Info TOP 20 of the World's Most Powerful CPU Cores - IPC/PPC comparison

[Richie Rich]

Added cores:

• A53 - little core used in some low-end smartphones in 8-core config (Snapdragon 450)
• A55 - used as little core in every modern Android SoC
• A72 - "high" end Cortex core used in Snapdragon 625 or Raspberry Pi 4
• A73 - "high" end Cortex core
• A75 - "high" end Cortex core
• Bulldozer - infamous AMD core

Geekbench 5.1 PPC chart 6/23/2020:

Pos​	Man​	CPU​	Core​	Year​	ISA​	GB5 Score​	GHz​	PPC (score/GHz)​	Relative to 9900K​	Relative to Zen3​
1​	Nuvia​	(Est.)​	Phoenix (Est.)​	2021​	ARMv9.0​	2001​	3.00​	667.00​	241.0%​	194.1%​
2​	Apple​	A15 (est.)​	(Est.)​	2021​	ARMv9.0​	1925​	3.00​	641.70​	231.8%​	186.8%​
3​	Apple​	A14 (est.)​	Firestorm​	2020​	ARMv8.6​	1562​	2.80​	558.00​	201.6%​	162.4%​
4​	Apple​	A13​	Lightning​	2019​	ARMv8.4​	1332​	2.65​	502.64​	181.6%​	146.3%​
5​	Apple​	A12​	Vortex​	2018​	ARMv8.3​	1116​	2.53​	441.11​	159.4%​	128.4%​
6​	ARM Cortex​	V1 (est.)​	Zeus​	2020​	ARMv8.6​	1287​	3.00​	428.87​	154.9%​	124.8%​
7​	ARM Cortex​	N2 (est.)​	Perseus​	2021​	ARMv9.0​	1201​	3.00​	400.28​	144.6%​	116.5%​
8​	Apple​	A11​	Monsoon​	2017​	ARMv8.2​	933​	2.39​	390.38​	141.0%​	113.6%​
9​	Intel​	(Est.)​	Golden Cove (Est.)​	2021​	x86-64​	1780​	4.60​	386.98​	139.8%​	112.6%​
10​	ARM Cortex​	X1​	Hera​	2020​	ARMv8.2​	1115​	3.00​	371.69​	134.3%​	108.2%​
11​	AMD​	5900X (Est.)​	Zen 3 (Est.)​	2020​	x86-64​	1683​	4.90​	343.57​	124.1%​	100.0%​
12​	Apple​	A10​	Hurricane​	2016​	ARMv8.1​	770​	2.34​	329.06​	118.9%​	95.8%​
13​	Intel​	1065G7​	Icelake​	2019​	x86-64​	1252​	3.90​	321.03​	116.0%​	93.4%​
14​	ARM Cortex​	A78​	Hercules​	2020​	ARMv8.2​	918​	3.00​	305.93​	110.5%​	89.0%​
15​	Apple​	A9​	Twister​	2015​	ARMv8.0​	564​	1.85​	304.86​	110.1%​	88.7%​
16​	AMD​	3950X​	Zen 2​	2019​	x86-64​	1317​	4.60​	286.30​	103.4%​	83.3%​
17​	ARM Cortex​	A77​	Deimos​	2019​	ARMv8.2​	812​	2.84​	285.92​	103.3%​	83.2%​
18​	Intel​	9900K​	Coffee LakeR​	2018​	x86-64​	1384​	5.00​	276.80​	100.0%​	80.6%​
19​	Intel​	10900K​	Comet Lake​	2020​	x86-64​	1465​	5.30​	276.42​	99.9%​	80.5%​
20​	Intel​	6700K​	Skylake​	2015​	x86-64​	1032​	4.00​	258.00​	93.2%​	75.1%​
21​	ARM Cortex​	A76​	Enyo​	2018​	ARMv8.2​	720​	2.84​	253.52​	91.6%​	73.8%​
22​	Intel​	4770K​	Haswell​	2013​	x86-64​	966​	3.90​	247.69​	89.5%​	72.1%​
23​	AMD​	1800X​	Zen 1​	2017​	x86-64​	935​	3.90​	239.74​	86.6%​	69.8%​
24​	Apple​	A13​	Thunder​	2019​	ARMv8.4​	400​	1.73​	231.25​	83.5%​	67.3%​
25​	Apple​	A8​	Typhoon​	2014​	ARMv8.0​	323​	1.40​	230.71​	83.4%​	67.2%​
26​	Intel​	3770K​	Ivy Bridge​	2012​	x86-64​	764​	3.50​	218.29​	78.9%​	63.5%​
27​	Apple​	A7​	Cyclone​	2013​	ARMv8.0​	270​	1.30​	207.69​	75.0%​	60.5%​
28​	Intel​	2700K​	Sandy Bridge​	2011​	x86-64​	723​	3.50​	206.57​	74.6%​	60.1%​
29​	ARM Cortex​	A75​	Prometheus​	2017​	ARMv8.2​	505​	2.80​	180.36​	65.2%​	52.5%​
30​	ARM Cortex​	A73​	Artemis​	2016​	ARMv8.0​	380​	2.45​	155.10​	56.0%​	45.1%​
31​	ARM Cortex​	A72​	Maya​	2015​	ARMv8.0​	259​	1.80​	143.89​	52.0%​	41.9%​
32​	Intel​	E6600​	Core2​	2006​	x86-64​	338​	2.40​	140.83​	50.9%​	41.0%​
33​	AMD​	FX-8350​	BD​	2011​	x86-64​	566​	4.20​	134.76​	48.7%​	39.2%​
34​	AMD​	Phenom 965 BE​	K10.5​	2006​	x86-64​	496​	3.70​	134.05​	48.4%​	39.0%​
35​	ARM Cortex​	A57 (est.)​	Atlas​	0​	ARMv8.0​	222​	1.80​	123.33​	44.6%​	35.9%​
36​	ARM Cortex​	A15 (est.)​	Eagle​	0​	ARMv7 32-bit​	188​	1.80​	104.65​	37.8%​	30.5%​
37​	AMD​	Athlon 64 X2 3800+​	K8​	2005​	x86-64​	207​	2.00​	103.50​	37.4%​	30.1%​
38​	ARM Cortex​	A17 (est.)​	​	0​	ARMv7 32-bit​	182​	1.80​	100.91​	36.5%​	29.4%​
39​	ARM Cortex​	A55​	Ananke​	2017​	ARMv8.2​	155​	1.60​	96.88​	35.0%​	28.2%​
40​	ARM Cortex​	A53​	Apollo​	2012​	ARMv8.0​	148​	1.80​	82.22​	29.7%​	23.9%​
41​	Intel​	Pentium D​	P4​	2005​	x86-64​	228​	3.40​	67.06​	24.2%​	19.5%​
42​	ARM Cortex​	A7 (est.)​	Kingfisher​	0​	ARMv7 32-bit​	101​	1.80​	56.06​	20.3%​	16.3%​

