News Intel GPUs - Intel launches A580

Page 33 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

blckgrffn

Diamond Member
May 1, 2003
9,110
3,028
136
www.teamjuchems.com
Intel insists that DG1 is a native die and not defective U parts. It is still likely the U design cut down to remove everything other than the memory controller and the IGP.

BTW when it comes to F processors, Dell is putting in at the very least the 1650 Super so I guess that's not that bad.

That's definitely a more useful combination. Switching from my RX 460 desktop GPU to my 8th gen i7 Laptop for even normal desktop work on three screens reveals how slow that onboard GPU is - I even have a quadro but during typical desktop use it hangs back.
 

Tup3x

Senior member
Dec 31, 2016
944
925
136
That makes it sound like a prototype which is the last thing an OEM wants installed.

My opinion is that despite Intel's denials, this is just a way to eat up dies with a dead cpu but a working iGPU. And they put the bare minimum into its development to get them working on platforms that already knew how to hook into iGPU so they can sell it with F series. When Intel basically was the entire market they didn't need to do this, they just threw away imperfect dies so the OEMs had an iGPU whether they wanted it or not. But now the OEMs have been buying F series and plopping cheapo discrete cards in there which Intel probably hates.
They don't have to include display connectors in motherboard if they slap in a card like this. It's a glorified display adaptor. If some OEM wants to target multi display use case, then it might do the job just fine. In any case this is not their actual entry in the discrete desktop GPU market.
 

PingSpike

Lifer
Feb 25, 2004
21,729
559
126
Well, with what the existing cards cost at the moment I can't think of a much better time to enter the market for Intel.

That said, isn't the higher end card using TSMC as well? Not sure that is going to play out well at the moment.
 

MrTeal

Diamond Member
Dec 7, 2003
3,554
1,658
136
Well, with what the existing cards cost at the moment I can't think of a much better time to enter the market for Intel.

That said, isn't the higher end card using TSMC as well? Not sure that is going to play out well at the moment.
Does it mine Eth well? No? That's too bad. Does it mine it at all? A little bit? Great, sold out!
 

Heartbreaker

Diamond Member
Apr 3, 2006
4,222
5,224
136
Well, with what the existing cards cost at the moment I can't think of a much better time to enter the market for Intel.

That said, isn't the higher end card using TSMC as well? Not sure that is going to play out well at the moment.

They don't have a higher end gaming card. It would be a great time, if they actually had a gaming product that wasn't total garbage, but they don't.

HPC part is completely disconnected from gaming. The low end part only has IGP performance, and is only available for a few crappy OEM systems.

So they have nothing...
 
Feb 4, 2009
34,494
15,729
136
I don’t get all the negativity, I want a 3rd player in graphics. I’d be happy if it was incapable of mining because there would be a chance you could find one and it would be priced reasonably.
 
Feb 4, 2009
34,494
15,729
136
They don't have a higher end gaming card. It would be a great time, if they actually had a gaming product that wasn't total garbage, but they don't.

HPC part is completely disconnected from gaming. The low end part only has IGP performance, and is only available for a few crappy OEM systems.

So they have nothing...

Then there is this.....
 

NTMBK

Lifer
Nov 14, 2011
10,208
4,940
136
They don't have a higher end gaming card. It would be a great time, if they actually had a gaming product that wasn't total garbage, but they don't.

HPC part is completely disconnected from gaming. The low end part only has IGP performance, and is only available for a few crappy OEM systems.

So they have nothing...

They have HPG, their high end gaming part made on TSMC 7nm, which is different from their HPC part.
 

gdansk

Golden Member
Feb 8, 2011
1,973
2,353
136
Xe HPC is the chiplet monstrosity Raja showed off earlier. Made by bunch of different fabs then glued together. The minimum performance for this is a known quantity, since they advertise the specifications of the Aurora supercomputer.

Xe HP is pretty similar but scaling up to 4 tiles instead of 2. This is the one they demonstrated the total flops and encoding ability in a 4 tile configuration. So performance should be around the advertised level.

Xe HPG is hopefully much simpler and made by TSMC. In Intel's own slides it targets "enthusiast, mid range". I think they'll hit the midrange but enthusiast is dubious. We've not really seen anything about it other than Raja's latest tweet.
 
Last edited:

Mopetar

Diamond Member
Jan 31, 2011
7,797
5,899
136
It has dragons. DRAGONS. Be content with your bread and circuses. Raja has no time for such insolence!

Historically speaking, "There be dragons" was usually a warning to stay away from something.

But it is good to get at least something even if it's not very substantial. Intel's GPUs were starting to seem far more discreet than they were discrete.
 
  • Like
Reactions: Tlh97 and coercitiv

Heartbreaker

Diamond Member
Apr 3, 2006
4,222
5,224
136

All depends on the clock speeds, and how well the architecture scales!

Doesn't their IGP have 96 EU? Then I don't buy 512 EU > AMD RX 6800.

Right now 96 EU IGP ~= Vega 8 APU.

512/96*8 ~= Vega 42, (not Navi 60).

I'm expecting it to fall short, very short, of RX 6800.
 

mikk

Diamond Member
May 15, 2012
4,111
2,105
136
Xe LP is bandwidth starved in notebooks, more so than Vega 8. And it is low clocked at 1.3 Ghz in comparison, AMDs iGPU or dedicated GPUs in general can go to like 2.0 Ghz. Frequency scaling, bandwidth scaling is unknown at this point for DG2 or Gen12HP.
 

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
I agree with above that Xe in tiger lake is still lower clocked than it will be in dGPUs. You're looking at Xe on 10sf with the optimizations of the process set for a CPU vs. Xe on a third party process that's optimized for GPU use. Looking at the information that we have on the discreet Xe LP cards, they clock a little higher with a reduced EU count. They also aren't much different in performance, despite having their memory controller all to themselves.

In the end, there's no point in having so many SKUs if they are very close in performance. If they have that many, then it should indicate that the next rung up is sufficiently more performant enough to justify it, which should indicate that they can hit most of the rings up the ladder.