Intel Announces the Arc B-Series Graphics Cards

[TheLostSwede]

Today, Intel announced the new Intel Arc B-Series graphics cards (code-named Battlemage). The Intel Arc B580 and B570 GPUs offer best-in-class value for performance at price points that are accessible to most gamers, deliver modern gaming features and are engineered to accelerate AI workloads. The included Intel Xe Matrix Extensions (XMX) AI engines power the newly introduced XeSS 2, comprised of three technologies that together increase performance, visual fluidity and responsiveness.

"The new Intel Arc B-Series GPUs are the perfect upgrades for gamers. They deliver leading performance-per-dollar and great 1440p gaming experiences with XeSS 2, second-generation ray tracing engines and XMX AI engines. We're delighted to be joined by more partners than ever so that gamers have more choice in finding their perfect design." -Vivian Lien, Intel vice president and general manager of Client Graphics



[Editor's note: Our preview of the Arc Battlemage Series is now live]


About the New Architecture: The Intel Arc B-Series GPUs use Intel's latest Xe2 architecture, optimized to deliver improved efficiency and higher performance per core with less software overhead. Second-generation Xe-cores deliver the solid compute capabilities required for modern workloads and include high-performance XMX AI engines. New Xe-cores are supported by more capable ray tracing units, better mesh shading performance and improved support of key graphics functions to significantly increase performance efficiency with the latest game engines.

About XeSS 2: XeSS 2 now includes three technologies: XeSS Super Resolution, XeSS Frame Generation and Xe Low Latency. XeSS Super Resolution is the key technology that underpinned first-generation XeSS, which has offered AI-based upscaling for more than two years and now supports more than 150 games. New AI-powered XeSS Frame Generation adds interpolated frames using optical flow and motion vector reprojection to provide higher fluidity gaming. And new Xe Low Latency integrates with the game engine to provide faster response to gamers' inputs. With all three technologies activated, XeSS 2 is capable of increasing the frames-per-second output by up to 3.9x to deliver high performance in demanding AAA games.

Why It Matters: Intel's graphics intellectual property (IP) enables modern graphics experiences at scale for a range of devices, including graphics cards, laptops and handhelds. Intel Arc B-series GPUs benefit from a mature software environment thanks to continuing investments across Intel's graphics portfolio.

Compared with the previous generation, the Intel Arc B-Series GPUs offer 70% better performance per Xe-core and 50% more performance per watt. The B580 GPU, when compared to the Intel Arc A750 GPU, is on average 24%5 faster at 1440p with some games up to 78% faster. When compared to the competition, the Intel Arc B580 GPU offers up to 32% better performance-per-dollar.

With 12 GB of dedicated GPU memory for the Intel Arc B580 and 10 GB for the Intel Arc B570, gamers can expect high-performance gaming at 1440p ultra quality settings, enhanced by AI-powered XeSS 2 technologies.

New Intel Graphics Software provides access to display settings including color and scaling modes and variable refresh rate (VRR) support. 3D graphics settings include a frames-per-second limiter and a driver-level low-latency mode. Performance controls include basic and advanced overclocking and metrics powered by the widely adopted open source PresentMon, which now includes support for frame generation and latency measurements.

The Details: Features of the Intel Arc B-Series GPUs include:

Powerful Media Engines: Intel's leading Xe Media engine with twin multi-format transcoders (MFX) is designed to support faster, higher-quality creation and playback of multimedia (video). The GPUs also include expansive hardware encode and decode codec support, such as AV1, H.265 including 4:2:2 10-bit, H.264 and VP9.
Support for AI Playground: Intel's all-in-one, easy-to-use free AI starter application is designed to run generative AI workloads locally to allow users to easily perform text-to-image generation, photo editing and upscaling, and customize chatbots with their own data.
Game On Driver Support: Intel Arc B-Series GPUs will continue to receive optimized software and performance for the most popular games on launch day.

