What is the ISA for the basic execution units? How are they getting sufficient bandwitdh and latency with 2000 units? Knights Mill only had 128 x86 cores. Is this a reduced instruction set?
It's more like a GPU than a CPU -- but I'm sure it uses some sort of proprietary ISA, of which I don't think a name has ever been mentioned publicly, at least not to my knowledge. Probably "The Xe ISA"
When you consider NVidia`s latest earnings show they made more money due to rapid growth in this segment as opposed to previously having the gaming segment being more profitable. Intel are onto something.
Worth noting that Nvidia's growth in that area was largely down to the purchase of Mellanox, otherwise they'd still be earning significantly more from gaming graphics.
Definitely a smart move from Intel, but they'll need to be compelling in an area that Nvidia aren't.
Even if Intel is successful, it will take time to unseat NVIDIA's CUDA frameworks and in the meantime, NVIDIA will be releasing new gear. I'm no fan of Intel (the last few years of shenanigans has been godawful) but it's great to see more competition here.
The secondary question is how much this market is really going to grow. I get the sense that personal assistants and internet startups have plateau'd and the global economy is now a big question mark, with the American market a double question mark.
Now all we need to know are the die/card size, temperature, power requirements, price and apples-to-apples comparisons to nVidia next generation enterprise GPU. What else was I expecting but incomplete reporting of unicorn hardware and the proverbial; "take these numbers with a grain of salt" NOT that these are pre-production numbers so that we'll assume to infinity and beyond for the final hardware.
Oh and for anandtech to stop being a 247 shill for intel.
Yup. What a yawn-fest. If it was available NOW that would be something to discuss. But this is nothing that AMD & NVIDIA cannot equal in the Ampere & RDNA 2 generations which will *ship* in the next few months (1T & 2T solutions, 4T equivalent would be multi-GPU). By the time Intel fixes their process or migrates the IP to TSMC both NVIDIA & AMD will be on next gen, chiplet architectures, likely 5nm, with performance that will beat this with ease. As someone else said - Intel is shrieking loudly about unicorn hardware running at undisclosed power levels with unverified benchmarks because they have NOTHING they can ship or even sample that is competitive.
What a sad state of affairs for a company with some great engineering talent.
Also sad that AT is broadcasting this propaganda with little to no comment pointing out the gaping holes in Intel's presentations & roadmaps.
"As someone else said - Intel is shrieking loudly about unicorn hardware running at undisclosed power levels with unverified benchmarks because they have NOTHING they can ship or even sample that is competitive." That's precisely what Intel have been doing for many years now, first for their 10nm fabbed CPUs and now for their GPUs.
RDNA 2? I wouldn't count on that one too much cause as long as i can remember AMD wanted to bring THE BIG GPU which they predictably failed! Ampere? well it is no currently at Xe league but next gen NVIDIA GPU might be who knows! Also a small note: intel 10nm is actually on par with TSMC 7nm and intel's 7nm would be on par with TSMC 5nm so the only thing AMD/NVIDIA or TSMC actually would have is smaller naming scheme nothing else!
@MojArch - so many citations needed. Intel's 10nm is *about to be on a par* with TSMC 7nm - it hasn't been up until now, as it's not lived up to Intel's original specifications. 10++ (or 10SF as it's now known) is getting them to that point.
As for RDNA 2 vs this, it's worth bearing in mind that from Tiger Lake projections, Xe gets Intel to a position of needing significantly more area to beat *Vega* in terms of power and performance. Scaling that up, they're either going to pay a die-size penalty or a power penalty to beat RDNA 2.
Projecting from Turing, Ampere is likely to be *way* outside of Xe's League on a pound-for-pound basis.
Ampere is the last monolithic Nvidia GPU. From its successor (Hopper) onward they are also switching to chiplets (or at least that's the current plan; it's possible that the very high transistor density of TSMC's 5nm node allows them to postpone switching to chiplets for one more generation). I doubt Hopper will be released before Q4 2021 - Q1 2022 though.
Xe-LP is on 10nm SuperFin and Xe-HP on 10nm Enhanced SuperFin. Judging from Tiger Lake benchmark leaks, the SuperFin process provided a pretty big performance bump vs Ice Lake. Intel says it also can now extend to a higher voltage and frequency range. I expect to see the extra range used in Tiger Lake-H in 2021.
Aside from yields, which they noticeably didn't mention, it looks like Intel's 10nm process is doing well.
This is news. We're reporting news that came out of the Architecture Day event, just as the same way we report news out of AMD's Tech Days when they give insights into future products. These products aren't finalised, but understanding where Intel is pitching its stake in the sand, at least at a holistic level, is better than not knowing.
Normally I'm called a shill for AMD, but what do I know.
