Intel Arc B580 Selling Like Hot Cakes, Weekly Restocks Planned

[Lionheart]

Well this is great news, Intel needed some positive light, judging by the reviews it looks like a solid card for the price, now just hope Intel can do the same with their CPU's....

 

[Scrizz]

Taken out of context, you miss my point.

The rest of your post was about you wanting AMD to lower their prices so "it would grab your attention." It still doesn't change the fact that there is nothing more to talk about at this moment.

 

[mechtech]

"as well as arguably better performance than its primary competitors, the RTX 4060 and the RX 7600."

Wouldn't its closest competitor on amd side be an rx6700xt/6750xt?

 

[yfn_ratchet]

Glad to see all the warm reception actually made enough of a splash for these cards. The B500 cards look like a litmus test for the rest of the range (I believe I've said something similar in a different thread but I digress) and with the absolute hit the B580 is I'm optimistic about the adjacent 300/700 segments.

Slot-powered B380 for cheap-ass HTPC emu-majiggers and self-hosted streaming, B310 to succeed the amiable A310 as a display adapter with some media processing grunt, and of course the B770 as a card I'd be happy to own for my main rig, granted it offers ~200% the performance of my mildly aged RTX 3060. It's a bit of a steep ask, but it's doable given the B580 already outpaces it in its own niche.

 

[londiste]

"as well as arguably better performance than its primary competitors, the RTX 4060 and the RX 7600."
Wouldn't its closest competitor on amd side be an rx6700xt/6750xt?

In practice, the competition is generally determined by MSRP price and current generation.
RX67x0XT is a previous-gen product of higher segment (wasn't it $500 range MSRP?) that has come down in price with clearing stock and I think there was a re-release to some degree?

 

[friocasa]

MLID said Intel was losing 20 bucks per card sold if I remember correctly

But even if that was the case, ain't that much and they could increase the price a little bit to OEMs, they would accept if there's still money be made and the alternative is no stock at all

The problem now is how long will it take to AMD and Nvidia to respond, B580 might become a horrible value proposition as soon as next month, so it's a risky bet, both for Intel(ordering new silicon to be made at TSMC), and OEMs(buying chips to make cards)

 

[boomheadshot8]

For 300€ you can buy a 3070
also b850 and A770 is 350€ in france not cheap at all ...

 

[Nostras]

You do know that AMD, Intel, Nvidia are American companies right?

You are aware AMD and Nvidia are pricing the cards properly right?

 

[Calenhad]

A message to GGforever:

>>...as well as arguably better performance than its primary competitors, the RTX 4060 and the RX 7600...

Really?

The actual performance of Intel Arc B580 is between NVIDIA GeForce RTX 2080 SUPER and NVIDIA GeForce RTX 3060 Ti.

Intel Arc B580 Specs

Intel BMG-G21, 2670 MHz, 2560 Cores, 160 TMUs, 80 ROPs, 12288 MB GDDR6, 2375 MHz, 192 bit

www.techpowerup.com

You do realise that both RTX 4060 and RX 7600 are in the same link you used as proof? At 94% and 93% performance... So what part of the initial statement is wrong?

Single Precision ( FP32 ) Processing Powers of these GPUs are as follows:

NVIDIA GeForce RTX 2080 SUPER = 11.15 TFLOPs
INTEL Arc B580 = 13.67 TFLOPs
NVIDIA GeForce RTX 4060 = 15.11 TFLOPs
NVIDIA GeForce RTX 3060 Ti = 16.20 TFLOPs
AMD Radeon RX 7600 = 21.75 TFLOPs

I'm Not against of Intel and more competition is better than NVIDIA-AMD monopoly.

There are real numbers, like Number of Streaming Processors and, as a result, a Peak Processing Power ( PPP ) of a GPU and you can Not speculate or exceed these numbers.

Unfortunately, GGforever clearly speculated.

Tell me you don't know how a GPU works, without telling me you don't know how a GPU works.