About Availability: The Intel Arc B580 Limited Edition graphics card and models from add-in board partners Acer, ASRock, GUNNIR, ONIX Technology, MAXSUN and Sparkle will be available starting Dec. 13 from $249.

Intel Arc B570 graphics cards from the same add-in board partners will be available starting Jan. 16, 2025, from $219.

View at TechPowerUp Main Site

 

[Guwapo77]

I don't like Intel's hardware, but they damn sure made some of the cleanest looking GPUs to have ever graced this sector.

 

[wNotyarD]

Why release the B570 only a month later?

 

[Nordic]

The budget market just got a lot more interesting.

 

[AcE]

So late to the market that now it will compete with the new architectures that it was not supposed to be compete with. It won’t do that well I fear. People thought Arc is dead because it was so late.

 

[Bobaganoosh]

New budget king? Seems like a decent upgrade on the old GTX 1660/50 series that still hasn't quite gone away from the budget-build sector.

I definitely want to see when the B780(770?...750?...not sure what highest might be) launches and where it lands from a price/performance perspective. It seems like with it not even part of this announcement that whatever the answer is, it will include "late".

 

[Durvelle27]

Price seems very stupid

why get the 570 when the 580 is only $30 more

 

[theouto]

Price seems very stupid

why get the 570 when the 580 is only $30 more

Because they want you to spend the $30 that it costs to go from the 570 to the 580.

 

[Onasi]

@theouto
Yeah, it’s a classic upsell strategy plus the chips that aren’t good enough for the B580 need to go SOMEWHERE, but I suspect pricing it even lower would cut too much into margins.

 

[AcE]

Because they want you to spend the $30 that it costs to go from the 570 to the 580.

AMD didn't even release the partially broken 7600 chips (so basically 7600 vanilla already is the "XT" and the "XT" is just a XT with doubled vram), I think it all went into laptop or workstation products.

 

[chrcoluk]

I don't like Intel's hardware, but they damn sure made some of the cleanest looking GPUs to have ever graced this sector.

Single 8 pin connector was nice and visible lol.

 

[lexluthermiester]

Single 8 pin connector was nice and visible lol.

Yeah, I think they're deliberately highlighting that. I like it! Intel is doing things right.

 

[AggressiveCombo]

And they have 2 more gens to release


Please be good

 

[Dawora]

New budget king? Seems like a decent upgrade on the old GTX 1660/50 series that still hasn't quite gone away from the budget-build sector.

I definitely want to see when the B780(770?...750?...not sure what highest might be) launches and where it lands from a price/performance perspective. It seems like with it not even part of this announcement that whatever the answer is, it will include "late".

atleast 1660 drivers works no suprises..
whit intel too many game still suffer bad drivers

if i have to buy Amd or Intel gpu i bought AMD

 

[Vincero]

@theouto
Yeah, it’s a classic upsell strategy plus the chips that aren’t good enough for the B580 need to go SOMEWHERE, but I suspect pricing it even lower would cut too much into margins.

Absolutely, indeed they do - I have no doubt that's the case. As mentioned though, the gap between the 580 and 570 is small so I'm guessing yields are pretty decent.

I love how it used to be / still is the normal MO with lower tier parts - downclock it, impose lower max TDP and power limits and make it a laptop part - obviously for parts that fail to operate at intended performance levels - and boom mobile Intel B580M. Hell, it's even only a PCIe x8 part - already circling that middle road of having a split personality in terms of where it belongs (desktop/laptop).

For those that can't even do that then to the B570 you go.... surprisingly, when you consider how much even Nvidia and AMD have harvested some GPU dies and cut down / fused off the (non)functioning components, the B570 still leaves a lot of working parts - retains 90% of shaders and TMUs but same ROP count.... I wonder if we'll see some lower items over time like a B560/550/etc., as more chips roll out and some aren't even able to make a B570 part.

