Friday, October 24th 2025

AMD Radeon AI PRO R9700 GPU Arrives October 27 at $1,299 for Retail

AMD has finally set the retailer date and the price tag for its professional Radeon AI PRO 9000 series graphics cards, suitable for tasks like professional visualization or workstations that can accommodate up to four cards. Set to launch on October 27 for retailers, the flagship Radeon AI PRO R9700 will arrive at $1,299. Previously, AMD focused solely on the OEM and system integrator segment, leaving retailers and individuals building their own workstations for AI and professional visualization workloads without options. The company aimed to launch this GPU for retailers by the end of July, but supply constraints prevented this, as OEMs and system integrators consumed the entire supply.

These Radeon AI PRO R9700 GPU features a 4 nm "Navi 48" GPU, equipped with 64 RDNA 4 compute units, providing 4,096 stream processors and 128 AI accelerators for enhanced matrix operations across various data formats. A key improvement over the RX 9070 XT gaming GPU is the increased memory capacity, now doubled to 32 GB. The memory subsystem includes 20 Gbps GDDR6 across a 256-bit wide interface, delivering 640 GB/s of memory bandwidth, supported by a 64 MB 3rd Gen Infinity Cache. The AMD Radeon AI PRO R9700 achieves up to 191 TeraFLOPS FP16 dense performance and up to 1531 TOPS INT4 sparse performance, all within a power footprint of up to 300 W. The card is packed in dual-slot design with blower-style cooler, so dense multi-GPU configurations can be easily achieved.
Source: via VideoCardz
Add your own comment

56 Comments on AMD Radeon AI PRO R9700 GPU Arrives October 27 at $1,299 for Retail

#26
Space Lynx
Astronaut
CheeseballNah, they just took Navi 48 XTX and gave it 32 GB of VRAM and slightly cut the boost clock. It should game fine since the driver is still Adrenalin.
man Nvidia really is king. because they have like 3 different kinds of cores on their gpu's, dedicated RT, and then dedicated AI with 5000 series, then regular cuda.

at least that is my understanding. AMD is really far behind... more than I realized.
Posted on Reply
#27
Cheeseball
Not a Potato
Space Lynxman Nvidia really is king. because they have like 3 different kinds of cores on their gpu's, dedicated RT, and then dedicated AI with 5000 series, then regular cuda.

at least that is my understanding. AMD is really far behind... more than I realized.
Ahh not really, especially with RDNA 4.0. This does have its own RT and Matrix cores, similar to the functionality in RT and tensor cores in the NVIDIA GPUs.
Posted on Reply
#28
Space Lynx
Astronaut
CheeseballAhh not really, especially with RDNA 4.0. This does have its own RT and Matrix cores, similar to the functionality in RT and tensor cores in the NVIDIA GPUs.
ah ok that is what i was asking, would this AI PRO card have increased Matrix cores? i guess not cause AMD is fail... wtf lol

as an AMD fanboy i am disappoint.
Posted on Reply
#29
Cheeseball
Not a Potato
Space Lynxah ok that is what i was asking, would this AI PRO card have increased Matrix cores? i guess not cause AMD is fail... wtf lol

as an AMD fanboy i am disappoint.
Ah I see. No, its the same GPU as the RX 9070 XT. The only technical difference is the bigger VRAM.

This Radeon PRO card is the only current workstation card they have at this time.
Posted on Reply
#30
igormp
Space Lynxah ok that is what i was asking, would this AI PRO card have increased Matrix cores? i guess not cause AMD is fail... wtf lol

as an AMD fanboy i am disappoint.
Fwiw, most Nvidia products have no difference either. A RTX PRO 5000 is just a cut down 5090 with extra VRAM, no difference in how the cores are made, nor any extra AI cores.
The only products that have an actual difference are their x100 lineup, such as the B100 found in the GB200/GB300, H100/H200, etc etc, for which AMD's counterpart are the Instinct lineup (MI300, MI350, MI450, etc).
Posted on Reply
#31
Noci
CheeseballAh I see. No, its the same GPU as the RX 9070 XT. The only technical difference is the bigger VRAM.
So, in summary: $ 600 for 16 GB DDR6 VRAM :eek: :roll:
Posted on Reply
#32
Space Lynx
Astronaut
NociSo, in summary: $ 600 for 16 GB DDR6 VRAM :eek: :roll:
yeah thats fucking wild to me, but eh i guess there is a market for it, just not us
Posted on Reply
#33
lambda
Availability will decide the actual pricing. Good card for the MSRP I think.

