Every day I wake up and thank God for having me be born 23 minutes away from a MicroCenter

[-]

ga239577@reddit

Isn't the R9700 a better buy at $1299.99-$1349.99?

According to Gemini the R9700 has over double the FP32 and close to double the FP16 compute and memory bandwidth is slightly higher on the R9700.

This should translate into faster performance on both PP and TG unless I'm missing something.

[-]

mr_zerolith@reddit

It's a much better deal per $

It's nice a to have a good amount of vram, but you also need speed to run the big models that ram enables.

That R9700 is the best value starter hardware you can get today.

[-]

skinnyjoints@reddit

This is a bit of a big question but I would really appreciate an answer from someone that knows what they’re talking about.

I keep hearing “NVIDIA has a monopoly” and “Only NVIDIA chips are good”. The reasoning provided is almost always “because they have CUDA”. Simultaneously, I see GPUs from Intel and AMD that offer the same performance as NVIDIA GPUs for thousands of dollars less. So it’s reasonable to conclude that CUDA does offer a lot of value.

What I don’t understand is what this value actually is? What does CUDA offer that non-NVIDIA chips don’t have? Can I run models on these alternatives? Fine tune them? Train them from scratch? Use llama.cpp, PyTorch, and other AI platforms?

Ultimately, what is the big deal with CUDA?

[-]

AlwaysLateToThaParty@reddit

A ziskin benchmarked them the other day, and the intel performed better.

[-]

ga239577@reddit

Yeah I'd say the Intel one would be more competitive at $699.99 ... $949.99 it's kinda DOA unless Intel can improve stock enough to make up for the differences

[-]

mr_zerolith@reddit

I still wouldn't buy it at that price, the problem is the speed is not there. The larger the model, the more speed you need.

You'd need a ton of them in parallel and then you get into server board territory which gets super expensive versus a dual x8 capable board and two much better cards ( 5090, RTX PRO 6000, AMD R9700 )

[-]

ga239577@reddit

I have one R9700 and it's fast enough for me to get several iterations of changes completed on a project per day.

Nowhere near as fast as Cursor or cloud AI, but it works for me.

[-]

exact_constraint@reddit

+1. Running an R9700 and looking to scale to multiple cards. If Intel knocked another few hundred bucks off, I’d probably take the performance hit. But the B70 is just too close to the price of the R9700 to justify the hit.

[-]

Tradeoffer69@reddit

Pretty similar bandwidth just a tad ahead for AMD.

Maybe thats due to the more mature ROCm?

Also Intel has lower power draw so more scalable?

[-]

SexyAlienHotTubWater@reddit

AMD probably just runs the RAM hotter, which means they can slightly bump the frequency.

[-]

EmPips@reddit

From almost every angle, yes.

But $949 for 32GB at that power draw and memory bandwidth.. this card will definitely find a home. It's not a bad buy at all.

[-]

TheBlueMatt@reddit

You're really betting on the drivers improving. There's a ton of headroom in the Intel drivers and optimization but who knows how far it'll go.

[-]

ga239577@reddit

Yeah I don't think the drivers can make up for the pp speeds since it's compute bound / and R9700 has 2x the compute.

[-]

Lancelot3777@reddit

One that is funny. Two curious if you run this on your main PC will it bog down your computer. I want to run Gemma 4.

Openclaw Anthropic Refugee.

[-]

79215185-1feb-44c6@reddit

RIP

[-]

gigaflops_@reddit (OP)

Keep checking every day. I've seen them come in stock 2-3 times since release. You're bound to find a restock one of these days if you keep checking.

[-]

ZippyHighway@reddit

For what its worth, Newegg had them in stock for most of this morning. I kept checking to see how fast they were going.

[-]

here_n_dere@reddit

Not sure if people would buy this for ai or gaming?

[-]

PuddleWhale@reddit

I thought it Nvideas were used almost exclusively to run models.

[-]

ImportancePitiful795@reddit

Nope. You can run models on Intel/AMD.

Training needs bit work because everything is about "nvidia". But can do it also.

Also at this point you get 4xB70s for the cost of a single 5090. Which is insane.

[-]

IntelligentFire999@reddit

Jezz... how much is a 5090 these days? I have one that i use to just play games lol.

[-]

ImportancePitiful795@reddit

Close to $4000/€4000 the cheapest one.

Roughly 4xB70s.

[-]

IntelligentFire999@reddit

Yikes. It was ~$2600 last year this time (msrp is/was $1999). Crazy times.

[-]

Frosty-Cup-8916@reddit

You can run models, training is harder though iirc

[-]

eFlick98@reddit

is it the one in Wayne? 👀

[-]

Hyiazakite@reddit

Bought 4 to compare these with my 3090's results were .. abysmal to say the least .. (Linux - tried IPEX, SYCL, OpenVINO - vLLM, llm scaler, llama.cpp builds) .. to some extent I believe it to be driver issues as the cards were only briefly reaching max TDP during inference so they will probably get better but the only models I got running were pretty outdated (Qwen3 being the most modern, got Qwen3.5 running in llama.cpp SYCL but really slow). Returned them today. I'll gladly pay a few extra 100 EUR for something that just works.

[-]

droans@reddit

You need to use OpenArc. The vLLM and llama.cpp implementations are horribly built and contain basically zero hardware optimizations.