TOP 10 - Performance Per Area comparison at ISO-clock (PPA/GHz)

Copied from locked thread. They try to avoid people to see this comparison how x86 is so bad.[/B]

Pos​	Man​	CPU​	Core​	Core Area mm2​	Year​	ISA​	SPEC PPA/Ghz​	Relative​
1​	ARM Cortex​	A78​	Hercules​	1.33​	2020​	ARMv8​	9.41​	100.0%​
2​	ARM Cortex​	A77​	Deimos​	1.40​	2019​	ARMv8​	8.36​	88.8%​
3​	ARM Cortex​	A76​	Enyo​	1.20​	2018​	ARMv8​	7.82​	83.1%​
4​	ARM Cortex​	X1​	Hera​	2.11​	2020​	ARMv8​	7.24​	76.9%​
5​	Apple​	A12​	Vortex​	4.03​	2018​	ARMv8​	4.44​	47.2%​
6​	Apple​	A13​	Lightning​	4.53​	2019​	ARMv8​	4.40​	46.7%​
7​	AMD​	3950X​	Zen 2​	3.60​	2019​	x86-64​	3.02​	32.1%​

It's impressive how fast are evolving the generic Cortex cores:

• A72 (2015) which can be found in most SBC has 1/3 of IPC of new Cortex X1 - They trippled IPC in just 5 years.
• A73 and A75 (2017) which is inside majority of Android smart phones today has 1/2 IPC of new Cortex X1 - They doubled IPC in 3 years.

Comparison how x86 vs. Cortex cores:

• A75 (2017) compared to Zen1 (2017) is loosing massive -34% PPC to x86. As expected.
• A77 (2019) compared to Zen2 (2018) closed the gap and is equal in PPC. Surprising. Cortex cores caught x86 cores.
• X1 (2020) is another +30% IPC over A77. Zen3 need to bring 30% IPC jump to stay on par with X1.