Nitpicking as usual but 48GB of VRAM + FP4 and FP6 support would've been very nice but I guess that's reserved for CDNA cards (for now).

Currently you'd need at least 3 of these to run GPT OSS 120B completely in memory and 4 to run a Qwen3 235B quant. Not bad but there's scope for improvement.
Posted on Reply
#34
Mescalamba
Good guy AMD.

1. doesnt gimp or cripple gaming vs computing GPUs
2. you can install regular drivers on PRO card and pro drivers on regular ones
3. memory tax is hilariously low compared to only existing competition (its ECC VRAM not regular, supposedly)

And yet, people complain.

If I should complain, I would like to know...

1. whats fp64 performance? (Im assuming around 9070 XT.. obviously, but still?)
2. is ECC 9700 AI PRO exclusive or not?

Apart that, AMD is finally working to make AI usable on their GPUs, so it will be decent solution, altho it will need probably some elbow grease to make a lot of AI related SW work. Thankfully a lot of it is in Python.
Posted on Reply
#35
lambda
Space Lynxah ok that is what i was asking, would this AI PRO card have increased Matrix cores? i guess not cause AMD is fail... wtf lol

as an AMD fanboy i am disappoint.
AMD is catching up fast in the inference space (in terms of software) but I have no idea about actual model training.

I guess Meta uses a decent amount of AMD chips.
Posted on Reply
#36
Mescalamba
lambdaAMD is catching up fast in the inference space (in terms of software) but I have no idea about actual model training.

I guess Meta uses a decent amount of AMD chips.
If you can run inference, you can train. And vice versa.
Posted on Reply
#38
xorbe
NociSo, in summary: $ 600 for 16 GB DDR6 VRAM :eek: :roll:
Yup, look at the 96GB RTX 6000 price, it's just scaling by vram, by an insane factor, due to AI demand.
Posted on Reply
#39
LabRat 891
Mescalamba1. whats fp64 performance? (Im assuming around 9070 XT.. obviously, but still?)

Stock:stock, Spec:spec, the R9700 is a SMIDGE slower than the 9070XT.
However, both of the R9700s I've had hands-on time with, reliably boosted beyond spec (w/o manual adjustment), and sustained it better than either of the 2x Pulse 9070XTs I've played with. -Under UE5 and other game loads, not LLMs, tho.
Mescalamba2. is ECC 9700 AI PRO exclusive or not?
Apparently, ECC VRAM support is only supported under a Linux+ROCm operating environment. But, is otherwise a common feature to R9700.
MescalambaApart that, AMD is finally working to make AI usable on their GPUs, so it will be decent solution, altho it will need probably some elbow grease to make a lot of AI related SW work. Thankfully a lot of it is in Python.
From what I've gathered, support is advancing rapidly.

To put things in-perspective, AMD favorably compared the R9700 32GB to the RTX 5080 16GB in their pre-release presentations. :laugh:
That sounds absurd at first but 'by the $', if your workload isn't massively advantaged by CUDA or you flat-out JUST need big VRAM, the R9700 is arguably economical for-purpose.

Still, it's not CUDA, and that's sufficient information for many to (understandably) completely dismiss the R9700.
If what 'AI things' one is doing are supported under ROCm or generic Vulkan, it's likely a very competitive option for 32GB of VRAM and competent computational performance.
Posted on Reply
#41
LabRat 891
CheeseballNah, they just took Navi 48 XTX and gave it 32 GB of VRAM and slightly cut the boost clock. It should game fine since the driver is still Adrenalin.
Pretty much, yes.

If AMD'd felt it profitable, we could've had an 'RX 9070 XTX 32GB'. -Basically, the R9700's VRAM and binnings but, with the 360w+ TDP (As-seen w/ the XT Taichi) and with 5080/5090-like PCIe slot-breaking coolers.

TBQH, I don't fault AMD for not making that. (despite, RX480->RX580->RX590 pulling off similar, pretty well)
A. All 9070XTs are already pretty far into 'diminishing returns', regarding power:clocks.
B. It seems like a waste of high-profit margin bin'd Navi48s. High$ products like Radeon AI Pro R9700 or a prospective W9700, are (probably) much better uses for those binnings.
C. The performance improvement from 9070 to 9070 XT is already small. The difference between AIB OC XT models, even smaller.
"Aligning the stars" for The Perfect Navi48 Card, would still net very little over the XT. [In 'retro' terms, think: 6800GT v. 6800Ultra 512MB]
mechtechIt's gone full circle ;)

www.techpowerup.com/gpu-specs/radeon-9700-pro.c50
AMD had to choose between Nostalgia Marketing and copying nVidia for this gen.

