I was right.View attachment 107291
Clock speeds Geekbench6
Multi CoreX1P-46-100 8 Phoenix cores @ 3.4 GHz
(4.0 GHz Single Core Boost)12,000 8 Gen 4 2 Phoenix-L cores @ 4.3 GHz
6 Phoenix-M cores @ 3.5 GHz10,000
X1P-46-100 was run in Windows. If it was Linux, the score would be ~10% higher, so about 13,000 multi-core.
That means X1P-46-100 is 30% faster in multi-core than 8 Gen 4.
Both are 8 core CPUs.
If the 8G4 scores less, despite it's Phoenix-M cores being clocked even higher than the Phoenix cores in the X Plus,
It must mean that the Phoenix-M cores have less IPC than Phoenix/Phoenix-L cores.
Phoenix-M is a different uarch.
Lots of vendors have already been down this road before and they failed miserably in creating a competitive Arm-based SoC using just the architectural license. Qualcomm's lead isn't going to change anyone's mind and magically make it possible again. The type of engineering resources required to make a competitive SoC in today's market are astronomical.So ARM is scrambling, flexing their legal muscles to potentially kill Oryon but also to discourage others from trying to follow Qualcomm's lead.
I think the non-constructive tone of ARM's negotiating tactics either speak to its desperation / lack of pricing power or to outside influences, one of which could very well be Apple: it has pursued legal action against Nuvia from its very inception after GW III's departure.What if its not ARM that wants to take down Qualcomm, but... Apple?
Remember, Apple is part owner of ARM, and has incredible influence over ARM.
Arm memory ordering model is different from x86. That's why Apple added an extension to help Rosetta2. Yeah Arm is different from x86 beyond decoders 😉I don't think ARM's case is very strong though: ARM's IP reach doesn't extend past the decoders of a CPU and the rest of the implementation is entirely Nuvia's efforts. Jim Keller had once said that AMD's cancelled K12 ARM CPU based on the Zen u-arch wouldn't involve much beyond replacing the CPU decoders (although the similar memory ordering, registers / flags, and calling convention between x86 and ARM certainly helps in creating common u-arch implementations.)
Total Store Order (TSO) ?Arm memory ordering model is different from x86. That's why Apple added an extension to help Rosetta2.
To be honest, ARM is also acting in bad faith.
It seems to me that they won't just be satisfied with extracting more $$$ from Qualcomm. Instead, they have a vested interest in destroying the Oryon CPU project.
With ARM’s Knowledge And Assistance Owed To Qualcomm Under Its ALA, Qualcomm Continued Its Work Developing CPU Cores After The Acquisition Closed
211. ... During the verification process, ARM knew that it was interacting with former NUVIA employees, and knew that Qualcomm was seeking to verify core designs that included technologies Qualcomm had acquired from NUVIA.
216. In or around late 2021, Qualcomm also introduced the Compute SoC into the parties’ weekly discussions. Like the parties’ discussions concerning the in-development Phoenix Core for the Server SoC, these discussions were transparent, and ARM was aware that these discussions included Qualcomm engineers formerly at NUVIA and related to Qualcomm’s ongoing development of the technologies it had acquired from NUVIA.
After so many years, Qualcomm finally brought Apple's technology to Android. This was once ridiculed and questioned, but the actual result was that the second-generation Oryon, which took three years to polish, accomplished the feat and the improvement was obvious. However, building from scratch requires trial and error, and the first generation had many problems. The second generation was a perfect answer in three years. But it is still ridiculed, which is really puzzling.
ARM is a company that focuses on IP development, with a large team. It loses to a small workshop in the same development cycle. If more people were given to Nuvia, it would have entered the iteration last year, and XE would not have been launched so late. Everything is difficult at the beginning. The next generation of SM8850 will also have a more eye-catching improvement, but the frequency will be higher. The mixed use of SF2 and N3P is worrying. There may be a relatively low-frequency and low-priced sub-flagship, and a high-specification, high-frequency and high-performance real flagship, and the price is also expensive.
The performance of Oryon on spec2017 can be tested on several more models, and they should be roughly the same. The large lead of GB6 is still due to the high-specification and large-scale CPU cluster. Looking only at the single-core spec, the lead is not large, but looking at the scale of the single core, next year will be a watershed. What can be revealed now is that both ST and MT have improved by more than 20%. With the introduction of SME and frequency increase, we can continue to appreciate the stack core strategy.
If I have to comment on the second generation Oryon Phoenix L and Phoenix M, it is probably reasonable but below expectations. It is rare for a small workshop to achieve this level. I should have said it exceeded expectations. But I can only give a reasonable evaluation. The CPU is too difficult, I can understand.
The original expectation was that the second generation single core performance would catch up with a17pro and energy efficiency would catch up with a16. At present, both are lower than my expectation. The next generation will face challenges in energy efficiency because there is no process node update. However, N3p does not need to worry too much about overclocking capability. It depends on whether the spec can be improved and continue to exceed expectations.