Comparison to Apple cores:

• AMD's Zen2 core is slower than Apple's A9 from 2015.... so AMD is 4 years behind Apple
• Intel's Sunny Cove core in Ice Lake is slower than Apple's A10 from 2016... so Intel is 3 years behind Apple
• Cortex A77 core is slower than Apple's A9 from 2015.... but
• New Cortex X1 core is slower than Apple's A11 from 2017 so ARM LLC is 3 years behind Apple and getting closer

GeekBench5.1 comparison from 6/22/2020:

• added Cortex X1 and A78 performance projections from Andrei here
• 2020 awaiting new Apple A14 Firestorm core and Zen3 core

Updated:



EDIT:
Please note to stop endless discussion about PPC frequency scaling: To have fair and clean comparison I will use only the top (high clocked) version from each core as representation for top performance.

 

[Hitman928]

You have no real idea of TDP of Gravitons2.
Under load the 7742 is stuck at 2.25, less than Graviton2.
You have already been proven wrong on the area/size.
You have already been proven wrong on price.

And while you admit the IPC is less, I even doubt that metric.
And the 128 and 256 core chips. Do you have any benchmarks at all to support this ? or that those chips even exist ?

Why do you keep posting this BS ? If you don't know what you are talking about, you should stop posting.

Ampere announced that they will be coming out with a 128 core chip but it won't release until some time next year, probably first half of next year but I haven't seen any hard date. The 256 core CPU on 5 nm is speculation on his part as far as I know.

 

[lobz]

Nice graph. But you have to take into account this:

• Price ............. Rome 7500 USD ........................ Gravion2 estimated 500 USD .............. way cheaper

256-core A78@ 3.0 GHz on 5nm TSMC process.... that's gonna be a fun

Let me be clear first, you haven't set the bar too high so far.

But this...

This has to be the most ridiculous thing I've ever read on this forum - and I've been lurking for years during the juanrga era.

 

[Markfw]

Let me be clear first, you haven't set the bar too high so far.

But this...

This has to be the most ridiculous thing I've ever read on this forum - and I've been lurking for years during the juanrga era.

Exactly. Here is a retail 7702p 64 core for $3850. And nobody knows what a Graviton2 sells for since they are not for sale. But that estimation is ridiculous. a 9900k 8 core sells for more than that.

AMD Epyc 7702P Server Processor (3.35 GHz, 64 Cores, Socket SP3) Tray - 100-000000047 for sale online | eBay

Find many great new & used options and get the best deals for AMD Epyc 7702P Server Processor (3.35 GHz, 64 Cores, Socket SP3) Tray - 100-000000047 at the best online prices at eBay! Free shipping for many products!

www.ebay.com

 

[Gideon]

You have no real idea of TDP of Gravitons2.
Under load the 7742 is stuck at 2.25, less than Graviton2.
You have already been proven wrong on the area/size.
You have already been proven wrong on price.

And while you admit the IPC is less, I even doubt that metric.
And the 128 and 256 core chips. Do you have any benchmarks at all to support this ? or that those chips even exist ?

Why do you keep posting this BS ? If you don't know what you are talking about, you should stop posting.

Besidse all this I'll add that the 128 core Altra is coming in 2021 and 5nm version is coming in 2022 (going against Genoa, that will also add cores and is 5nm) so a really stupid comparison.

 

[Richie Rich]

Last year:

• Last year everybody was laughing to ARM server attempts like Graviton1 based on weak A72.
• Last year everybody was laughing to Apple's A13, saying it's fake benches and there is no way Apple will move to ARM.

This year:

• This year ARM's 64-core server CPU Graviton2 based on A76 smashed AMD Zen1 Naples and outperformed Rome in performance per thread.
• This year ARM released Cortex X1 core with +30% IPC jump (outperforming Zen2 by 40% IPC). Ignored but nobody laughs anymore.
• This year Apple announced transition of ENTIRE billion business to ARM ISA which means that performance was not fake.
• This year Japanese Fugaku super computer has became No.1 in TOP500 chart and it's based on ARM CPUs with new SVE vectors 2048-bit capable, effectively smashing GPU based supercomputers (and rising big question mark to entire GPU computation business, also note that Tesla dumped Nvidia GPUs and developed their own chip, so GPU computation might be death end road).
• 
  
• This year we expect new server platform Neoverse N2 (based on A77 or A78/X1).
• This year we expect new Apple A14 and ARM MacBook. Big benchmark fun ahead

Next year:

• Next year new ARMv9 ISA is coming, SVE2 instruction set 2048-bit capable.
• Next year new Matterhorn core line up (3 new cores, Big.Little + X2), attacking Apple's A12/A13 performance, including SVE2 vectors.
• Next year new server platform Neoverse N3 (based on Matterhorn?).