E.g. the Nvidia GP104 (used for the GTX 1080) dies were harvested a lot for lower tier products with much bigger jumps - the GTX 1070 is only 75% of the GTX 1080 (again except for ROP count). In some cases, maybe they were to fulfill orders, but the likelihood is they were making the best out of what they got, e.g.:

Tier: Product - Shaders:TMUs:ROPs
Good chips: GTX 1080 - 2560:160:64
Bad chips: GTX 1060 3G - 1152:72:48

If it wasn't for the additional logic for NVENC, display output, PCIe interface, etc., it would literally be over 50% of the chip inactive....

Basically Intel have a lot of headroom for more filler products - not that people want them specifically - maybe consign the waste of money type of cards like the RX 6400 to history.

 

[Launcestonian]

Does anyone know if & when the B300 series will launch? I'm interested in dGPU that only uses motherboard slot power.

 

[AcE]

Tier: Product - Shaders:TMUs:ROPs
Good chips: GTX 1080 - 2560:160:64
Bad chips: GTX 1060 3G - 1152:72:48

GTX 1060 3G isn't the same chip as 1080, it's not GP104, it's a part deactivated GP106. https://www.techpowerup.com/gpu-specs/geforce-gtx-1060-3-gb.c2867

 

[Vincero]

GTX 1060 3G isn't the same chip as 1080, it's not GP104, it's a part deactivated GP106. https://www.techpowerup.com/gpu-specs/geforce-gtx-1060-3-gb.c2867

Normally that would be the intended approach...
However I'm guessing as it's a cut down part in either case, Nvidia didn't care about which harvested die is used as long as there is an OEM with a matching PCB to stick it on:

NVIDIA GeForce GTX 1060 3 GB GP104 Specs

NVIDIA GP104, 1708 MHz, 1152 Cores, 72 TMUs, 48 ROPs, 3072 MB GDDR5, 2002 MHz, 192 bit

www.techpowerup.com

 

[AcE]

Normally that would be the intended approach...

Exceptions prove the rule, my friend. It was 99% GP106. GP104 in a 1060 is a great waste, broken silicon is mostly used as 1070s.

 

[Vincero]

Exceptions prove the rule, my friend. It was 99% GP106. GP104 in a 1060 is a great waste, broken silicon is mostly used as 1070s.

True, but kind of missing the point - the B570 is only a 10% reduction in shader compute power and texturing capability - there is scope for the same die to pad out some lower offerings at further cost reduction, offering hopefully something better than the steps between the existing Arc A3xx series and the A5/7xx products.

The previous Arc A580 was a cut-down A7xx GPU die - this time they have a specific IC for this range. I'd assume they'll have a specific IC for the B3xx series and laptop products, like they did with the A3xx range, as even this B570/580 isn't exactly small / low power.

That said, if the yields are really good then lower tier products are a loss of profit if they are artificially binned down. Maybe whatever the B3xx series brings will hopefully be a bit more potent for Intel (relatively speaking) than the A3xx series and actually be a compelling option at the low-end - I don't think there has been a compelling (i.e. usable, doesn't completely suck, and even if around $100 actually feels like it's worth it) low-tier graphics product since the GTX 1050.

 

[AcE]

True, but kind of missing the point - the B570 is only a 10% reduction in shader compute power and texturing capability - there is scope for the same die to pad out some lower offerings

That's kinda you missing the point. The lower they reduce the die compute wise, the lower they have to sell it for. That's why that is typically not done. Mostly chips are sold as "full" and "partially deactivated" meaning 10% or so, to make it usable because most broken silicon isn't much broken, just small parts. After that it makes more sense to use a different, smaller die for different SKUs such as a theoretical B560 and 550. That is how it is typically done, rarely they use same die for 3 different SKUs as this is a money losing business then. Nvidia can typically afford it more than other companies to cut a die harder and sell it, as their margins are bigger anyway. Companies like AMD and Intel will barely be able to afford it, their margins already start way lower typically.