Appreciate it while it lasts. We fear the forum will shut down someday, and it could be soon.To be honest, I did not expected that this forum has so much for offering to newbies like me,I am loving every second I am spending here.Truly a gem community.
If your implementation does both the weaker data dependency ordering of ARM and the TSO of x86, it more or less does come down to decoders when it comes to getting correct implementations of either ISA. Those concepts themselves aren't unique to ARM's IP; I guess I meant to say that implementations can be made compatible here without the hassle of totally redesigning caches for instance.Arm memory ordering model is different from x86. That's why Apple added an extension to help Rosetta2. Yeah Arm is different from x86 beyond decoders 😉
I think Oryon-M is fine.I find the M core a bit disappointing, in terms of efficiency. Seems like P cores are much better for efficiency, I understand the main reason is for area savings but still.
That's true for D9400 and A18 Pro too.Even at the lower part of the curve, like at 1W, the P core is still more efficient than the M core.
What if its not ARM that wants to take down Qualcomm, but... Apple?
Remember, Apple is part owner of ARM, and has incredible influence over ARM.
Shhh. No no no.Appreciate it while it lasts. We fear the forum will shut down someday, and it could be soon.
That Apple has no financial or legal clout in the industry (even with rivals much less companies it has small stakes in) is laughable. During their attempted knifing of Qualcomm, they were able to force phone manufacturers to withhold contractually guaranteed licensing fees (the complaints by the OEMs used language nearly identical to Apple's complaint.) They were able to sway Samsung to their side:Apple was one of ARM's founders and a 1/3 owner at first, but they sold off those shares long ago. They went in on ARM's IPO along with other companies like Nvidia, but at a mostly symbolic amount in the $25 to $100 million range (in an IPO that valued ARM at over $50 billion) A shareholder that holds between 0.05% and 0.2% of your shares is not exercising any management control, especially in a matter as important as "do we sue one of our biggest customers". Apple may have influence over them in technical decisions like the future of the ISA, but they don't have any influence financially or legally.
This is all SoftBank's call, as they still own 90% of ARM. Given that its market cap has gone from ~$50 billion at IPO to $160 billion at the moment it needs to grab a LOT of money from somewhere to justify that ridiculous valuation. I mean, the valuation was ridiculuos at IPO, but now it is in the category of bizarre IMHO (P/E of 346)
The stock is down 7% today as of this writing so apparently the market is not a fan of their ultimatum to Qualcomm.
At the conference in Idaho, according to documents Qualcomm filed earlier this year, Apple saw an opportunity to put itself in front of investigators. Qualcomm claims that at the event—almost certainly the Allen & Co. conference in Sun Valley, which both Cook and Lee attended—the Apple executive urged Samsung to pressure South Korean antitrust regulators to intensify an investigation into Qualcomm that had been open since 2014. “Get aggressive,” the Apple executive said, according to Qualcomm’s filing, adding that this would be the “best chance” to get Qualcomm to lower its prices.
Got it and it makes sense. I wonder how much one would leave on the table by having TSO always on; perhaps there's some way to test the on Apple or Qualcomm machines?If your implementation does both the weaker data dependency ordering of ARM and the TSO of x86, it more or less does come down to decoders when it comes to getting correct implementations of either ISA. Those concepts themselves aren't unique to ARM's IP; I guess I meant to say that implementations can be made compatible here without the hassle of totally redesigning caches for instance.
Qualcomm is certainly rumored to want Intel (and most likely an x86 license...):Maybe Intel should give Qualcomm a call: "Yo, want x86 license?"
Not going to happen though... But if they did, that would be very interesting turnaround.
Aren't cross licensing agreements between Intel and AMD valid as long as one or the other won't be bought by 3rd party? I wonder what the consequences would be in such case.Maybe Intel should give Qualcomm a call: "Yo, want x86 license?"
Not going to happen though... But if they did, that would be very interesting turnaround.
Under the terms of the agreement between the x86 chip developers, if the two companies merge with other companies on certain terms, enter into certain kind of joint-venture agreements that effectively change their ownership or get acquired, the cross-licensing agreement is terminated. On Wednesday, it was reported that the agreement is terminated only for the party that changes its control. However, according to Drew Prairie, director of corporate communications at AMD, once ownership of either AMD or Intel changes, the whole agreement is terminated for both parties. As a result, after a transaction happens, the companies need to negotiate a new cross-licensing agreement.
“Actually, the agreement is pretty clear – if there is a change of control for either company the agreement is terminated,” said Mr. Prairie. “That does not mean a new agreement could not be reached, but in a change of control the agreement is terminated.”
The world's been getting crazier of late, it seems.Yeah Qualcomm isn’t getting an x86 license unless something very very crazy happens