IPC increase:

ARM:

• ARM Cortex A75 -> A76 was 30% IPC jump
• ARM Cortex A76 -> A77 was 20% IPC jump (8% higher than Zen2 according to SPECint)
• ARM Cortex A77 -> X1 was 30% IPC jump (40% higher than Zen2 according to SPECint)

AMD:

• AMD Zen -> Zen+ was 5% IPC jump
• AMD Zen+ -> Zen2 was 15% IPC jump
• AMD Zen2 -> Zen3 will be 17(?)% IPC jump

.

 

[peanutbridges]

As a random forum spectator, the oddest thing about all this is that on various forums across the interwebs, there seems to be the same (type of) person arguing for ARM and Graviton and all that. All of a sudden appearing around May 2020 and overzealously promoting the supposed benefits of said platform without an ounce of humility.

So it goes.

 

[Gideon]

I can feel the fear and panic among x86 fanatic fans

Oh what trolling. I'm actually really hopeful of ARM making a breakthrough in server performance and start taking massive market share (for that to really happen we probably have to wait for Nuvia).

I think few here argue that in a couple of years this will start to manifest itself and in the 5-10 year timeframe x86 might indeed be the dinosaur of old mainframe but it's far from set to stone.

What I really do hate is when trolls come to forums, monopolize and hijack any discussion about ARM. Produce endless threads with tables that are absurdly biased and of questionable accuracy and present themselves as the only all-seeing messiah that allowed to preach about the "goodness of ARM" with every other forum member being a "zealous x86 fanatic".

 

[coercitiv]

What I really do hate is when trolls come to forums, monopolize and hijack any discussion about ARM. Produce endless threads with tables that are absurdly biased and of questionable accuracy and present themselves as the only all-seeing messiah that allowed to preach about the "goodness of ARM" with every other forum member being a "zealous x86 fanatic".

Trolls don't continuously produce endless walls of text which expose their own lack of critical thinking. This is a completely different beast, and will likely get worse before it gets better.

 

[DrMrLordX]

• Last year everybody was laughing to ARM server attempts

Now we're just laughing at you.

 

[NTMBK]

• This year Japanese Fugaku super computer has became No.1 in TOP500 chart and it's based on ARM CPUs with new SVE vectors 2048-bit capable, effectively smashing GPU based supercomputers (and rising big question mark to entire GPU computation business, also note that Tesla dumped Nvidia GPUs and developed their own chip, so GPU computation might be death end road).

Fugaku isn't some massive paradigm shift, it's a continuation of Fujitsu's existing designs- the K computer was also a CPU-only vector monster, it just ran on SPARC instead of ARM.

You also neglected to mention that it is no more energy efficient than Summit, using ~2.8X power for ~2.8X performance... and Summit used a 14nm CPU and a 12nm GPU. I don't think the GPGPU is going anywhere.

 

[Richie Rich]

Fugaku isn't some massive paradigm shift, it's a continuation of Fujitsu's existing designs- the K computer was also a CPU-only vector monster, it just ran on SPARC instead of ARM.

You also neglected to mention that it is no more energy efficient than Summit, using ~2.8X power for ~2.8X performance... and Summit used a 14nm CPU and a 12nm GPU. I don't think the GPGPU is going anywhere.

Fair point but few things to note:

1. - ARM Fujitsu A64FX has 2x 512-bit SVE SIMD units. And SVE/SVE2 is capable of 2048-bit. So it's just the beginning. Next Fujitsu's ARM CPU could be much bigger beast than current one (maybe 1024-bit SIMD width).
2. - GPGPU computing is NOT used in mobile devices which means there is BIG question mark about efficiency. Mobile devices are using specialized NPUs, much faster and efficent solution.
3. Also Tesla and Google abandoned Nvidia's GPU in favor of specialized NPUs which give them one order of magnitude better performance and efficiency. For AI/ML is GPGPU definitely a wrong way.
4. Don't forget that big GPGPU success is probably related to x86 stagnation for whole decade thanks to Intel. While Nvidia pushed performance every year significantly up. So GPGPU was kind of the way of the least resistance. Now when CPU performance was unleashed again thanks to ARM and SVE2 vectors there will be some re-balancing in computing (probably less GPGPU in favor of NPU and CPU). And all ARM vendors has specialized NPUs and x86 world has nothing (HiSilicon integrated NPU into his x86, but Intel and AMD has nothing).