It took me around ~45s to get a response from a 1k token prompt with llama.cpp whereas it took less than five seconds with OpenArc.

[-]

Jolly-Event7578@reddit

So this B70 it's not as good as a 3090 ? You would pay 100 extra to get what instead ?

[-]

SexyAlienHotTubWater@reddit

R9700

[-]

HugoCortell@reddit

Europeans can buy a used 3090 (half burnt) for the same price.

[-]

JLX_973@reddit

I bought my 3090 for 400€ in France second-hand, it's like new 😇

[-]

Jolly-Event7578@reddit

Probably just before the a.i boom and you were very lucky , average price for a 3090 non ti in whole europe it-s 550-900€

[-]

redditorialy_retard@reddit

ti doesn't really matter no in AI?

got mine for about 600 bucks in 2025

[-]

ImportancePitiful795@reddit

Actually there is a company selling them around €890+VAT

[-]

Far-Chest-8821@reddit

Which? Source?

[-]

ImportancePitiful795@reddit

Intel Arc Pro B70 AI & Workstation 32GB (33P01IB0BB)

€869.80 actually. (+VAT depending your country).

[-]

siete82@reddit

Wtf, the GPUs are half price than in Spain, including VAT. Is that store legit? Do they ship to all the EU?

[-]

ImportancePitiful795@reddit

Is a store in Denmark belonging to a Norwegian company if remember correctly. Google it, they are legit having very competitive prices and good reviews.

[-]

sammyranks@reddit

With your 24% VAT, don't expect price miracles.

[-]

SexyAlienHotTubWater@reddit

If you're using it for work you don't pay VAT.

[-]

hahncholo@reddit

Can it run Crysis

[-]

FastHotEmu@reddit

2006: Can it run Crysis?
2026: Can it vibe code Crysis?

[-]

ImportancePitiful795@reddit

Yes. 😊

[-]

CrypticZombies@reddit

Intel ones are garbage. Nothing to brag about on this one sorry

[-]

somethedaring@reddit

What can it be used for? I do a lot of cuda

[-]

Nugs_@reddit

It’s an unfair advantage fr

[-]

skrshawk@reddit

I am grateful I live within daytrip distance but not right near a MC. I would blow my budget on stuff I don't need.

[-]

Prize_Negotiation66@reddit

And I'm born in the most sanctioned country in the world, locked up like a prisoner. There is no justice.

[-]

MerePotato@reddit

Worlds only North Korean Reddit user

[-]

mizhgun@reddit

North Korea is not even close to be the most sanctioned country in the world.

[-]

cloudcity@reddit

Wait wut

[-]

Iwaku_Real@reddit

I guess >!Russia!<

[-]

Slave669@reddit

Crying in Australian.

[-]

last_llm_standing@reddit

dude i. have a microcenter 15 mins away from me, i never saw any such deals. how do you find one?

[-]

titpetric@reddit

Inspect element, edit element, delete prefix number, take screenshot. Best I can get is a RX7900 at about the same price, but, 24GB vram and its an XL size card with 3 fans

[-]

Iwaku_Real@reddit

I cant figure out what you mean idk why

[-]

gigaflops_@reddit (OP)

Bruh if you need to prove it to yourself, change the store location until you find somewhere that has it. Mine has had 2-3 days since release where they were in stock. MicroCenter is the 🐐 for not allowing online orders and limiting in-store purchases to one-per-customer per day to stop scalpers. I bought a couple of the 3000 series cards there back in the day.

[-]

Frosty-Cup-8916@reddit

This is a real deal tho

[-]

last_llm_standing@reddit

ill prolly get just the fan

[-]

titpetric@reddit

Safe bet, cash in on that fan shortage

[-]

pfn0@reddit

In stock here in silicon valley (if it were good, it'd be sold out instantly).

[-]

gigaflops_@reddit (OP)

Check the website every day, and make sure you select your store's location or it doesn't show up (since it's in-store only). My location has gotten stock at least 2-3 times since release.

[-]

GreenDavidA@reddit

I look at it as a destructive curse … for my wallet.

[-]

Open-Impress2060@reddit

How would this perform in gaming vs like a 9070xt?

[-]

intergalactic_watch@reddit

what a weir title

[-]

Polaris_debi5@reddit

Excellent VRAM/price ratio for the B70. However, with OpenVINO 2026.0 and vLLM optimizations, Intel's scaling with llm-scaler for multi-GPU setups is starting to look better for massive models on a budget.

[-]

IntelligentFire999@reddit

Can I join you in prayer next time? I live about 30 mins away from a Microcenter as well.

[-]

Proof_Nothing_7711@reddit

Hello, one question I’ve a Rx 7900 XTX with 24 gb vram. My interest is more for big context than pure speed. Could me useful for me use this Intel Arc Pro B70 with 32 gb VRAM ? Speed will change significantly? Thanks

[-]

Blastronomicon@reddit

Help these guys

https://geohot.github.io/blog/jekyll/update/2023/05/24/the-tiny-corp-raised-5M.html

[-]

ImportancePitiful795@reddit

The Intel stack atm needs to use vLLM and is a gen behind, but 32GB vs 24GB there is no competition.

Though at this point the R9700 make good case also.

[-]