• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Intel Updates Q1 2022 (GPU, CPU and Foundry)

Grim5

Grim5

Grim5

Soldato
Joined
6 Feb 2019
Posts
20,093
Intel has provided updated roadmaps and details for upcoming products today, including GPUs and CPUs

Starting with GPUs:

* Intel ARC is slightly delayed, with desktop GPUs pushed back to a Q2 launch. Intel plans to manufacture 4 million desktop GPUs this year.

This 4 million figure is unlikely to inspire confidence, with Nvidia moving 10s of millions of Ampere GPU's, its a number from Intel that may not be able to provide significant downward pressure on GPU prices



FLzyhcuVcA0WEPn



https://cdn.wccftech.com/wp-content/uploads/2022/02/2022-02-17_22-20-53-1030x579.png











* Project Endgame: Intel will be launching a new service that allows gamers to access the power of an Intel Arc GPU. I don't know how well this will work, you play games on your PC, but all the graphics processing is done remotely in an Intel server housing an ARC GPU that you pay a fee to access.


* Intel Celestial architecture: Dominating the ultra high end market. Intel's next architecture is Battlemage, thats launching next year, but the year after in 2024 Intel is launching Celestial which it claims will take the ultra high end gaming performance crown.


- Intel is taking a slow stepped approach here - Alchemist is designed to compete with Nvidia and AMD 700 series GPUs (e.g 6700xt/RTX3070). Battlemage is designed to compete with 800 series GPUs (7800XT/RTX4080) and Celestial designed to beat 900 series GPU's (8900XT/RTX5090)




https://cdn.videocardz.com/1/2022/02/Intel-Roadmap.png










Foundry Updates
  • - Intel 4: Ready for mass production in 2022, delivers 20% increase in transistor density and performance per watt.
  • - Intel 3: Ready for mass production in 2023, delivers further 18% improvement in density and perf/watt.
  • - Intel 20a: Targeting 2024, delivers further 15% improvement in density and perf/watt

https://cdn.wccftech.com/wp-content/uploads/2022/02/2022-02-17_22-25-41-1030x579.png







CPU: In 2025 Intel will launch a new CPU architecture based on it's tiled MCM technology, it aims to be able to fit 5 chiplet/tiles on one die. These can be configured in a variety of ways, you can 5 chiplets of CPU cores, 2 CPU and 3 GPU or any other combination of CPU and GPU chiplets.


https://cdn.wccftech.com/wp-content/uploads/2022/02/2022-02-17_22-26-01-1480x833.png






Performance:

* Intel claims its Saphire Rapid SP chips are up to 150% faster than the just released AMD Milan-X, only shown one benchmark though (simulation of fluids)

https://cdn.wccftech.com/wp-content...U-vs-AMD-EPYC-Milan-X-Benchmarks-1030x579.png






* Intel claims its tiled Ponte Vecchio GPU is up to 260% faster than Nvidia in computing Black Scholes and Monte Carlo financial systems modelling


https://cdn.wccftech.com/wp-content...ecchio-GPU-vs-NVIDIA-Ampere-A100-1030x579.png

**Do Not Hotlink Images **
 
Last edited by a moderator:
Interesting times ahead. Making some bold claims there about the high end market. I can’t see either AMD or Nvidia giving up easily so hopefully this all bodes well for competition…
 
rare said:
Interesting times ahead. Making some bold claims there about the high end market. I can’t see either AMD or Nvidia giving up easily so hopefully this all bodes well for competition…
Click to expand...

depending on how much intel is going to charge for these cards.

but hopefully it will put the boot up the backside of nvidia and amd and push them to go that little bit further with generational improvements.
 
Honestly I'm just glad there's a strong third player in the market now. More options means our money goes further in theory, as we can pick from a wider range of products for the same budget. Assuming Intel price their cards aggressively.
 
RTX 3070 Beats Intel Flagship Arc Alchemist GPU In OpenCL

https://www.tomshardware.com/uk/news/nvidia-rtx-3070-beats-intel-arc-alchemist-gpu-opencl-benchmark

The flagship Intel Arc Alchemist graphics card hits 2.4 GHz in a new OpenCL benchmark. Nonetheless, the DG2 graphics card was still no match for Nvidia's GeForce RTX 3070 (Ampere), one of the best graphics cards on the market.

Barring any setbacks, Intel's DG2 family of discrete desktop graphics cards will arrive in the second quarter of the year. The flagship Arc Alchemist SKU will wield up to 512 execution units (EUs) and up to 16GB of GDDR6 memory. The graphics card has scaled up to 2.4 GHz in the new Geekbench 5 submission (via Benchleaks), likely the graphics card's boost clock. But, of course, it's an engineering sample, so the clock speed isn't final.

If we look at the performance on paper, the flagship Arc Alchemist should deliver around 20 TFLOPs of FP32 performance. Of course, it's not the best metric for gaming, but it's the only information available. Based on the FP32 performance, the DG2-512 theoretically performs similar to the often referenced GeForce RTX 3070. The AMD equivalent would be Radeon RX 6700 XT (Big Navi).

tJ6Hlgf.jpg


Intel Arc Alchemist (Image credit: Primate Labs Inc.)

