Wednesday, August 12th 2020
NUVIA Phoenix SoC is 40-50 Percent Faster Than Zen 2 for a Third of Power
Last year, in November of 2019, a startup company called NUVIA Inc. broke out of the stealth mode and decided to reveal itself to the public. Focused on "re-imagining silicon", the company is led by some of the brightest minds in the semiconductor industry. Some people like Gerard Williams III, the CEO of the company, previously served as a chief CPU architect at Apple and has spent over 10 years at Arm before that. Others like Manu Gulati and John Bruno serve as senior vice presidents of silicon and system engineering respectively. Together, their people are forming a company full of well-known industry names. Of course, there are more and you should check out this page.
NUVIA Inc. promises to deliver only the best performance and "re-imagine silicon" as they say. Today, we got some bold claims from the company regarding the performance of their upcoming Phoenix SoC. Using Geekbench 5, the company has provided some simulated results of how the Phoenix SoC will perform. Being that it runs on Arm ISA, the SoC can run at very low power and achieve good performance. NUVIA has run some simulations and it expects its Phoenix SoC to be 40-50% faster in single-threaded performance than Zen 2/Sunny Cove at just a third of the power, 33% of the percent of power to be precise. In the graph below, NUVIA has placed its SoC only in 5 W range, however, the company said that they have left the upper curve to be disclosed at later date, meaning that the SoC will likely compete in high-performance markets and at higher power targets. While these claims are to be taken with a grain of salt, it is now a waiting game to see how NUVIA realizes its plans.
Source:
AnandTech
NUVIA Inc. promises to deliver only the best performance and "re-imagine silicon" as they say. Today, we got some bold claims from the company regarding the performance of their upcoming Phoenix SoC. Using Geekbench 5, the company has provided some simulated results of how the Phoenix SoC will perform. Being that it runs on Arm ISA, the SoC can run at very low power and achieve good performance. NUVIA has run some simulations and it expects its Phoenix SoC to be 40-50% faster in single-threaded performance than Zen 2/Sunny Cove at just a third of the power, 33% of the percent of power to be precise. In the graph below, NUVIA has placed its SoC only in 5 W range, however, the company said that they have left the upper curve to be disclosed at later date, meaning that the SoC will likely compete in high-performance markets and at higher power targets. While these claims are to be taken with a grain of salt, it is now a waiting game to see how NUVIA realizes its plans.
28 Comments on NUVIA Phoenix SoC is 40-50 Percent Faster Than Zen 2 for a Third of Power
Well, I guess I just described startups in general.
NUVIA Phoenix Targets +40-50% ST Performance Over Zen 2 for Only 33% the Power
Because in 2-3 year, they will fight with very different beast. That remind me all these news GPU startup in the 2000 that were suppose to have amazing tech beating current tech. Their main problem was always that by the time they were ready to go to silicon, they were outdated.
Also, Geekbench. LOL. Way to kill any credibility you might have had, NUVIA.
There have already been a few vendors for the last 3-4 years entering with ARM based solution like Cavium, Ampere, Amazon and others. The market is huge so many want to get a piece of it and it will be race during the upcoming 10 years to grab market shares and trying to become dominant player like Intel are at the moment.
While X86 isn't dead, it is a dying tech as it doesn't seem to be able to scale up performance much and can't scale down it's energy consumption much either.
Intel havn't manage to increase the IPC since Skylake 5 years ago, during this time ARM based CPU's have more than doubled the IPC with no higher energy requirements and there seem to be no slowdown on development for the upcoming years.Can't say it is a post-epedemic as Nuvia was founded before any heard of Covid-19. Also founded by former employees at Apple that been involved in the development of their chips. Apple doesn't seem to care for the datacenter market and these people seeing the potential of the chips they developed and wanted to capture a piece of the huge data center market.
Of course all of these vendors will not succeed, there will be a race and some will not make and either go under or be bought up but calling it an investor scamming scheme are a bit to much.
you can simulate all you want but until it's actual hardware being tested your results don't mean shit if anything yall just sound like another Transmeta wanna be
Fact: Skylake IPC has increased and continues to do so; Ice Lake increases it by 18%. That's completely ignoring Zen and its derivatives.
Fact: Arm has been capped at ~3Ghz clocks for years despite using much smaller process nodes.
Fact: x86 can be low-power at high performance, Lakefield demonstrates this.
ARM and other companies using their designs/ISA look like they have better efficiency for a very simple reason that has nothing to do with how well they make their chips, they just use lower clocks. Lower clocks means lower voltages but the relationship between power and voltages is not linear so when you have an ARM chip running at lower clocks vs your average 5 Ghz Intel chip or whatever it's efficiency looks much, much better.For the last time, no they don't. I've explained above why.