 

[Vincero]

That's kinda you missing the point. The lower they reduce the die compute wise, the lower they have to sell it for. That's why that is typically not done.

Wow, you literally quoted just the bit of my post and ignored the rest where I acknowledge that exact point.

That a) they have a specific IC for the B5xx range now, and b) artificially binning chips for lower SKUs is a profit losing activity.

I'm not suggesting they do a 50% non-functional die product, simply that a bridge product could exist between whatever the best B3xx may be and the B570 - I find it highly likey they will have chips with more defects than just the reductions made for the B570 - there will likely be chips with bad ROPs for example that could be further binned to a lower product also. Not selling it at all in another crappy SKU is even more of a money looser than just artificially binning chips.

 

[AcE]

I'm not suggesting they do a 50% non-functional die product, simply that a bridge product could exist between whatever the best B3xx may be and the B570 - I find it highly likey they will have chips with more defects than just the reductions made for the B570 - there will likely be chips with bad ROPs for example that could be further binned to a lower product also.

I don't know why you make so much mental gymnastics for such a small thing that is unlikely to happen, but okay. Whatever. As I said, mostly it's 2 SKUs, 1 full activated, 1 10-20% deactivated and that's it, no need to discuss this as it's a historical fact, aside from the technical side that supports it. Why, because TSMC is so good, you simply don't need more than 2 SKUs to sell most of what is produced, there will be barely any leftover, and thus no 3rd SKU for the same chip. Maybe with Samsung foundry this would be different, more really broken chips, more SKUs.

Not selling it at all in another crappy SKU is even more of a money looser than just artificially binning chips.

I don't think it is artificial. I think they bin the chips to sell more or less all of what is produced with the wafer. If the leftover, of even more broken chips, is 10%, they probably throw it away, you're basically talking about those <10% that are leftover. With big dies, there won't even be a lot of them, it was different with "Ellesmere" for example (RX 480/580), with so many small chips this under 10% number will be significant enough, so it was a third SKU, the RX 460/560, warranted. Won't happen here, as the chips eat too much of the wafer, making it not worth wile.

 

[Vincero]

I don't think it is artificial. I think they bin the chips to sell more or less all of what is produced with the wafer. If the leftover, of even more broken chips, is 10%, they probably throw it away, you're basically talking about those <10% that are leftover. With big dies, there won't even be a lot of them, it was different with "Ellesmere" for example (RX 480/580), with so many small chips this under 10% number will be significant enough, so it was a third SKU, the RX 460/560, warranted.

Either throw them or they become an OEM special - witness the spread of cut-down RX 580s, 1060 5GB, 4090D, etc. - yeah sure the last one is due to 'export restrictions' but it wouldn't surprise me if there's a few extra defects on them vs the normal 4090. Also, the B580 die is only fractionally bigger than the Polaris die (~50 mm²) used on the RX 480/580, so why wouldn't the same rules apply...?
At the end of the day, if you have the PCB design already there, shipping any inventory that are low/no-profit is better than a total loss. Look at how AMD have been hoarding Zen3 CPUs that may not have made the cut and selling them as gradually worse/cheaper AM4 CPUs - I would expect GPU manufacturers to do the same as a matter of course rather than just bin anything.

Fair enough, a lot of this is anecdotal as we have no real knowledge of the non-public roadmap and yield data, etc., and whilst TSMCs 5nm yields may be great as an average, not every wafer is as good as the best. Just because they are not likely to do it doesn't mean Intel don't have the option to if push come to shove.

Now the review results are actually out, it actually looks like they could easily drop the features down to 75-80% of the full B580 and still compete in some titles with the RTX 3060/4060 - sell at a price point where you aren't taking a hit and you can offset the losses vs throwing the dies away.
Intel may not ship these, but AIBs might be given the option to sell them in certain markets - a practice that has gone on for a while.