Only a 15% uplift in ST in Cinebench R23 vs a 5950X. Considering Zen 2 to Zen 3 had bigger uplift and AMD waited 2 years to deliver Zen 4.
Keep in mind this is on TSMC 5nm.
Keep in mind this is on TSMC 5nm.
Last edited:
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
AMD 7000 Series - A disappointment!
- Thread starter exoticSpice
- Start date
- Sort by reaction score
It seems that the IPC increase is so low that AMD has decided to mask it and instead decided to reveal the overall performance increase of the new CPUs.
Has AMD reached the limit of the x64 ISA and can't innovate anymore? Could the PC world move to a RISC (ARM or RISC-V) ISA more quickly?
Has AMD reached the limit of the x64 ISA and can't innovate anymore? Could the PC world move to a RISC (ARM or RISC-V) ISA more quickly?
Looks like it.
We have to see how Intel will do with Meteor Lake and Arrow lake first. Then we have Zen 5 as well. But I think it's more of a limit of Zen not x86.
Reading between the lines it appears like the are no significant improvements in the core architecture itself, mostly harnessing the improvements of 5nm to push the clocks higher while maintaining high sustained performance. Yeah. Nothing to write home about. If that’s it, then both Intel and AMD are stuck mostly playing the balancing game.
Why do you think that ARM or RISC-V won't have the same problem?
Let's see what kind of performance Apple will deliver with M2. The curious thing is that top single-core performance is more or less comparable between all the most advanced CPUs, regardless of their architecture or design. Maybe we are simply hitting the limit of CPU technology, as in running out of tricks to make single-threaded tasks go faster with the current manufacturing processes. Apple's advantage is that they can get to this peak using 3-4x less power than Intel or AMD, but it doesn't mean they will have an easy time breaking it.
AMD will probably turn the wick up if the architecture is as good as it gets. No one complains when Intel drives ludicrous amounts of power through the chips.
Who is looking forward to 850W+ rigs?
Who is looking forward to 850W+ rigs?
AMD64 (what both Intel and AMD use) is an archaic ISA that has to be brute forced to run at any speed with modern CPU design. Long pipelines, complex branch prediction and the use of a lot of power is what hide this for the casual user.
All other things being the same a proper RISC ISA will outperform both AMD and Intel by a margin, think about what a M1 could do if Apple were to push it past 5GHz alone.
ARM and RISC-V will eventually run into the same problem but a higher absolute level of performance.
I don’t expect more than 10-20% on the single core for the M2 especially if it is based on the A15.
Perhaps this cycle is AMD's "Tick"? I don't think they would purposefully limit their computational capacity unless they had a reason. AMD has never been one to shy away from increased performance.
Environment aside, the bigger these PSUs get, the more companies want to draw power instead of focusing on making designs efficient. Pretty soon we'll be using Nuclear Reactor cooling towers to cool down computers.
The biggest problem of x86 is non-trivial instructions decoding, but it's a problem that can be solved in constant time. It just costs extra die space (probably even not much) and power. Long pipelines is an implementation detail, as x86 vendors want to go as fast as possible without going too wide with power consumption being an afterthought at best. As to use of complex branch prediction and other complicated performance enhancing shenanigans, there is barely a CPU maker on the market that has more of them than Apple.
What are you basing these claims on? There are no RISC CPU that outperforms x86 CPUs on the per core basis. The "higher absolute level of performance" of RISC is a conjecture at best. I mean, RISC has many benefits, but it is not immediately clear to me why RISC CPUs should offer higher absolute performance. It's something people like to mention (at least since Apple Silicon came out and now RISC-V is gaining traction), but it has never been practically demonstrated or even elaborated. Personally,
As of now, the most obvious demonstrable advantage of ARMv8 is that it can be used to build CPUs with same peak performance but much lower power consumption. It is yet unclear how much of the latter is due to ARM ISA itself and how much due to Apple's secret source or reliance on advanced technologies with no costs spared (designing an ultra-wide CPU like Apple did is no easy or cheap enterprise). I mean, even the newest official ARM cores are still significantly behind what x86 offers, despite using latest and greatest ARM ISA. Intel's Gracemont cores for example seem to have performance comparable to ARM X1 (in a Snapdragon 888) at similar power consumption, despite it being an x86 core - both do around 1100 GB5 at roughly 5 watts.
Why didn't Apple push it past 5Ghz then? Surely if it were possible they would do it, at least on their desktop machines, to have a commanding lead over x86 offerings? Instead they only boosted the clock by meagre 50mhz or so. I think this is where we come to design tradeoffs. Apple goes ultra wide, which (along with their other secret sauce) gives them top-notch performance with top-notch power consumption, but their design likely limits the maximal achievable frequency. It's ways better than the "let's go fast and very vey hot" general approach of x86 makers, but it is not clear at all that Apple's approach can achieve higher per-core throughput.
Na, it's going to be a new core. Using A15 at this point would make no sense at all.
a) it would require design changes that would go against what is best for laptops and they use the same chips for both
b) it would have used much more power and cooling to the point where the Ultra would not have been possible in the Studio (at that size) and Apple does care a lot about these things
c) the extra heat and power would have made it impossible to run all cores that fast all the time negating much of the benefits in multithreaded applications
d) M1.... is still 1st gen quite possible that Apple will push up the clocks later on or with systems that have a biggere thermal envelope (like the next MacPro)
I feel that all what AMD did is the bare minimum work needed to fit 7nm Zen 3 into 5nm socket, nothing more or less (similar to what they did with 6nm Zen for Laptops).
And I speculate that they were waiting for their Xilinx acquisition to finalize to start working on new architecture.
Xilinx provide FPGA chips that are used for co-processors/accelerators for various tasks (similar to M1 Pro/Max media engines for example) which should offload tons of CPU work to efficient specilaized accelerators.
And I speculate that they were waiting for their Xilinx acquisition to finalize to start working on new architecture.
Xilinx provide FPGA chips that are used for co-processors/accelerators for various tasks (similar to M1 Pro/Max media engines for example) which should offload tons of CPU work to efficient specilaized accelerators.
Last edited:
It's not 15% IPC but 15% ST uplift