Also this is a long game. I'm not saying these guys or especially Tachyum will live up to the promises, but them trying to innovate is not a bad thing.He literally finished with "take this wit a grain of salt".Intel stagnation. They have been around for a few years.
Well, maybe not 5 years, but it was almost 5 years before Ice Lake got here after Sky Lake and on desktop and server it still are Sky Lake with very small addition. Zen played catchup to Sky Lake until Zen 3, not saying AMD did bad with Zen but what did they the 8-9 years before that, Phenom and Bulldozer wasn't that good.
Yes, ARM CPU's are designed for lower clocks, but look around on the X86 servers CPU's the have all core turbo boost around 3GHz and using high amount of power/core (5-15W).
The Samsung Galaxy Book S with Lakefield was not that impressive, it wasn't much faster than it's Snapdragon 8cx predecessor. I wait to see more implementation of Sunny Cove cores before getting a opinion.Aarch64 was designed with modern compilers in mind and can utilize different optimizations and consistency models than possible on X86, so far less transistors are needed to branch prediction and micro-ops decode and less transistors requires less power.
Are almost no data center CPU's running in 5GHz and anything with over 20 cores runs around 3GHz with all core turbo.
An other indicator ARM gaining in the data center market are the Top500 list where X86 have had a massive part for the last 15 years with some odd Sparc and Power based system now and then, now suddenly the 1st posistion are a ARM based system.It isn't dead, much like Sparc, Power, MIPS are still around but it's market share will most certain slowly decrease over the coming 10 years if not something big happens. Software in active development can move to other architectures often with very little effort, but moving some older C/C++ code from X86 to ARM can be problematic with Aarch64 different consistency model but C11/C++11 are a almost 10 year old standard now so much software should already use it.
Alot of software are developed on somewhat portable frameworks like Java, .Net, Go or using interpeters as Node.Js, Python or similar so they are often fast to move.
Then there are of course some software that isn't on active development and they are harder, there you will need X86 but if the software is that old, it was developed for slower systems and maybe emulation is enough to run it anyway.
All of this are for data center and the same are not true for Desktop or high performance mobile space, I don't see any competition for a X86 Windows turning up in at least 5 years, on the Apple side I'm uncertain, will there be a PowerMAC with ARM in the next 2-3 years? not sure but it will interesting to see what the first MacBook with ARM can do for certain.
x86-64 is extension of x86, so you're being pedantic for no reason.
You think Intel/AMD stuff their CPUs with so much complicated OoO , branch prediction, fetch-decode logic for no reason ? You need all that to provide sustainable lasting high performance platforms.Funny you mention those Power and Sparc systems because ARM is basically in the same spot they were, yet they're somehow better positioned according to you. You are aware those guys tried the same things right ? Streamlined architectures, reliance on compiler optimization. Look were they ended up.
Maybe if you build a system for an exact specific task all that works, but if you want a platform that can be used generically, it just doesn't. Many have tried.You are not impressed by that because all ARM and Apple and everyone else did these last 10 years is mimic x86 designs to get more performance and catch up while Intel and AMD reared their head in hard limits of what can be done. These ARM chips will hit the same wall very soon, ILP has a limit, there is no magic recipe.
Which led to Intel charging ridiculous prices for their CPUs, especially server CPUs, e.g. www.anandtech.com/show/15955/how-to-save-6000-dollars-intel-xeon-8280-vs-6258r - that's the primary driver behind the adoption of Arm in datacentres. Now that Intel has competition again, the faltering push towards Arm is likely to slow back down again.
The other driver, of course, is lock-in. Sooner or later we're going to see different vendors offering different extensions to their Arm CPUs that will make certain software faster on those CPUs, and they won't share those extensions with others. So if you want your particular workload to run faster, you gotta use that vendor's CPUs, which means their cloud. And once they have you in there, much like Apple they're going to make sure you don't get out.
No. Fiction. There's nothing inherent about ARM's ISA or core designs that limits them to lower clocks. There's just no point going way past the most efficient clock / voltage on the curve, unless you have to brute force everything because IPC is so bad .... this has been the case for the better part of a decade and a half with x86. AMD have finally made some pretty efficient ULV CPUs with good IPC and clocks which aren't too bad. Intel have yet to prove they're capable of this - the AMD ones are still nowhere near as efficient as the best ARM chips though. People are likely in for a bit of a shock when the ARM based Apple laptops come out ... as there will finally be an easy way for 'noobs' to benchmark cutting edge ARM chips against Intel chips aimed at the same class (which admittedly look REALLY bad vs the AMD equivalents currently). Intel laptop chips have about 50% of the performance per watt of AMD chips, and I'd be surprised if the AMD chips aren't at about 50% of what Apple come out with - unless their first gen product is low-balling.
Lakefield's a pile of crap ...
I simulate its 1 billion times worse. Prove me wrong