A custom AI chip is an order of magnitude more performant and less power hungry than a GPU-based system to run complex neural networks which are at the core of a machine-learning system.


Google ended up with the same conclusion when it revealed 2 years ago that it had secretly built its own AI chip (Tensor Processing Unit or TPU) for the company's data centers, delivering an order of magnitude better-optimized performance per watt for machine learning.


Similarly to Apple, with its own AI chip

Why Tesla Dropped Nvidia's AI Platform For Self-Driving Cars And Built Its Own

Musk's revelations about Tesla's own AI chip development is a blow to the dominance of Nvidia's autonomous vehicle Drive platform in the automotive industry as it proves to the chipmaker's current 370 customers that 3 years and tens of millions of dollars of investment could bring them there too.

www.forbes.com

 

[NTMBK]

Fair point but few things to note:

1. - ARM Fujitsu A64FX has 2x 512-bit SVE SIMD units. And SVE/SVE2 is capable of 2048-bit. So it's just the beginning. Next Fujitsu's ARM CPU could be much bigger beast than current one (maybe 1024-bit SIMD width).
2. - GPGPU computing is NOT used in mobile devices which means there is BIG question mark about efficiency. Mobile devices are using specialized NPUs, much faster and efficent solution.
3. Also Tesla and Google abandoned Nvidia's GPU in favor of specialized NPUs which give them one order of magnitude better performance and efficiency. For AI/ML is GPGPU definitely a wrong way.
4. Don't forget that big GPGPU success is probably related to x86 stagnation for whole decade thanks to Intel. While Nvidia pushed performance every year significantly up. So GPGPU was kind of the way of the least resistance. Now when CPU performance was unleashed again thanks to ARM and SVE2 vectors there will be some re-balancing in computing (probably less GPGPU in favor of NPU and CPU). And all ARM vendors has specialized NPUs and x86 world has nothing (HiSilicon integrated NPU into his x86, but Intel and AMD has nothing).

Why Tesla Dropped Nvidia's AI Platform For Self-Driving Cars And Built Its Own

Musk's revelations about Tesla's own AI chip development is a blow to the dominance of Nvidia's autonomous vehicle Drive platform in the automotive industry as it proves to the chipmaker's current 370 customers that 3 years and tens of millions of dollars of investment could bring them there too.

www.forbes.com

Yeah, AI is a bit of a special case. I think NVidia has done a good job of riding the wave, and adding enough AI-specific stuff (i.e. tensor cores) to keep relevant. But a lot of AI solutions will be best served by dedicated hardware- especially on the inferencing side, not the training side. (Which are two very different things!)

But for "general purpose" HPC, GPGPU isn't going anywhere. I suspect we're going to see a lot more systems with ARM host CPUs and GPGPU accelerators- for instance NVidia has brought the entire CUDA stack to ARMv8.

 

[Thunder 57]

Let me be clear first, you haven't set the bar too high so far.

But this...

This has to be the most ridiculous thing I've ever read on this forum - and I've been lurking for years during the juanrga era.

juanrga was a member here? I'm sure he had plenty of insightful knowledge and was tact with his predictions.

 

[Saylick]

juanrga was a member here? I'm sure he had plenty of insightful knowledge and was tact with his predictions.

I hope that's sarcasm because if so my sarcasm meter must be off. Oh, he had plenty of insightful knowledge alright... Back in the day, the dude was adamant that Zen could only muster Sandy Bridge levels of IPC if I remember correctly, and we all know how well that prediction went...

 

[Thunder 57]

I hope that's sarcasm because if so my sarcasm meter must be off. Oh, he had plenty of insightful knowledge alright... Back in the day, the dude was adamant that Zen could only muster Sandy Bridge levels of IPC if I remember correctly, and we all know how well that prediction went...

It very much was. I debated with him on ExtremeTech about that but he seems to have disappeared from there too. I tried searching for his handle here but came up with noting. Guess he deleted his account or was just banned.