Those are raw floating point operations per second numbers. Which would be a wonderful measure if GPUs only did floating point calculations.

Let me throw a different metric at you:
NVIDIA GeForce RTX 2080 SUPER = 288.0 GB/s
INTEL Arc B580 = 456.0 GB/s
NVIDIA GeForce RTX 4060 = 272.0 GB/s
NVIDIA GeForce RTX 3060 Ti = 448.0 GB/s
AMD Radeon RX 7600 = 288.0 GB/s

Oh man, the B580 is so much better than all those other cards. Only the RTX 3060 Ti comes close. I mean, it is almost 160% faster than a RX 7600! What a bargain!

Oh wait, memory bandwidth is also just one part of the puzzle.

 

[londiste]

All these specs have become really bad indicators as of late.

Processing power actually does scale pretty well with gaming performance, except the part that @AcE pointed out above where AMD and Nvidia both have dual ALUs in a shader. Due to various reasons the performance boost from having dual ALUs does exist generally seems to be in the 30% range (a bit more on Nvidia and a bit less on AMD side for now).

The big caches added right next to memory controllers on CPUs lower the impact of low memory bandwidth considerably. Until they don't which usually happens at large resolutions. AMD did that 2 generations back, Nvidia with the last generation and it has allowed both of them to reduce the VRAM bus width by about a third compared to previous gen of the same segment. Meaning in this case the B580 with its 192-bit bus competes against 128-bit bus from AMD and Nvidia. Yet, at least at 1080p and mostly 1440p that relative lack of memory bandwidth really does not show up. It kind of seems to at 2160p but that is beyond these cards anyway.

 

[Vayra86]

It all sounds nice, but as long as the product itself isn't available anywhere, it doesn't really mean anything.

But it is...

 

[AusWolf]

But it is...

Where? One major UK retailer lists two models, but both are on pre-order with no ETA. The rest of them don't sell Intel GPUs at all.

 

[freeagent]

You are aware AMD and Nvidia are pricing the cards properly right?

Go back to sleep.

 

[TechBuyingHavoc]

If they need a 300mm² *5nm* bleeding edge foundry (for) GPU(s), to compete with 1-2 gen old stuff, it just shows they *can't* bring faster stuff anyway, huge chips, super expensive to produce, not efficient, like this one. Can it even be done? Would it make them money? It will only compete with 7800 XT-ish anyway, so I'm highly sceptical if Intel can even bring something substantially better than this, without losing money on it. They're already losing money on this. This is like a 7800 XT size-wise, but has way less performance and costs ~half of it.

I suppose the thinking inside Intel is that yes the current product is not competitive (profit-wise) but we have a product roadmap that will close the gap and actually make money for us, improve our mindshare, and maybe even fill future production volume in our underutilized fabs. I can see all of that, the b580 is significantly improved vs its predecessor. But Intel may not have the cash reserves to get to that promised future.

It really feels like Intel is 5 years late to this race, the time to start dipping your toes in the GPU ocean was back in 2015 since it would take years for that first product to come out. Intel spent $7 billion on McAfee back in 2010. Then they spun it back out in 2017. That $7 billion would have gone a far way into getting a GPU product line up and running.

 

[AcE]

So there we are @AcE , you don't want to act like one, but you do nothing other than repeating youtubers bullshit, you can't bring a single of you own thought.

Wtf? I don't even watch youtube.

The big caches added right next to memory controllers on CPUs lower the impact of low memory bandwidth considerably. Until they don't which usually happens at large resolutions.

It works so far very well for Nvidia, with AMD it works very well with 7800 XT and 7900 XTX, the others have some bandwidth constraints. Also the 6900 XT lost unusually much at 4K that's why it had hard times competing with 3090 at 4K but killed it at 1440p and 1080p. Yep, Radeon was partially faster back then depending on res.

