMD says Q3 for Gorgon Halo, while Nvidia has simply said “fall” for RTX Spark.
Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.

Jake Roach is the Senior CPU Analyst at Tom’s Hardware, writing reviews, news, and features about the latest consumer and workstation processors.
-
usertests Reply“I’m actually curious about what done, but when I look at their specs, their specs are 128 gigs of local memory. We’ve done it on Strix Halo. Their specs are a 20-core CPU. We have a 16-core / 32-thread CPU in here,” Tikoo said. “So, if you just compare the specs, I don’t see… now, Gorgon Halo, which is coming out in Q3, is going to be a better product.”
Strix/Gorgon Halo can't be considered a gaming platform because of inflated prices. That leaves AI, and RDNA3.5 is way behind RDNA4/5 in AI performance per CU, as well as Spark. But Spark's theoretical performance is often held back by having similar memory bandwidth to Strix/Gorgon Halo.
I could see someone talking themselves into going AMD for its broad x86 compatibility. -
RagingAardvark They'll never admit it, but the SW ecosystem gap is still HUGE. I saw how the sausage is made at AMD, and even when they have better HW specs they can not compare to Nvidia for AI SW stacks. We had to beg borrow and plead to get any perf libraries supported with ROCm. Basically you've got two choices with AMD for AI/ML:Reply
1) Use Microsoft's HW-agnostic libraries (and AMD has a good team working with them).
2) Use only bare bones SW libraries that ROCm will support on a specific SKU, but forget using any perf libraries for performance or memory reduction. We were desperate to get any support for FlashAttention(1) on Radeon, while FlashAttention3 was already supported on CUDA.
Want to do inference? Fine, you can do that (mostly) with AMD. Want to do almost anything more? Just use NVidia. I yanked out my AMD GPUs and rolled back to a 5 year old 12GB NV card to do basic training tasks out of sheer frustration. Was it slow? Yes. Did it work? YES! Couldn't say that about AMD. -
hotaru251 Reply
thats like saying a maxed out imac isnt an imac because its vastly mroe expensive than the base model..usertests said:can't be considered a gaming platform because of inflated prices
"gaming" tax has always existed.
the fact that its more costly doesnt mean it isnt a gaming platform (i mean just because a 5090 is so costly doesnt mena its not a gaming gpu)..it just emans its got a high cost.
while it will likely beat amd in "ai" like features (as few spend mroe on that stuff than nvidia) in stuff not using that it will likely perform worse. (and again this is vs already out stuff not whatever is upcoming from apple/amd and doubt nvidia is going to update this regularly like they do) -
usertests Reply
Gaming is just not the relevant use case for Strix Halo right now. You don't get these boxes with 128 GB RAM (soon 192 GB with Gorgon) for gaming, you get em for AI. But the Halo/Sparks with lesser 32 GB are also inflated in price from what I've seen, and people should head for laptops with dGPUs instead. In the handheld realm, Panther Lake looks superior on performance at low/typical TDPs, and Strix Halo handhelds saw massive price hikes a couple months ago.hotaru251 said:thats like saying a maxed out imac isnt an imac because its vastly mroe expensive than the base model..
"gaming" tax has always existed.
the fact that its more costly doesnt mean it isnt a gaming platform (i mean just because a 5090 is so costly doesnt mena its not a gaming gpu)..it just emans its got a high cost.
while it will likely beat amd in "ai" like features (as few spend mroe on that stuff than nvidia) in stuff not using that it will likely perform worse. (and again this is vs already out stuff not whatever is upcoming from apple/amd and doubt nvidia is going to update this regularly like they do)
AYANEO NEXT 2 with Ryzen AI Max+ “Strix Halo” costs up to $4,299, weighs 1.4kg, and may require airline approval formsAYANEO suspends NEXT 2 sales, Ryzen AI Max+ 395 handheld cost now too high
Strix Halo feels like a beta test of the new InFO-oS "sea of wires" packaging that will be used with Zen 6 desktop CPUs. RDNA3.5 is showing its age. But it is going to be the correct product for some people even at $4k+. And if prices plummeted, then it would be great for many people. But I think Medusa Halo, whenever it comes out, will solve some of Strix's problems.
Nvidia showed a Spark roadmap with 2 additional generations based on Rubin and Feynman. Coming in 2028 and 2030 if I'm reading it right. -
kealii123 "You know, we were the only game in town for almost two years now"Reply
Apple: "Am I a joke to you? " -
kealii123 But seriously, once you meet some minimum basic requirements, the #1 thing influencing the speed of LLMs is the memory speed, and these two have exactly the same.Reply
Meanwhile apple silicon devices can be triple or quadruple the memory bandwidth. -
hotaru251 Reply
they were.kealii123 said:Apple: "Am I a joke to you? "
gaming on apple is still bad (playable but for $ you putting down its bad) even on the HALO products. -
Gururu Not sure how popular Strix Halo is, the coverage treats it with gloves and keeps it in the closet. There are some legit concerns being raised about the lack of RTX Spark performance data. All hype and no results might make for quite the popularity descent as we reach Fall.Reply -
-Fran- Reply
Wait until you see the prices nVidia want you to pay for the N1X.usertests said:Strix/Gorgon Halo can't be considered a gaming platform because of inflated prices. That leaves AI, and RDNA3.5 is way behind RDNA4/5 in AI performance per CU, as well as Spark. But Spark's theoretical performance is often held back by having similar memory bandwidth to Strix/Gorgon Halo.
I could see someone talking themselves into going AMD for its broad x86 compatibility.
Remember Jensen thinks we all have $10K battlestations at home. Think about that for a second.
Strix Halo can absolutely game and very well. nVidia showed jack at their presentation about actual games, ergo they haven't been able to make them run reliably. You know how much Jensen loves demo'ing stuff during his ego-trips on stage. Gorgon Halo will be even better, so I doubt AMD has much to worry about it.
Regards.