Some people will always get upset not just at the level of obfuscation in a company's marketing - which is sometimes fair - but at the sites that relay it, which is just bizarre. How else are we supposed to have a laugh at the desperation if we don't get to see it? 😬
I've never created an account since I never felt the need to post a comment. I created one just now because I just wanted to say Ian that for every Internet Troll posting that you are "a shill" there are a hundred of us quiet readers who enjoy your news coverage, the quality, depth and insight. Keep doing it and don't feel the need to feed the trolls :)
I don't in the least agree that Anandtech are a "247 shill for intel". They're definitely passing the figures on with a degree of enthusiasm intact, but at no point did I feel like this was different from how they treat any other company's products.
What is clear is that Intel are engaging in a broad-spectrum marketing onslaught with a bunch of un-launched products right now. They're allowed to do that, and you're allowed to add whatever dose of seasoning you find appropriate.
Since these are early engineering samples running at 1300 MHz we can assume that gaming graphics cards will contain either a single high clocked tile or two lower clocked tiles. If Intel can achieve a peak/turbo clock of 1900 - 1950 MHz at a reasonable TDP with their single tile variant (for a peak FP32 performance of 15.5 to 16 TFLOPs) they might not need a dual tile gaming variant to compete with AMD and Nvidia*, at least for the first generation of cards.
*except the cutting edge xx80 Ti / xx90 card from Nvidia.
I'm expecting Xe HPG to take up roughly the role in the GPU market that RDNA played for this generation - not remarkable in any respect, but performant enough at the right price. I'm sincerely hopeful that they'll be at least moderately competitive just to end this absurd price inflation that Nvidia have been pushing in the mid-range.
Yeah, except no... :-D nVidia and AMD are preparing next gen. Plus Teraflops are meaningless for gaming, because it really doesn't scale and you can't really compare it between different architectures. Plus AMD and nVidia have decades of gaming optimizations behind them, I ntel doesn't
Bang for buck is the easiest area to compete, because if you're desperate enough for market share then you can just sell at whatever price your performance enables. The performance per area / performance per watt side of the equation are the hard parts, as they dictate your profit margins and your performance ceiling.
To summarise: PP$ is the most likely place for Intel to actually be competitive. Based on the projections for Xe from Tiger Lake, they're almost certainly not going to be particularly competitive on any other front.
Tiger Lake will have support for lpddr5, which could give the GPU a further bump.
Also, the recent slides indicate the new Xe-DG1 10nm SuperFin process enables clocking up to 1.7 GHz, which is higher than I've seen leaked in early benchmarks.
Both of those "further bumps" are already baked into their performance projections. The first batch of devices won't have LPDDR5, which means they'll be running at a deficit compared with Intel's projections.
Basically they're playing fast-and-loose with their marketing and selling Jam Tomorrow, as per usual.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
41 Comments
Back to Article
TopHatProductions115 - Friday, August 21, 2020 - link
But can it run Crysis?!I'll see myself out now :P
Ian Cutress - Friday, August 21, 2020 - link
We have CPUs that can run Crysis through software rendering now. Not well, but they do run :)https://www.anandtech.com/bench/CPU-2020/2755
The_Stopher - Friday, August 21, 2020 - link
Can it run Crysis remastered? ;)kaspar737 - Saturday, August 22, 2020 - link
The better question now is "Can it run Microsoft Flight Simulator?"Machinus - Friday, August 21, 2020 - link
What is the ISA for the basic execution units? How are they getting sufficient bandwitdh and latency with 2000 units? Knights Mill only had 128 x86 cores. Is this a reduced instruction set?extide - Friday, August 21, 2020 - link
It's more like a GPU than a CPU -- but I'm sure it uses some sort of proprietary ISA, of which I don't think a name has ever been mentioned publicly, at least not to my knowledge. Probably "The Xe ISA"Spunjji - Monday, August 24, 2020 - link
It's almost certainly reduced when compared with x86, as the EUs don't need to do nearly as many different kinds of work.KimGitz - Friday, August 21, 2020 - link
When you consider NVidia`s latest earnings show they made more money due to rapid growth in this segment as opposed to previously having the gaming segment being more profitable. Intel are onto something.Spunjji - Monday, August 24, 2020 - link
Worth noting that Nvidia's growth in that area was largely down to the purchase of Mellanox, otherwise they'd still be earning significantly more from gaming graphics.Definitely a smart move from Intel, but they'll need to be compelling in an area that Nvidia aren't.
nico_mach - Wednesday, August 26, 2020 - link
Even if Intel is successful, it will take time to unseat NVIDIA's CUDA frameworks and in the meantime, NVIDIA will be releasing new gear. I'm no fan of Intel (the last few years of shenanigans has been godawful) but it's great to see more competition here.The secondary question is how much this market is really going to grow. I get the sense that personal assistants and internet startups have plateau'd and the global economy is now a big question mark, with the American market a double question mark.