:laugh:
Posted on Reply
#42
JimmyDoogs
Can anyone actually find a gaming title that uses more than 16GB in 4k? Maybe in 8k I could see that but I game in 4k and have yet to find a game that uses over 16GB.

Just a few titles running in 4k as well: Most don't go over 12GB but I wouldn't use 12 for 4k just in case.

I do know this is not for gaming so I know it's slightly off topic but I'm always curious about VRAM usage.
Posted on Reply
#43
LabRat 891
JimmyDoogsCan anyone actually find a gaming title that uses more than 16GB in 4k? Maybe in 8k I could see that but I game in 4k and have yet to find a game that uses over 16GB.
UE5-driven MW5: Clans can eat up 18GB+. Problem though (on my 5800X3D) 7900XTX, 9070XT, and R9700 all still come up short on performance.
OtOH, Beam.NG/Beam.MP can use more than 16GB VRAM and, the R9700 performs notably 'better' than a 9070XT. Not sure why but, it is repeatable.
JimmyDoogsJust a few titles running in 4k as well: Most don't go over 12GB but I wouldn't use 12 for 4k just in case.
Most games I've tried out, seemingly try to stay inside 12-16GB, even beyond 4K (VSR).
Seems by 'where the numbers hang out' that these are intentional "optimizations" by game devs and/or the driver.
JimmyDoogsI do know this is not for gaming so I know it's slightly off topic but I'm always curious about VRAM usage.
AFAIK, the R9700s are the only 2-slot blower Navi 48 cards on the market/soon-to-market. These occupy a position similar to those nVidia RTX not-quadro low-profile cards. They're "professional" cards but, due to some specific overlaps in needs, ends up desirable to select Gamer-Enthusiast persuasions (SFF builds, LLMs, etc.).
Posted on Reply
#44
Tomorrow
watzupkenWow, the MSRP more than doubled over the RX 9070 XT @ 600 bucks, despite the fact that it’s the same GPU under the hood and just for the extra VRAM. They are really trying to milk their customers who are into this AI craze with a steep premium.
Go look at Nvidia workstation card prices and you'll see that +600 and 1300 total is pennies compared to what Nvidia charges for their workstation cards with extra VRAM.
igormpNot that bad, way cheaper than an used 4090 and considerably cheaper than the 5090, assuming this price actually holds in stores instead of being inflated by retailers.
However, if intel ever manages to deliver the B60 widely, a supposedly dual B60 would offer 48GB at a slightly lower price point.
Exactly. People who compare this to gaming cards only see the price go up , but there's zero benefit in 99% of today's games going from 16GB to 32GB.
In workstation/AI going from 16 to 32 is presumably much bigger impact. By the time 16GB+ becomes useful there are cheaper consumer cards available with 18-24GB.
LabRat 891BTW, the Navi48s on these cards seem better-binned than the 9070(XT) dies. I never saw 3.2-3.4Ghz boosts or sustained 3GHz+ on my 9070XT.
(Two example R9700s, both exhibited same behavior)
I constantly see sustained 3Ghz on up to 3.4Ghz boost on my XT. Could be your Pulse samples just not being very good.
Not that my Gigabyte Gaming OC is anything special...
CheeseballThis Radeon PRO card is the only current workstation card they have at this time.
In the RDNA4 family, yes. In previous families they're always had Pro cards too.
igormpFwiw, most Nvidia products have no difference either. A RTX PRO 5000 is just a cut down 5090 with extra VRAM, no difference in how the cores are made, nor any extra AI cores.
The only products that have an actual difference are their x100 lineup, such as the B100 found in the GB200/GB300, H100/H200, etc etc, for which AMD's counterpart are the Instinct lineup (MI300, MI350, MI450, etc).
People are expecting that from Nvidia. Thus no one bats an eye when they do it, but if AMD does it then they're greedy.
NociSo, in summary: $ 600 for 16 GB DDR6 VRAM :roll:
Go look at Nvidia workstation card prices and you'll see that +600 and 1300 total is pennies compared to what Nvidia charges for their workstation cards with extra VRAM.
LabRat 891UE5-driven MW5: Clans can eat up 18GB+. Problem though (on my 5800X3D) 7900XTX, 9070XT, and R9700 all still come up short on performance.
OtOH, Beam.NG/Beam.MP can use more than 16GB VRAM and, the R9700 performs notably 'better' than a 9070XT. Not sure why but, it is repeatable.