 

[Systems analyst]

There have been endless arguments over the years in this forum about ARM vs X86.
Cost of a server SoC: In the past, ARM estimated $50 -$200 per SoC. This suggests that Graviton2 could easily have a UPC less than $500.
Furthermore, the use of ARM Physical IP reduces the cost to Amazon of development, and boosts performance and reduces time to market. This is not available to X86, presumably.
Amazon only pay the UPC for the SoCs. They can issue RFPs to sub-contractors to build-to-print blades and cabinets etc, with the design done by Amazon engineering teams, driving down costs.
R&D costs have favourable treatment in various tax jurisdictions, minimising the cost of design to Amazon.
Amazon have increased control of the computing stack, as per Apple, improving overall performance.
Amazon can target use-cases as appropriate.
On the other hand, using an X86 node, they pay the price of the node; having worked in industry, on applications of computer technology, the sales price to a customer was an integer multiple of the UPC. X86 cannot compete with this pricing.
ARM technology has enabled Fugaku to reach the top TOP500 in HPC. It is in production use already.
ARM commented that the Green TOP500 placing of FX64 showed that GPGPUs were not necessary for HPC.
Applications run on the Isambard project in the UK show that the Marvell/Cavium TX2 has better I/O capacity than the Intel chip tested. At increasing scale, the CPU and I/O performance of the TX2 and Intel chip were similar.
Apple have outstanding performance in phones and tablets, using ARM technology.
Mobile phones in general use ARM technology.
You can even buy a single-blade FX64 dual-node system now.
Given that there is a trend to using the cloud, it seems quite likely that X86 will be squeezed over the next few years, and may loose a significant amount of market share.
On the other hand,the ARM compilers may need improvement for gather/scatter in large unstructured codes.

 

[Markfw]

There have been endless arguments over the years in this forum about ARM vs X86.
Cost of a server SoC: In the past, ARM estimated $50 -$200 per SoC. This suggests that Graviton2 could easily have a UPC less than $500.
Furthermore, the use of ARM Physical IP reduces the cost to Amazon of development, and boosts performance and reduces time to market. This is not available to X86, presumably.
Amazon only pay the UPC for the SoCs. They can issue RFPs to sub-contractors to build-to-print blades and cabinets etc, with the design done by Amazon engineering teams, driving down costs.
R&D costs have favourable treatment in various tax jurisdictions, minimising the cost of design to Amazon.
Amazon have increased control of the computing stack, as per Apple, improving overall performance.
Amazon can target use-cases as appropriate.
On the other hand, using an X86 node, they pay the price of the node; having worked in industry, on applications of computer technology, the sales price to a customer was an integer multiple of the UPC. X86 cannot compete with this pricing.
ARM technology has enabled Fugaku to reach the top TOP500 in HPC. It is in production use already.
ARM commented that the Green TOP500 placing of FX64 showed that GPGPUs were not necessary for HPC.
Applications run on the Isambard project in the UK show that the Marvell/Cavium TX2 has better I/O capacity than the Intel chip tested. At increasing scale, the CPU and I/O performance of the TX2 and Intel chip were similar.
Apple have outstanding performance in phones and tablets, using ARM technology.
Mobile phones in general use ARM technology.
You can even buy a single-blade FX64 dual-node system now.
Given that there is a trend to using the cloud, it seems quite likely that X86 will be squeezed over the next few years, and may loose a significant amount of market share.
On the other hand,the ARM compilers may need improvement for gather/scatter in large unstructured codes.

Did you read the post a few posts above that detail exactly why $500 is way too cheap for that CPU ? How to you dispute that data ? Do you have specific objections ?

This post: https://forums.anandtech.com/thread...ores-ipc-ppc-comparison.2580622/post-40215959

 

[lobz]

It very much was. I debated with him on ExtremeTech about that but he seems to have disappeared from there too. I tried searching for his handle here but came up with noting. Guess he deleted his account or was just banned.

He spreads the truth on twitter ever since then. Right after Zen came out, he jumped ship to advocating for ARM.

 

[lobz]

Amazon only pay the UPC for the SoCs. They can issue RFPs to sub-contractors to build-to-print blades and cabinets etc, with the design done by Amazon engineering teams, driving down costs.

That is a horrendous argument in a x86 vs ARM debate, as it applies exclusively to Amazon, the company. Not a great way to system analyze the thread.

 

[Thunder 57]

He spreads the truth on twitter ever since then. Right after Zen came out, he jumped ship to advocating for ARM.

Meh, I don't do twitter. As that great Onion video once said; I have nothing more to add.

 

[name99]

Fugaku isn't some massive paradigm shift, it's a continuation of Fujitsu's existing designs- the K computer was also a CPU-only vector monster, it just ran on SPARC instead of ARM.

You also neglected to mention that it is no more energy efficient than Summit, using ~2.8X power for ~2.8X performance... and Summit used a 14nm CPU and a 12nm GPU. I don't think the GPGPU is going anywhere.