Everett F Sargent - Friday, August 21, 2020 - link
Four Tile: 42277 GFLOPs (42.3 TF) of FP32 (3.993x)https://developer.nvidia.com/blog/nvidia-ampere-ar...
So ~2X of the nVidia A100.
Now all we need to know are the die/card size, temperature, power requirements, price and apples-to-apples comparisons to nVidia next generation enterprise GPU. What else was I expecting but incomplete reporting of unicorn hardware and the proverbial; "take these numbers with a grain of salt" NOT that these are pre-production numbers so that we'll assume to infinity and beyond for the final hardware.
Oh and for anandtech to stop being a 247 shill for intel.
shabby - Friday, August 21, 2020 - link
By the time the 10nm process is mature Nvidia will be on the next architecture.TimSyd - Friday, August 21, 2020 - link
Yup. What a yawn-fest.If it was available NOW that would be something to discuss. But this is nothing that AMD & NVIDIA cannot equal in the Ampere & RDNA 2 generations which will *ship* in the next few months (1T & 2T solutions, 4T equivalent would be multi-GPU).
By the time Intel fixes their process or migrates the IP to TSMC both NVIDIA & AMD will be on next gen, chiplet architectures, likely 5nm, with performance that will beat this with ease.
As someone else said - Intel is shrieking loudly about unicorn hardware running at undisclosed power levels with unverified benchmarks because they have NOTHING they can ship or even sample that is competitive.
What a sad state of affairs for a company with some great engineering talent.
Also sad that AT is broadcasting this propaganda with little to no comment pointing out the gaping holes in Intel's presentations & roadmaps.
Santoval - Friday, August 21, 2020 - link
"As someone else said - Intel is shrieking loudly about unicorn hardware running at undisclosed power levels with unverified benchmarks because they have NOTHING they can ship or even sample that is competitive."That's precisely what Intel have been doing for many years now, first for their 10nm fabbed CPUs and now for their GPUs.
MojArch - Friday, August 21, 2020 - link
RDNA 2? I wouldn't count on that one too much cause as long as i can remember AMD wanted to bring THE BIG GPU which they predictably failed!Ampere? well it is no currently at Xe league but next gen NVIDIA GPU might be who knows!
Also a small note: intel 10nm is actually on par with TSMC 7nm and intel's 7nm would be on par with TSMC 5nm so the only thing AMD/NVIDIA or TSMC actually would have is smaller naming scheme nothing else!
Spunjji - Monday, August 24, 2020 - link
@MojArch - so many citations needed. Intel's 10nm is *about to be on a par* with TSMC 7nm - it hasn't been up until now, as it's not lived up to Intel's original specifications. 10++ (or 10SF as it's now known) is getting them to that point.As for RDNA 2 vs this, it's worth bearing in mind that from Tiger Lake projections, Xe gets Intel to a position of needing significantly more area to beat *Vega* in terms of power and performance. Scaling that up, they're either going to pay a die-size penalty or a power penalty to beat RDNA 2.
Projecting from Turing, Ampere is likely to be *way* outside of Xe's League on a pound-for-pound basis.
Santoval - Friday, August 21, 2020 - link
Ampere is the last monolithic Nvidia GPU. From its successor (Hopper) onward they are also switching to chiplets (or at least that's the current plan; it's possible that the very high transistor density of TSMC's 5nm node allows them to postpone switching to chiplets for one more generation). I doubt Hopper will be released before Q4 2021 - Q1 2022 though.Spunjji - Monday, August 24, 2020 - link
I think you're probably right about that release date. Nvidia seem to be quite firmly sticking to a 2-year cadence with mid-cycle refresh now.JayNor - Monday, August 24, 2020 - link
Xe-LP is on 10nm SuperFin and Xe-HP on 10nm Enhanced SuperFin. Judging from Tiger Lake benchmark leaks, the SuperFin process provided a pretty big performance bump vs Ice Lake. Intel says it also can now extend to a higher voltage and frequency range. I expect to see the extra range used in Tiger Lake-H in 2021.Aside from yields, which they noticeably didn't mention, it looks like Intel's 10nm process is doing well.
Ian Cutress - Friday, August 21, 2020 - link
This is news. We're reporting news that came out of the Architecture Day event, just as the same way we report news out of AMD's Tech Days when they give insights into future products. These products aren't finalised, but understanding where Intel is pitching its stake in the sand, at least at a holistic level, is better than not knowing.Normally I'm called a shill for AMD, but what do I know.