I suppose the thinking inside Intel is that yes the current product is not competitive (profit-wise)

Intel is a true mess, as the other guy pointed out, these chips are very expensive, this wasn't the case before when they used their own nodes. So expensive + expensive to produce + bad performance = utter fail. .....

 

[rv8000]

Agreed; and let’s not forget as iv said in other posts BMG is not an iteration of alchemist. It’s an entirely new design. The fact they could even reach this level of performance on not only gen 2 but on a totally new architecture is staggering to me. I know some will argue the performance is bad compared to there pedigree citing there iGPUs. However it was never as simple as “just make little GPU big GPU.” From feature levels to features themselves it was a massive lift. That’s just the cards. As someone who owned a A380 like the first week of ARC the driver team? Well I haven’t seen dedication or improvement like that literally ever. From AMD or Nvidia.

For the simple reason that the hardware moved at the pace of software. Intel is playing catchup in every conceivable way. From hardware to drivers, and they aren’t competing with someone who has a week head start, someone they can overtake by running there horse into the night. Nvidia and AMD have easily 2+ decades on Intel in discrete land.

They are in there infancy; there performance and hurdles they discuss publicly kind of solidify that. I don’t think B7xx will be earth shattering and I don’t think celestial will be either. But they are trying. Even Tom mentioned Celestial is in the hands of engineering but the architects are already on Druid.

It’s easy to pull for the consumer; that is a big part of having a third player. But I am hoping Nvidia and AMD catch the hint. I know many want lower prices, and who doesn’t? I am hoping though that Nvidia and AMD get the fire back in them. The drive to stay ahead. Improving and adding useful features. Then everyone will win no matter what brand you want to buy.

It won’t happen next year; but between the 3 of them I am hopeful it will start in the next 5 years. I think the hidden win is it shows that entering the market can be done. Even if Intel ultimately decides to pull out. Who knows who was watching and decided they will try there hand.

Except the only “useful” features being pushed now are IQ destructive upscaling techniques to cover up for the majority of pointless Ray Tracing implementations. All of said features are quickly becoming a crutch the gaming industry is relying on over any sort of actual optimization; latest personal example to me was the MH:Wilds Beta, yes it’s a beta, but running the game without use of DLSS/DLAA and or turning off TAA made the game look like actual garbage. But that just goes along with the “minimize time/cost/quality, and maximize profit” industry mindset.

 

[londiste]

Also the 6900 XT lost unusually much at 4K that's why it had hard times competing with 3090 at 4K but killed it at 1440p and 1080p. Yep, Radeon was partially faster back then depending on res.

That is actually a good example for the problem with wider bus vs large cache. Yes, this is not necessarily the only factor but all data suggests this is the major one.
While very similar in performance also quite similar in other specs, 3090 has a 384-bit memory bus and something like 6MB L2 cache while 6900XT has newfangled 256-bit memory bus and comparatively massive 128MB L2 cache. After that Nvidia very quickly did the same thing with bus size and cache with 40 generation.

 

[Vayra86]

Where? One major UK retailer lists two models, but both are on pre-order with no ETA. The rest of them don't sell Intel GPUs at all.

Here in NL, they are up for order, but I'm not paying +70~80 EUR on top of MSRP for this Several stores have them in stock too.

Vergelijk videokaart-prijzen en deals - Tweakers

Ontdek de beste prijzen en de nieuwste videokaarten. Bekijk en vergelijk videokaarten op Tweakers en maak de beste keuze voor jou.

tweakers.net

 

[Nostras]

Here in NL, they are up for order, but I'm not paying +70~80 EUR on top of MSRP for this Several stores have them in stock too.

Vergelijk videokaart-prijzen en deals - Tweakers

Ontdek de beste prijzen en de nieuwste videokaarten. Bekijk en vergelijk videokaarten op Tweakers en maak de beste keuze voor jou.

tweakers.net

View attachment 376470

For our American friends, if you want a comparison just replace the euros with a dollar. The 4060 and 7600 are almost exactly the same price in euros here (including tax) as dollars in US. Assume it has a 320$ MSRP in europe.

 

[SSGBryan]

For our American friends, if you want a comparison just replace the euros with a dollar. The 4060 and 7600 are almost exactly the same price in euros here (including tax) as dollars in US. Assume it has a 320$ MSRP in europe.

That also includes VAT, does it not? Your VAT is higher sales tax in the US.

When I include sales tax, the AsRock Challenger B580 goes to $296.

Nah, that would be 7800 XT/4070 Super, upper mid range cards do that, they can do even 4K (with some downsides) and aren't that expensive. I'm also pretty sure they sell the most. Mid range.

The B580 is a 1440p high card - that is 90% of the market.

Most most aren't dropping $500 on a video card. People posting here aren't most gamers.

"as well as arguably better performance than its primary competitors, the RTX 4060 and the RX 7600."

Wouldn't its closest competitor on amd side be an rx6700xt/6750xt?

View attachment 376383

The 6700 series draws nearly 60% more power, for roughly the same performance - and they are still more expensive than a B580.

 

[Nostras]

That also includes VAT, does it not? Your VAT is higher sales tax in the US.

When I include sales tax, the AsRock Challenger B580 goes to $296.

You're overthinking it.
7600 is 260€

Vergelijk videokaart-prijzen en deals - Tweakers

Ontdek de beste prijzen en de nieuwste videokaarten. Bekijk en vergelijk videokaarten op Tweakers en maak de beste keuze voor jou.

tweakers.net

4060 is 290€

Vergelijk videokaart-prijzen en deals - Tweakers

Ontdek de beste prijzen en de nieuwste videokaarten. Bekijk en vergelijk videokaarten op Tweakers en maak de beste keuze voor jou.

tweakers.net

B380 is 320€

Vergelijk videokaart-prijzen en deals - Tweakers

Ontdek de beste prijzen en de nieuwste videokaarten. Bekijk en vergelijk videokaarten op Tweakers en maak de beste keuze voor jou.

tweakers.net

If we remove taxes we get 215€, 240€ and 265€ respectively. Converted to $ we get 225$, 250$ and 275$.

If we look at American prices we get 250$ for the 7600

https://pcpartpicker.com/products/video-card/#c=554&sort=price&page=1

295$ for the 4060

https://pcpartpicker.com/products/video-card/#sort=price&c=552

and ???$ for the B580 (but let's assume 250$ once they're back in stock)

https://pcpartpicker.com/products/video-card/#sort=price&c=585

GPU in $	7600	4060	B580
US	250$	295$	250$
Europe (NL)	225$	250$	275$

With a major caveat that the B580 is out of stock in the US.
Perhaps this illustrates my frustration?
Depending on your view you could argue that either Intel is overpricing the card in Europe or AMD and Nvidia are giving major discounts in Europe.
I could also argue that the 7600 and 4060 have been having flash sales here as well, the 7600 could've been had for as cheap as 235€ and the 4060 250€ in November.

ACER Predator BiFrost AMD Radeon RX 7600 OC, 8GB GDDR6 voor €235,99 bij NBB.com | Pepper.com

nl.pepper.com

MSI GeForce RTX 4060 VENTUS 2X BLACK 8G OC voor slechts €249,10 bij Bol | Pepper.com

nl.pepper.com

Perhaps now @freeagent understand my pov instead of telling me to go to sleep (wtf?)

 

[AcE]

That is actually a good example for the problem with wider bus vs large cache. Yes, this is not necessarily the only factor but all data suggests this is the major one.
While very similar in performance also quite similar in other specs, 3090 has a 384-bit memory bus and something like 6MB L2 cache while 6900XT has newfangled 256-bit memory bus and comparatively massive 128MB L2 cache. After that Nvidia very quickly did the same thing with bus size and cache with 40 generation.

The fact is, even 128 MB was not enough, back then AMD had a presentation and the "curve" of the L3 cache was sweet spot with 128 MB but it ofc scaled further with even more cache, so it would've probably needed 160-192 MB L3 Cache to make it behave like a 3090 (no caveat with 4K). RTX 40 series has no such issues, the GPUs normally scale like expected, with RX 7800 XT and 7900 XTX this is the same, both have a lot of normal bandwidth, the L3 caches aren't very big.

The B580 is a 1440p high card - that is 90% of the market.

It's not, it's firmly a 1080p card, 1440p is a stretch for a weak card like that. 4060 Ti is way stronger, that's considered a "weaker" 1440p card.

 

[londiste]

The fact is, even 128 MB was not enough, back then AMD had a presentation and the "curve" of the L3 cache was sweet spot with 128 MB but it ofc scaled further with even more cache, so it would've probably needed 160-192 MB L3 Cache to make it behave like a 3090 (no caveat with 4K). RTX 40 series has no such issues, the GPUs normally scale like expected, with RX 7800 XT and 7900 XTX this is the same, both have a lot of normal bandwidth, the L3 caches aren't very big.

From the cards we have - not really. AMD reduced cache in the 7000 series. While 6800XT/6900XT had 128MB cache for their 256-bit bus, 7800XT has 64MB cache for similar 256-bit bus. And even 7900XTX with its 384-bit bus only gets 96MB. Basically - 6000 generation mid-range and up had 16MB cache for each 32-bit memory bus/chip. 6600 and lower had 8MB. 7000 generation settled to 8MB for the entire range.

Nvidia settled at the same or even lower amounts of cache although they seem to have put the cache in a slightly different place and not that directly next to memory controllers.

I am sure these amounts are optimized for something that is not necessarily overall best performance but the boost to memory bandwidth seems to behave similarly for AMD 6000 and 7000 generation as well as Nvidia's 40 gen. This hints at a point in cache amount that is enough and further increases do not help (as much if at all).

 

[AcE]

From the cards we have - not really. AMD reduced cache in the 7000 series. While 6800XT/6900XT had 128MB cache for their 256-bit bus, 7800XT has 64MB cache for similar 256-bit bus. And even 7900XTX with its 384-bit bus only gets 96MB. Basically - 6000 generation mid-range and up had 16MB cache for each 32-bit memory bus/chip. 6600 and lower had 8MB. 7000 generation settled to 8MB for the entire range.

The cache sizes aren't that important, effective bandwidth is (for the umpillionth time ) - this is a combination of regular bandwidth plus cache bandwidth, and it was ALWAYS like this, just more important now that the cache sizes are bigger. Important is just one thing: that the card doesn't lose too much performance on higher resolutions. Everything else is irrelevant and extremely moot to discuss.

I am sure these amounts are optimized for something that is not necessarily overall best performance

A generalised comment like this can only be wrong. RTX 40 cards have shown to work very well, AMD for the most part, just not always and not all cards. This is some luxury critic: the 6900 XT for example loses like 5% more performance in 4K than it should. Not a big issue.

Nvidia settled at the same or even lower amounts of cache although they seem to have put the cache in a slightly different place and not that directly next to memory controllers.

Not quite, depends which cards you mean. the 4090 has a ton of cache for example, the 4080 has more than enough. etc. And L2 cache can not directly be compared to L3 cache since it is simply better.

 

[londiste]

The cache sizes aren't that important, effective bandwidth is (for the umpillionth time ) - this is a combination of regular bandwidth plus cache bandwidth

Can you explain what is effective bandwidth?

Not quite, depends which cards you mean. the 4090 has a ton of cache for example, the 4080 has more than enough. etc.

No, 4090 does not have a ton of cache. What makes you say that? 4090 has 72MB, 4080 and 7800XT have 64MB, 7900XT has 80MB, 7900XTX has 96MB.

And L2 cache can not directly be compared to L3 cache since it is simply better.

Not going to google right now but IIRC AMD caches have more bandwidth almost throughout the whole thing. Latency varies with the different cache levels on one or the other but is at least comparable.