*You* neglected to mention some other interesting points. Like
https://www.top500.org/lists/green500/2019/11/

You don't think it's of some significance that a more traditional and easier programming model manages to match (and slightly exceed) the rather more difficult and specialized GPGPU model?
What does flexible mean? Well let's consider another supercomputer benchmark:
https://graph500.org

As to why some people keep posting about ARM on this forum, that's like asking "Why do you keep caring about the truth?"
The mainstream has been happy to post everything from selective quoting to ignoring facts to outright lies for years now; apparently hoping that some combination of avoiding and misrepresenting reality will change it. That strategy has become ever less viable now (though it's certainly still in play) so people are trying alternatives, which tend to be some combination of
- attack the individuals, regardless of what they say.
- complain that they are posting too much (what? really? there are TWO threads about ARM along with ten thousand threads about x86; and about 2/3 of what's in the ARM threads is ARM hate-skepticism)

 

[name99]

He spreads the truth on twitter ever since then. Right after Zen came out, he jumped ship to advocating for ARM.

Perhaps (hard to understand, I know) he's more interested in reality than in tribalism?
He advocated AMD's technology as the likely shape of the future back, now he's advocating ARM's technology as the likely shape of the future. And he's correct in both cases!

AMD's HW side (chiplets, advanced packaging) probably does define the shape of that dimension of computing going forward, certainly for larger SoCs.

ARM's ISA, for whatever reason (IMHO it's not this tiresome idiotic CISC vs RISC nonsense, it's more about memory ordering, precise semantics of inter-core interaction, ease of decode, ease of device validation/debug, and a willingness every so often to restructure the ISA to drop what doesn't work well so that it stops weighing you down) likewise probably will define a different aspect of computing going forward; obviously mobile and IoT already, but soon a reasonable fraction of desktop and server computing. (Already large by mindshare, expected by some of us to also become large and larger by various finance metrics).

 

[Hitman928]

*You* neglected to mention some other interesting points. Like
https://www.top500.org/lists/green500/2019/11/

Why are you pointing to the Green500 list from 8 months ago? This is the latest one:

June 2020 | TOP500

www.top500.org

You don't think it's of some significance that a more traditional and easier programming model manages to match (and slightly exceed) the rather more difficult and specialized GPGPU model?
What does flexible mean? Well let's consider another supercomputer benchmark:
https://graph500.org

Yes, the Fujitsu supercomputer will be easier to program for and get closer to max FLOP rate, that is definitely an advantage. However, it also depends on what the purpose of the machine is. Just like pretty much everything, it's a trade-off. If you have specialized code that can already make very effective use of GPGPU, then the coming supercomputers based around x86 and GPUs will dominate the Fujitsu supercomputer. If you need more flexibility, then the Fujitsu computer may be the better option, but what is the cross section of programs/use cases that can make effective use of the wide Fujitsu vector paths that can't effectively use a GPGPU code path? I don't know. Clearly there's a market for both solutions, but it seems to me like the vast majority of HPC supercomputers are being built with a GPGPU model, so it seems like that is still the preferred path for most of the people who are using these systems. If I'm wrong, then I'm sure we'll start to see more and more A64FX based systems and/or A64FX like designs popping up as the basis for these giant HPC systems.

The other variable is AMD's and Intel's attempts at a unifying programming space to make it easier to program for and better utilize these GPGPU based systems. We'll see how much success they have with this.

As to why some people keep posting about ARM on this forum, that's like asking "Why do you keep caring about the truth?"
The mainstream has been happy to post everything from selective quoting to ignoring facts to outright lies for years now; apparently hoping that some combination of avoiding and misrepresenting reality will change it. That strategy has become ever less viable now (though it's certainly still in play) so people are trying alternatives, which tend to be some combination of
- attack the individuals, regardless of what they say.
- complain that they are posting too much (what? really? there are TWO threads about ARM along with ten thousand threads about x86; and about 2/3 of what's in the ARM threads is ARM hate-skepticism)

People posting about ARM is fine. Posting a lot about ARM is fine. Coming into threads that have nothing to do with ARM and inserting ARM into the discussion over and over again and creating multiple threads on the same topic is not fine. That's what people are complaining about.

 

[name99]

Why are you pointing to the Green500 list from 8 months ago? This is the latest one:

June 2020 | TOP500

www.top500.org

The date does not change the point I was making. Both Summit and Fugaku get ~the same perf/W BUT Fugaku achieves that with a rather more flexible programming model. You don't think that's an important point?

