Discussion in 'Frontpage news' started by Hilbert Hagedoorn, May 29, 2020.
time to retire my ol 8700k ands 1080ti pair, they served me well
5nm cpu and 7nm gpu here i am
Honestly, working at TSMC is probably one of the safest workplaces in the world. No virus is gonna get through their cleanrooms.
Nothing to see informative with your post as well. We don't care what you and you're wife "karma777police" is going to buy on your next wedding. Anyhow we like that AMD is improving on a yearly basis and we're not being fed with the same thing as Intel does. But clearly since the word "smart" is not in your vocabulary, Intel will love milking you for years to come.
To AMD, just ignore these kind of people, just continue what you're doing, we like the innovation, we've liked what you've brought in the CPU world.
My wallet is ready and willing to retire my 8700k.
The server market, then the XT's that close the gaming gap to nearly zero(especially above 1080p), now they are going after the laptop market hard and their APU's. AMD is going for the jugular.
This might be my next new buy
If they effectively remove the CCX, and keep the 8-core chiplet, they remove the penalties associated with in-CCX communication.
Unlike rumors, they actually have announced an 8-core CCX, and transitioning to 5nm. The question is what they decide to produce there, nothing else.
Apple makes the best performance/watt CPUs in the market. They are literally in a different league.
Maybe we have forgotten it because it isn't happening. I don't think it's understood what "doubling" the IPC of Core would mean. Even some of the greatest jumps Intel ever managed (like... Core, or even the Pentium), were in the 10-50% IPC improvement range, in the best of cases. With the exception of Pentium 4, improvement usually came with frequency, and frequency also needs a mature process.
IPC from Intel increased with the Pentium, the Pentium Pro, it dropped with the P4, it rose again with Core, it rose again with Sandy and a tiny bit with Skylake.
Now we're at today. What kind of uplift do you expect at this point?
Where do you get this exactly? Willow Cove is a 10nm part, Rocket Lake is a 14nm++++++ part.
Check the Wikichip entry for Rocket Lake. Reported compilation target for GCC for it is -march=skylake -mtune=skylake.
Is this some sort of fever dream? Are you channeling Gordon Moore?
A new factory takes years to start producing properly, Intel hasn't even sorted their 10nm production, and that US factory (which is a political move, as location doesn't matter for things like these) will take years before it can produce a single wafer.
That won't happen for at least the next couple of years, and if you want PCIe5.0 from Intel, you will probably need to wait more than three years.
I suppose they need this change to improve per core performance over current generation and Intel chips.
The 7nm 4000 series seem to only bring 5-10% more performance rather than the expected 15%+
(leaked 4000 desktop apu score)
Link? If you are not able to provide link. Are you able to provide test used?
TDP limit of APU? Was iGPU under any load while testing?
It is important to remember that while one can set AMD's CPU to tick at specific clock manually, default operation is boost within set TDP limit.
AMD's CPUs are not going over this limit as long as they are measuring their own power draw properly. Which is the case unless EDC=1A bug is used intentionally.
While intel does give long term all core power draw at minimum clock for usual workload they define (which nobody outside intel knows) as TDP=SDP. And their CPUs have "short term" (read long term) SDP boost which is even twice as high.
Then again 4000 APUs are... out for mobile, and those are Zen2 based, not Zen3.
As far as benchmarks go, I have seen 3DMark benches for G-APUs (desktop) and those did not run with dedicated GPU, therefore iGPU ate power in each test from CPU. (Affecting even Physics results.)
Zen 4000 so far are APU only, so they are based on Zen 2. It's been like that for series 2000 & 3000 too. Those APU were based on the previous gen architecture (Zen & Zen+) and branded on the "next" series numbers.
It is odd, it can lead to mistake sadly. But that is AMD choice since it began the Zen era.
If is IPC improvements, clock has to be factored out. Is instructions per clock and not instructions per core.
Regarding the cache, no they can't. Once the CPU is design complete, you do not start from scratch adding more l1 cache.
I assume that needs replanning and redoing and retesting.
30% IPC is a LOT LOT of performance and likely is not gonna happen. 15% would be already a great result
Architecture does not need to be exclusive to process. It is well known that they are bringing Willow Cove to 14nm for Rocket Lake. And Willow Lake will have IPC improvements over Skylake, cache redesign, and PCIE 4.0
I think that this unexpected move to 5nm is the last step to the "Intel is dead". This will be the perfect marriage of a enhanced Zen3 core plus 5 nanometer that will result in a GREAT performance and low wattage.
This new CPUs will match Intel gaming performance that is the last thing that Intel has to offer now.... with 30% more multitasking performance costing 20% less who in earth will buy any Intel CPUS? Only employees!
this year. But there has been a c Tutuapp 9Apps ShowBox hange, the series was supposed to be released based upon an optimized 7nm+ fabrica...
Do we have any sources for this? Also bringing something to an older process other than the one it was made, sounds like a huge down grade.
Intel not only is not dead, they literally can't make enough chips to satisfy demand.
Going forward Intel is trying to make the architecture process agnostic as much as possible. So they can easily back port to an older process. There latest roadmaps speak to the back porting. it is a downgrade but when you can't meet demand they will have to keep making parts on 14nm++ even when they get 10nm sorted out.
Backporting literally means redesigning, it's not exactly as if they apply an expand filter in Autocad
This is genuinely interesting. Anybody else remember that AMD was also making an ARM Zen? Did they actually finish that?
And many people don't know that fact which means it is misleading. There was no real reason to use "confusing" numbering scheme other then fishing for non-informed customers which is probably 90%+.
Thing with these rumors is not "if", it is "when" 6+ core APUs will get released. I can safely assume that many people will pay extra for these because of highest first number and pair them with dGPUs.