The Arc Alchemist scored 85,448 points in the OpenCL test, which isn't the best indicator of gaming performance, of course. Nevertheless, the level of performance would put the flagship Arc Alchemist in the same alley as the GeForce RTX 2070 as far as OpenCL workloads go. If we compared it to Nvidia's latest offerings, Intel's graphics card is around 40% slower than the GeForce RTX 3070. Being engineering samples and unreleased hardware, we should take the performance figures with a truckload of salt.

similar DG2-512 sample, which maxed out at 2.1 GHz, had surpassed Nvidia's GeForce RTX 3070 Ti in some workloads. It would seem that Intel has been making progress in optimizing the Arc Alchemist's clock speeds. The early rumors pointed to a 2.4 GHz to 2.5 GHz maximum boost clock for the flagship DG2 part.

The only benchmark that's remotely close to a gaming workload came in the shape of an Ashes of the Singularity submission back in December. However, a mysterious Arc Alchemist sample had put up numbers that were very close to a GeForce RTX 3070, which lends some credence to the rumor that the DG2-512 SKU could potentially offer gaming performance that's close to Nvidia's mid-range performers.
 
Purgatory said:
RTX 3070 Beats Intel Flagship Arc Alchemist GPU In OpenCL

https://www.tomshardware.com/uk/news/nvidia-rtx-3070-beats-intel-arc-alchemist-gpu-opencl-benchmark

The flagship Intel Arc Alchemist graphics card hits 2.4 GHz in a new OpenCL benchmark. Nonetheless, the DG2 graphics card was still no match for Nvidia's GeForce RTX 3070 (Ampere), one of the best graphics cards on the market.

Barring any setbacks, Intel's DG2 family of discrete desktop graphics cards will arrive in the second quarter of the year. The flagship Arc Alchemist SKU will wield up to 512 execution units (EUs) and up to 16GB of GDDR6 memory. The graphics card has scaled up to 2.4 GHz in the new Geekbench 5 submission (via Benchleaks), likely the graphics card's boost clock. But, of course, it's an engineering sample, so the clock speed isn't final.

If we look at the performance on paper, the flagship Arc Alchemist should deliver around 20 TFLOPs of FP32 performance. Of course, it's not the best metric for gaming, but it's the only information available. Based on the FP32 performance, the DG2-512 theoretically performs similar to the often referenced GeForce RTX 3070. The AMD equivalent would be Radeon RX 6700 XT (Big Navi).

tJ6Hlgf.jpg


Intel Arc Alchemist (Image credit: Primate Labs Inc.)

The Arc Alchemist scored 85,448 points in the OpenCL test, which isn't the best indicator of gaming performance, of course. Nevertheless, the level of performance would put the flagship Arc Alchemist in the same alley as the GeForce RTX 2070 as far as OpenCL workloads go. If we compared it to Nvidia's latest offerings, Intel's graphics card is around 40% slower than the GeForce RTX 3070. Being engineering samples and unreleased hardware, we should take the performance figures with a truckload of salt.

similar DG2-512 sample, which maxed out at 2.1 GHz, had surpassed Nvidia's GeForce RTX 3070 Ti in some workloads. It would seem that Intel has been making progress in optimizing the Arc Alchemist's clock speeds. The early rumors pointed to a 2.4 GHz to 2.5 GHz maximum boost clock for the flagship DG2 part.

The only benchmark that's remotely close to a gaming workload came in the shape of an Ashes of the Singularity submission back in December. However, a mysterious Arc Alchemist sample had put up numbers that were very close to a GeForce RTX 3070, which lends some credence to the rumor that the DG2-512 SKU could potentially offer gaming performance that's close to Nvidia's mid-range performers.
Click to expand...


Saw that earlier, completely disregarded it. Don't care about open CL and it's pre release drivers
 
Quartz said:
If it's good enough for 1080p gaming at a decent price then that's all that's needed.
Click to expand...

And being available for gamers to buy. It's going to be mainly about price and availability. It won't have much impact even if it is good enough gaming performance at a decent price if too few gamers can buy them.
 
Alchemist is a weird one, Intel appear to be well behind schedule on the release but have £££ to invest in getting it finished, I hear they're throwing the cheque book at it but who knows..
 
Quartz said:
If it's good enough for 1080p gaming at a decent price then that's all that's needed.
Click to expand...
Aside from Intel maybe wanting to come in a bit below the others, if their design in bad in regards perf/transistors then their scope for being much cheaper may be very limited. Especially if using TSMC's expensive and scares nodes.

Yes, "there's no such thing as bad product, just a bad price" holds, but if for example DG2-512 performs below 3070 or 6700 (which both co-incidentally use 17.x billion transistors) but uses 30 billion transistors to do so (GA102 is 28.3 billion, Navi 21 is 26.8 billion) then we could consider it a failure.

Not that I can see any recent rumours of die sizes. DG2-128 was supposed have (TPU's GPU database guess) 8 billion. Obviously four times - 32 billion - would be bad for even 3070/6700 performance, but while GPUs are massively parallel some parts of a design don't just double/quadrupole.

TPU's DG2-512 placeholder says 18 billion which would be very good.

So despite Intel's latest slides, it is still all in the air even a few months before launch?
 
You must log in or register to reply here.
Back
Top Bottom