The newer list you provide shows the same point. The new top machine wins by means of a dedicated matrix accelerator, which is great. But once again substantially limited in the range of tasks it can accelerate.

The point is not that accelerators don't have value. Everyone has always agreed that
- accelerators are good for some tasks
- not for others
- are a pain to code for
- but worth it because they use less power
So it's kinda remarkable when you can basically match them on the power front, without having to give up as much generality and programming ease.

Yes, the Fujitsu supercomputer will be easier to program for and get closer to max FLOP rate, that is definitely an advantage. However, it also depends on what the purpose of the machine is.

That's a reasonable response if your goal is "I need to go out and buy a supercomputer today to perform task X". But that's not what we are doing. What we are doing is trying to understand something of the future direction of technology; and for that purpose it is highly significant that SVE appears to be able to perform this sort of FLOPS-dense arithmetic at reasonable power levels.

People posting about ARM is fine. Posting a lot about ARM is fine. Coming into threads that have nothing to do with ARM and inserting ARM into the discussion over and over again and creating multiple threads on the same topic is not fine. That's what people are complaining about.

Given that Richie Rich started this thread, and started it with an explicit "Here's how ARM is doing well" agenda, it seems a bit strange to complain about people "hijacking it" and inserting ARM into the discussion...
If that's happening in an essentially x86 thread (I mostly don't read those) complain in that thread. But complaining "ARM thread is being used to talk about ARM. OMG!!!" makes very little sense.

 

[Hitman928]

The date does not change the point I was making. Both Summit and Fugaku get ~the same perf/W BUT Fugaku achieves that with a rather more flexible programming model. You don't think that's an important point?

Not as much when Summit is 2 years old now and built around 14/16 nm technology compared to a just deployed system built entirely on 7 nm. When the x86/GPGPU systems come out that are built on 7 nm starting later this year, the Fujitsu system is not going to look so hot, at least from a peak performance and perf/w perspective.

The newer list you provide shows the same point. The new top machine wins by means of a dedicated matrix accelerator, which is great. But once again substantially limited in the range of tasks it can accelerate.

Yes this is true, but did you look at the #2 in the list? That's an almost 40% increase in peak perf/w over the full Fujitsu system when the x86/GPGPU solution moves to 7 nm. There's more of that coming later this year and next.

The point is not that accelerators don't have value. Everyone has always agreed that
- accelerators are good for some tasks
- not for others
- are a pain to code for
- but worth it because they use less power
So it's kinda remarkable when you can basically match them on the power front, without having to give up as much generality and programming ease.

Again, they matched it when comparing the brand new system against the 2 year old system giving the Fujitsu a significant node advantage. That node advantage won't last long and the Nvidia system is just the beginning of what is to come. I agree, if you can get a more general purpose architecture that is easier to program for to match the performance, or at least the perf/w, of a completely specialized or even GPGPU system, then that would be remarkable and basically obsolete the other systems. But that's not what has happened here. You have a first to market system utilizing the advantages of 7 nm that held a moment in the spotlight but is quickly getting eclipsed by more powerful GPGPU/specialized systems as they move to 7 nm as well.

Those GPGPU systems are also (maybe) becoming easier to program for with new unified programming models but we'll have to see if that works out the way AMD/Intel hopes they will or not. If further developed pure ARM HPC systems can catch up, that'd be awesome for the end users and I'm sure they would welcome such a thing with open arms, but that's not what we've seen happening, at least not yet. Maybe a new gen of ARM cores with even further developed SVE and DL extensions will get us there. I have my doubts but am willing to be open to that possible future.

That's a reasonable response if your goal is "I need to go out and buy a supercomputer today to perform task X". But that's not what we are doing. What we are doing is trying to understand something of the future direction of technology; and for that purpose it is highly significant that SVE appears to be able to perform this sort of FLOPS-dense arithmetic at reasonable power levels.

It's one and the same. The future is shaped by who is buying these massive systems.

Given that Richie Rich started this thread, and started it with an explicit "Here's how ARM is doing well" agenda, it seems a bit strange to complain about people "hijacking it" and inserting ARM into the discussion...
If that's happening in an essentially x86 thread (I mostly don't read those) complain in that thread. But complaining "ARM thread is being used to talk about ARM. OMG!!!" makes very little sense.

I don't think you are aware of what happened. This is now a merged thread because Richie Rich kept creating new threads on the same topic so the mods merged the latest two threads and told him to keep the relevant ARM discussions in this thread and stop putting it in multiple threads that he created or spreading it to other threads that aren't even ARM related.