Santoval - Friday, August 21, 2020 - link
Perhaps you are a shill for both, they are not necessarily mutually exclusive :p/s
Slash3 - Friday, August 21, 2020 - link
I appreciate the coverage, regardless of which company a given event may feature.Spunjji - Monday, August 24, 2020 - link
Some people will always get upset not just at the level of obfuscation in a company's marketing - which is sometimes fair - but at the sites that relay it, which is just bizarre. How else are we supposed to have a laugh at the desperation if we don't get to see it? 😬Solidstate20 - Wednesday, August 26, 2020 - link
I've never created an account since I never felt the need to post a comment. I created one just now because I just wanted to say Ian that for every Internet Troll posting that you are "a shill" there are a hundred of us quiet readers who enjoy your news coverage, the quality, depth and insight. Keep doing it and don't feel the need to feed the trolls :)HardwareHut - Friday, August 21, 2020 - link
Those numbers (42.3 TF of FP3) of the Xe-HP is with matrix math instructions.The A100 peaks at 156 TF of FP32 with tensor cores, and at 312 TF of FP32 with sparsity.
The 19.5 TF of FP32 that you are citing for the A100 is of FMA only - no tensor cores (matrix math).
Ian Cutress - Saturday, August 22, 2020 - link
These Intel numbers are through the ALUs, not XMX.p1esk - Saturday, August 22, 2020 - link
Tensor cores are FP16 multipliers with FP32 accumulators. Apples to oranges.CiccioB - Tuesday, August 25, 2020 - link
With Ampere the tensor cores are now FP32, not FP16.And can also do FP64 math, at a reduced speed.
Spunjji - Monday, August 24, 2020 - link
I don't in the least agree that Anandtech are a "247 shill for intel". They're definitely passing the figures on with a degree of enthusiasm intact, but at no point did I feel like this was different from how they treat any other company's products.What is clear is that Intel are engaging in a broad-spectrum marketing onslaught with a bunch of un-launched products right now. They're allowed to do that, and you're allowed to add whatever dose of seasoning you find appropriate.
JayNor - Monday, August 24, 2020 - link
Raja said the Xe-HP products go up to 300W range. However, that was back in Dec, 2019.https://youtu.be/-kWiRrf2o6Q?t=2156
MonkeyPaw - Friday, August 21, 2020 - link
I wonder how this compares to the Afterburner Accelerator card? Apple claims that card can handle 23 4K streams at once.SirDragonClaw - Saturday, August 22, 2020 - link
Those things are not comparable, also the Accelerator card is just a cheap fpga.Santoval - Friday, August 21, 2020 - link
Since these are early engineering samples running at 1300 MHz we can assume that gaming graphics cards will contain either a single high clocked tile or two lower clocked tiles. If Intel can achieve a peak/turbo clock of 1900 - 1950 MHz at a reasonable TDP with their single tile variant (for a peak FP32 performance of 15.5 to 16 TFLOPs) they might not need a dual tile gaming variant to compete with AMD and Nvidia*, at least for the first generation of cards.*except the cutting edge xx80 Ti / xx90 card from Nvidia.
Ryan Smith - Friday, August 21, 2020 - link
Xe-HPG will be a monolithic chip. Intel has already confirmed that it won't use any advanced packaging techniques (e.g. EMIB).Spunjji - Monday, August 24, 2020 - link
I'm expecting Xe HPG to take up roughly the role in the GPU market that RDNA played for this generation - not remarkable in any respect, but performant enough at the right price. I'm sincerely hopeful that they'll be at least moderately competitive just to end this absurd price inflation that Nvidia have been pushing in the mid-range.SethNW - Saturday, August 22, 2020 - link
Yeah, except no... :-D nVidia and AMD are preparing next gen. Plus Teraflops are meaningless for gaming, because it really doesn't scale and you can't really compare it between different architectures. Plus AMD and nVidia have decades of gaming optimizations behind them, I ntel doesn'tSethNW - Saturday, August 22, 2020 - link
I hit submit by mistake, don't see edit option, sorry for multi post.Anyway, I just don't see them coming on to for gaming. Or coming on top for bang for the buck. Maybe work wise, they will do better though.
Spunjji - Monday, August 24, 2020 - link
Bang for buck is the easiest area to compete, because if you're desperate enough for market share then you can just sell at whatever price your performance enables. The performance per area / performance per watt side of the equation are the hard parts, as they dictate your profit margins and your performance ceiling.To summarise: PP$ is the most likely place for Intel to actually be competitive. Based on the projections for Xe from Tiger Lake, they're almost certainly not going to be particularly competitive on any other front.
JayNor - Monday, August 24, 2020 - link
Tiger Lake will have support for lpddr5, which could give the GPU a further bump.Also, the recent slides indicate the new Xe-DG1 10nm SuperFin process enables clocking up to 1.7 GHz, which is higher than I've seen leaked in early benchmarks.
Spunjji - Friday, August 28, 2020 - link
Both of those "further bumps" are already baked into their performance projections. The first batch of devices won't have LPDDR5, which means they'll be running at a deficit compared with Intel's projections.Basically they're playing fast-and-loose with their marketing and selling Jam Tomorrow, as per usual.