Read more.Quote:
It would easily eclipse the 3 to 5 per cent IPC gains seen in the move from Zen to Zen+.
Printable View
Read more.Quote:
It would easily eclipse the 3 to 5 per cent IPC gains seen in the move from Zen to Zen+.
So having an IPC gain of 13% over Zen+ means that if you had a 2700X and 3700X freq locked to 4GHz then Zen 2 would be 13% faster in varying workloads?
Might have got this wrong but if a 3-5% IPC uplift gave 10% all core performance uplift... a 13% IPC uplift could potentially give way over 20% uplift....Quote:
Originally Posted by HEXUS
Well, that 10% uplift was partially built on clock increases of between 5% and 10%, so it depends on how exactly what clock speed they can get out of the 7nm process. A process move/shrink doesn't always improve clock speeds (think Richland -> Kaveri), but a 5% clock speed increase plus 13% IPC increase would work out to around 19% actual performance boost.
AMD net income (2017): $43 million ...intel net income: $9.6 BILLION !!.....lets do the math 9,600,000,000 / 43,000,000 = 223 TIMES. AMD needs a SUPER miracle to take 50% of global PC/server sales
I don't think that's comparable because Intels spiders web of product revenue streams covers far more than AMDs. And it doesn't need a Super miracle, it just needs to provide a good value proposition.
AMD aren't planning on taking on Moby Dick, Lisa said this quite clearly in their recent interviews. They just want to release good products that show up the competition in a different light and let it grow from there.
I think you are focusing on the wrong thing here.
If you build it they will come approach works if the value proposition is good. I suspect Nvidia is about to find that building something awesome at any cost will not necessarily translate into profits.
all it will take is a 30-40% sales over all and a lot will happen (if it's not atm) to intel teams ...
intel can't afford a drop in sales or heads will roll very quickly ..
they were short sighted and amd did a nice broadside as long as they can keep going they may get to 35% before intell regain composure
Why do i get the impression that 'Bits and Chips' picked up on the news that Zen disabled FMA4 because it's suspected to be broken and automatically assumed Zen2 would fix it and that would result in a 13% increase in IPC for 'scientific task' workloads / benchmarks.
Out of curiosity what's 'Bits and Chips' reputation like when it comes to leaks and rumors?
VEGA was a compute card, it absolutely smashed everything for mining. Doing 2000HS in moner / 45MH ETH for under 120w! When a Titan was doing 1300hs Monero / 31MH Eth at the same power but costing x3 the amount. This is why they sold out for over a year solid. AMD couldn't make them fast enough. From a business perspective that's a huge success.Quote:
They were quite close on Zen and utterly crap with Vega.
Just because its not what you want to use it for, doesn't mean its not a success elsewhere.
I know this as i have two AMD gaming rigs all paid for by mining on a bunch of vegas.
Technically the instructions aren't disabled, they only removed it from the CPUID flag list. The CPU will still accept the instructions if the software issues them. Presumably this is for compatibility's sake, can't have software compiled for AMD CPUs crashing out because of a buggy SIMD instruction set. It could account for the cited IPC improvement, but it'd come off as a bit of flakey marketing to me.
this % gain is fine, if the price increases are only an additional same amount. if they are 35-60% higher, then amd is can go kiss my butt, and they are no better than intel. funded by saudi filth money anyway
The irony of posting on a forum when Saudi weapons sales have helped prop up our own defence industry to over £43 billion over the last few decades.
Also getting confused when it was Abu Dhabi who did the investment and not Saudi Arabia and most of that was in Global Foundries.
Still my favourite graph of all time:
https://trends.google.com/trends/exp...ated%20quickly
Since IPC is the main area AMD are slightly behind on this sounds promising.
Do you think there'll be new mobos released too even though it's AM4 compatible?
Show us on the dolly where AMD touched you.
AMD is widely reporting to be in touching distance of the share of market split between Intel and AMD. Clearly Intel are involved in plenty of other markets/sectors/industries. Also if it WERE the case that Intel revenue was so high, AMD so low and AMD were still stripping the market share lead from Intel, that's rather to AMD's favour.
Edit: My prediction. It's the Zen2@4.5 = 8700k@4.9 thought train, not my own but it seems eminently sensible.
13% seems entirely reasonable to me. Zen was a completely new microarchtecture, and ANY new grounds-up design will have warts. Fix those problems and you get a nice boost in IPC. Not a surprise at all. Zen+ only got a minor bump in IPC because they could only make minor changes to the design, it's basically Zen with a bit of tuning. But Zen2 is where they go fix all the stuff that wasn't working right :-)
Probably not this time. The biggest performance boost in Zen+ came from reducing the cross-CCX memory latency, an improvement that doesn't help single core IPC at all but speeds up multi-core benchmarks and applications. That's one reason that AMD didn't bother with the full Zen+ improvements in the APUs (2200G and 2400G, Mobile Ryzen, and Embedded Ryzen), even though they are part of the second generation; those parts only have one CCX so the cross-CCX changes are meaningless.
AMD's net income is low for three reasons. They're pouring a lot of money into R&D, they are pricing their CPUs aggressively to gain market share, and GLobalFoundries isn't all that good at making chips. AMD's yields are likely to be far worse than Intel's. Intel has had problems over the years at designing chips, but they are really good at MAKING them.
Yields of AMD chips from TSMC remain to be seen. Also unknown is what sort of clock rates AMD's designs will be able to hit on TSMC's 7nm process. It could turn out to be a disappointment, or something that will shake Intel to the multiple cores.
They didn't improve cross-CCX latency specifically, although that was one of the systems effected, they 'improved' (they actual fixed a bug but that's academic) in the memory subsystem so the improvement (3% IPC) was seen across the board in single and multithreaded workloads, and everything from Epyc to mobile and embedded.