Most games I've tried out, seemingly try to stay inside 12-16GB, even beyond 4K (VSR).
Seems by 'where the numbers hang out' that these are intentional "optimizations" by game devs and/or the driver.
I did see about 14GB used in modded Skyrim at 1440p. Not sure how much the extra VRAM effected the performance as i had a 11GB 2080 Ti before and with 9070 XT performance improved by 60% in the 0.1% lows. 12% in averages and 20% in maximums. Im sure that at 4K i would be able to saturate 16GB.
But that's just one modded game.

By any chance do you know of 9700 has DP 2.1 UHBR 80 ports?
Previous Radeon Pro cards have had higher bandwidth display outs. Mostly miniDP.
I wonder if the same is true this time or have they cheaped out as XT has UHBR 13.5
Posted on Reply
#45
Rover4444
Too bad it sucks in image generation when compared to NVIDIA. For LLMs it might be good but you can stack four MI50s at that price and get 128GB LMAO
Posted on Reply
#46
LabRat 891
TomorrowBy any chance do you know of 9700 has DP 2.1 UHBR 80 ports?
Previous Radeon Pro cards have had higher bandwidth display outs. Mostly miniDP.
I wonder if the same is true this time or have they cheaped out as XT has UHBR 13.5
The ASRock Creator model has 4xDP2.1a (full size). I don't have a DP80-capable device, or I'd test that.
I use a Club3D UHBR13.5(DP54)->HDMI2.1 48gbps hub to drive my 4K OLED.

I had assumed the R9700 has DP80 but, monitors w/ that capability are few and far-between. So, I'm really not sure, either.
Posted on Reply
#47
Tomorrow
LabRat 891The ASRock Creator model has 4xDP2.1a (full size). I don't have a DP80-capable device, or I'd test that.
I use a Club3D UHBR13.5(DP54)->HDMI2.1 48gbps hub to drive my 4K OLED.

I had assumed the R9700 has DP80 but, monitors w/ that capability are few and far-between. So, I'm really not sure, either.
Hmm i could have swore GPU-Z had a Display page in the Advanced tab in the past that listed the maximum link speed, but now i cant find it.
Posted on Reply
#48
LabRat 891
TomorrowHmm i could have swore GPU-Z had a Display page in the Advanced tab in the past that listed the maximum link speed, but now i cant find it.
Same. First thing I checked, too.


Seems like without very clear specification from AMD, the only way to find out would be to try a UHBR20/DP80 display.

Gigabyte Aorus FO32U2 Pro
MSI MPG 322URX QD-OLED

MPG 272URX QD-OLED
ASUS PG27UCDM
are
is the only US/CAN market displays I'm finding w/ DP80/UHBR20.

Prior to release it'd been rumored to support UHBR20/DP80 but, I feel like it would be safer to assume "No."
A. The only AMD cards w/ advertised full UHBR20/DP80, were/are the W7800 and W7900.
B. The R9700 is not even supported by the same driver branch as the W-series. I suspect the DP80 feature would be 'left out' of the R9700.
Posted on Reply
#49
Tomorrow
Yeah without explicit support in the specs i would assume no. Same as ECC only in Linux. I assume it's some kind of software based ECC. If it were hardware then it would be OS agnostic. Shame. AMD could have put more work into this card. Any idea about SR-IOV ?
Posted on Reply
#50
eidairaman1
The Exiled Airman
MescalambaGood guy AMD.

1. doesnt gimp or cripple gaming vs computing GPUs
2. you can install regular drivers on PRO card and pro drivers on regular ones
3. memory tax is hilariously low compared to only existing competition (its ECC VRAM not regular, supposedly)

And yet, people complain.

If I should complain, I would like to know...

1. whats fp64 performance? (Im assuming around 9070 XT.. obviously, but still?)
2. is ECC 9700 AI PRO exclusive or not?

Apart that, AMD is finally working to make AI usable on their GPUs, so it will be decent solution, altho it will need probably some elbow grease to make a lot of AI related SW work. Thankfully a lot of it is in Python.
On another note the full focus is on UDNA anyhow.
Posted on Reply
Add your own comment
Nov 20th, 2026 17:07 PST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts