Intel Arc "Battlemage" BMG-G31 B770 GPU Support Lands in Mesa Driver

[Solaris17]

I still really hope they accelerate the Celestial launch. I am hoping they are doing this to fast ship the remaining silicon. I am excited for this and Intel to keep pushing despite resistance in the GPU industry because we all need it, even those that wont ever buy one; but there release schedule is not great.

 

[Squared]

16GB 9060XT is probably the best (well, only) deal going right now if you need > 1080p at sub $400. I just would have expected it to beat a 7700 XT, which is really just a tier above it, and it doesn't really do that.

It's just not enough of an upgrade for me personally vs my 6700 XT. I need to go up one more tier to make any expenditure worthwhile, but that jumps to $600+.

Sad thing is, coulda bought a 7800 XT a year ago for $400.

That is a bummer. There was like 1 day where I saw the 7000-series options and thought it might be time to buy but I passed it up and then supply disappeared. Multiple generations have passed me like this which is why I rushed out to get the 8600 9060 XT. I was coming from a 5600 XT so this is a worthy upgrade and a better fit inside my expansion slot and power constraints than the 7700 XT was, and cheaper too.

To be clear, I'm expecting the B770 to outperform the 9060 XT and go after something more like the 5070, 9070 XT, or perhaps even the 5080. But when it comes the market it competes in might have interested me if I hadn't of just upgraded.

On an unrelated topic, the width of my GPU memory interfaces over time is interesting. GPU memory bus sizes are shrinking over time. Except Vega, which came out between RX 400/500 and RX 5000. Vega 56 had a 2048-bit bus. "Only" 409.6 GB/s of bandwidth though.

GPU	Memory Bus	Memory Bandwidth***
RX 480 4GB	256 bit	224 GB/s*
RX 5600 XT (6GB)	192 bit	288 GB/s
RX 9060 XT 16 GB	128 bit	322.3 GB/s**

*The 8 GB RX 480 had faster memory.
**The 9060 has 32MB of L3 cache so it does a lot better than its memory bandwidth would suggest.
***Not shown is memory clock speed is increasing which is what allows the bandwidth to increase even as the bus shrinks.

 

[RandallFlagg]

That is a bummer. There was like 1 day where I saw the 7000-series options and thought it might be time to buy but I passed it up and then supply disappeared. Multiple generations have passed me like this which is why I rushed out to get the 8600 9060 XT. I was coming from a 5600 XT so this is a worthy upgrade and a better fit inside my expansion slot and power constraints than the 7700 XT was, and cheaper too.

To be clear, I'm expecting the B770 to outperform the 9060 XT and go after something more like the 5070, 9070 XT, or perhaps even the 5080. But when it comes the market it competes in might have interested me if I hadn't of just upgraded.

On an unrelated topic, the width of my GPU memory interfaces over time is interesting. GPU memory bus sizes are shrinking over time. Except Vega, which came out between RX 400/500 and RX 5000. Vega 56 had a 2048-bit bus. "Only" 409.6 GB/s of bandwidth though.

GPU	Memory Bus	Memory Bandwidth***
RX 480 4GB	256 bit	224 GB/s*
RX 5600 XT (6GB)	192 bit	288 GB/s
RX 9060 XT 16 GB	128 bit	322.3 GB/s**

*The 8 GB RX 480 had faster memory.
**The 9060 has 32MB of L3 cache so it does a lot better than its memory bandwidth would suggest.
***Not shown is memory clock speed is increasing which is what allows the bandwidth to increase even as the bus shrinks.

Memory bus width within a tier has gone up and down over time. XX60 series for example have ranged from 128 to 256 bit.

I think it has a lot to do with where we are at in advancing memory standards. This time around, Nvidia does not need as wide a bus as AMD due to using GDDR7. Last gen they also had GDDR6X. I'm sure there are a few engineers whose job it is to analyze the benefits of using faster (more expensive) memory on a narrower bus vs the extra cost of implementing a wider bus (more PCB space, more traces, more cost).

You can also get away with 12GB instead of 16GB with 192bit vs 128 / 256 bit. While 8GB usefulness seems to have finally come to an end for those running > 1080P, I haven't seen anything indicating 12GB is hobbled vs 16GB. And given how long it took for 8GB to show it limits, 12GB may be good for 5+ more years.

 

[jallenlabs]

Keep on keeping on Intel. We need those gpus.

 

[lexluthermiester]

I still really hope they accelerate the Celestial launch. I am hoping they are doing this to fast ship the remaining silicon. I am excited for this and Intel to keep pushing despite resistance in the GPU industry because we all need it, even those that wont ever buy one; but there release schedule is not great.

Perhaps they'll do a dual release? B750/B770 and the C750/C770? That would not be a terrible thing.

 

[soulphie]

imo with how late this card would be if it it its clear that this card was not planed. Intel arc themselves had said that they would not make the same mistake again of releasing a lower end card before their strongest (like they did with the a380) because it hampers hype. Tom peterson had said on GN that he would release the A770 first if he had the chance to redo alchemist.
I am guessing that intel saw that arc built up good will with the low price of the b580 and want to actually bring out something that can make money as opposed to the b580 that certainly loses intel money on MSRP

 

[capdauntless]

I'll wait for reviews but it's highly likely I'll be picking one of these up.

 

[kondamin]

Perhaps they'll do a dual release? B750/B770 and the C750/C770? That would not be a terrible thing.

If they had it in own production with spare capacity sure, but they don’t

 

[Bomby569]

Intel has been rather late with B770. Ideally, it should have arrived shortly after the B580. It has been six months since that launched and we still don't have a confirmed release date for its larger sibling.

considering how bad the price/availability of the already released cards is, it's pointless to release more cards.

 

[lexluthermiester]

If they had it in own production with spare capacity sure, but they don’t

You don't know what their capacity is. Let's let Intel worry about Intel things.

 

[kondamin]

You don't know what their capacity is. Let's let Intel worry about Intel things.

for the B series it's 0 as they have TSMC make it

 

[ModEl4]

That's strange, i thought that since B580 is already CPU limited if you pair it with older CPUs they wouldn't attempt a B770.
Did they had a breakthrough with their software team?
Anyway it must mean good news then, unless they are willing to sell G31 (bigger than Navi48 357mm²) at a loss.
Imo Navi44 is so good that it doesn't make sense if G31 can't hit the below performance level and with the current drivers is questionable if it can hit it.(32 Xe2)
And even if it can hit the performance it must be lowered priced vs Navi44 to make impression.
Below performance levels in current TPU testbed:

		1080	1440	2160
5060Ti 16GB	$429	111.5	112.5	118.5
9060XT 16GB	$349	105	105.5	109.5
9060XT 8GB	$299	100	100	100
5060 8GB	$299	96.5	94.5	83.5
B770 16GB	$329-$349	106	112.5	130
B750 16GB	$299-$319	96.5	101.2	115.3
B580 12GB	$249	78.5	81.5	91.5

 

[lexluthermiester]

for the B series it's 0 as they have TSMC make it

Once again, YOU don't know what their capacity is. Let's let Intel do the Intel thing.

 

[wolf]

Come on Intel, we're counting on you for saving us from Nvidia, and AMD Nvidia minus $50!