from 100€ to 200€ then, instead. Hope they can keep up with those prices
Intel ARC GPUs, Xe Architecture for dGPUs [2018-2022]
- Thread starter DavidGraham
- Start date
-
- Tags
- intel
- Status
https://ascii.jp/elem/000/004/069/4069704/
TSMC was chosen because Intel advanced nodes wouldn't have enough capacity at the time, next chip could be Intel made but it's not set in stone yet.
XeSS is again confirmed to work with Xe-LP (DP4A version)
AIBs will bring custom cards just like AMD & NVIDIA
TSMC was chosen because Intel advanced nodes wouldn't have enough capacity at the time, next chip could be Intel made but it's not set in stone yet.
XeSS is again confirmed to work with Xe-LP (DP4A version)
AIBs will bring custom cards just like AMD & NVIDIA
D
Deleted member 13524
Guest
The fact that they're having AIBs in the mixture makes me lose some hope on the cards finding their way to gamers.
D
Deleted member 13524
Guest
So the Xe cards will go to miners.
Way to go Intel.
At this point it looks like they're only going to be good for occupying valuable N6 waffers that would otherwise be better spent on relieving the console market.
Way to go Intel.
At this point it looks like they're only going to be good for occupying valuable N6 waffers that would otherwise be better spent on relieving the console market.
It’s probably because they know it won’t work. Nvidia hasn’t been able to figure it out.
Who knows how good or bad they're in mining?
Man from Atlantis
Veteran
D
Deleted member 13524
Guest
Mining efficiency in GPUs has been a pretty straightforward equation of VRAM bandwidth divided by the graphics card's power consumption. Architecture and floating point performance have little to do with hashing performance so far.
The bottleneck is always on the GPUs' VRAM, which is why miners usually downclock and downvolt the vcore while overclocking the memory. Cards with HBM2 are especially good for this, which is why XFX is repurposing Navi 12 leftovers to put dedicated mining cards in the market.
If Intel's Alchemist has a 512GB/s bandwidth in a 6nm chip that can probably clock low and consume little power, and there are no hashing locks in the hardware/software, it will probably be good for mining.
I'm fairly sure those geometry numbers for Navi 22 are wrong. It has 2 shader engines with 2 shader arrays each, and each shader array has a primitive unit that culls 2 triangles per clock. This results in up to 8 culled triangles per clock, just like Navi 10.
Regardless, if DG2-512 with those specs doesn't perform substantially better than Navi 22, it's either clocking really low or Intel has a dud in their hands, in what relates to gaming performance.
Bondrewd
Veteran
keeping the delay -> moar flops cadence going woohooo; power is still horrendous tho and oh they've added more racks for that.
Treat it as an Aurora devkit or something.
jiiiiiiiiiiiiiiiiiiiiiii.
Treat it as an Aurora devkit or something.
Going by this report be the next platform, the lopsided race between AMD and Intel for the exaflops machine seems to be one for third place:
https://www.nextplatform.com/2021/10/26/china-has-already-reached-exascale-on-two-separate-systems/
https://www.nextplatform.com/2021/10/26/china-has-already-reached-exascale-on-two-separate-systems/
Bondrewd
Veteran
Linpack machines; useless.
DavidGraham
Veteran
Intel showcased their XeSS AI upscaling tech running already in two games: Hitman 3 and Riftbreaker.
https://videocardz.com/newz/intel-d...on-technology-in-hitman-3-and-the-riftbreaker
https://videocardz.com/newz/intel-d...on-technology-in-hitman-3-and-the-riftbreaker
Last edited:
If they're anything like the current Sunway system (very likely IMHO), then they are probably not really that good at real world workloads.
While the Linpack performance of Sunway (#4 on June 2021 Top500 list) is similar to Sierra (#3 on the same list), Sierra is 3.7x as fast on HPCG tests. Fugaku (current #1) is 33x as fast as Sunway on HPCG.
IMHO HPCG is more representative of real world workloads then Linpack is. Of course, I don't know what they are doing using Sunway, maybe it's a good fit, but in most cases HPC is not just about packing more ALUs.
For those interested, here's the latest list of HPCG results from June 2021:
https://www.hpcg-benchmark.org/custom/index.html?lid=155&slid=310
Bondrewd
Veteran
Good news; they are.
It's s LINPACK machine lmao.
Very interesting numbers there. From the non A64/NEC-stuff it seems Infiniband is the interconnect of choice for systems with higher HPCG efficiency.
And yet, AMD and Intel are pursuing this apparently meaningless number like no tomorrow.
- Status
