Discussion in 'Article Discussion' started by Gareth Halfacree, 3 Feb 2016.
Could support 64 threads per socket.
I think Zepplin is an interesting choice for a name. The original was heralded for being a great human achievement, but it also had a catastrophic failure. AMD's architecture will have only one of these outcomes.
Names don't mean anything really. Performance talks BS walks. It could be called Royal Douche, if it's good I wouldn't care
take two identical products, one called i9000 x pro elite, the other called Royal Douche.
I'll take two of the latter please...
as for AMD, well intel got it bang on when they pushed for the best fabs in the world decades ago. The have the best sand -> money machines and nobody can better them.
Now, if only they'd put a 'Led' infront of that 'Zeppelin', they'd have a Whole Lotta Lovin' Going on...
So the real problem I have with AMD is that the multiple core strategy flopped in *2008*. The multi-core Opertons absolutely failed to capture the market in light of Intel's growing IPC gain.
8 years later, the lead has gotten enough where the top chip from AMD is at best selling for *less* than a locked version of Intel's quad cores despite the FX series having double the cores. Not to mention that AMD still hasn't moved away from the Southbridge design.
It looks to be then end it seems.
So why did http://ark.intel.com/products/84682/ come out last year, if massively multicore was a big flop?
There's a market for stupidly high core counts, and I assume the Zeppelin part is destined for the server market.
AMD used to do quite well out of their Opterons. Sure they weren't as good as the Intels but you got far more cores for less money. As for the amount of cores? well you could have a quad 16 core set up so yeah, I would say they can be used alright.
The problem is as time went by the Opterons became less and less competitive so AMD just seem to have stopped making them completely. I'm actually surprised that Zen server parts haven't started rolling out already, given that we usually tend to get them long before the desktop hand me down version. Maybe they're going to launch the whole lot at once?
Any way, even in desktop systems the core march stomps on. Broadwell E is coming soon, with 10, 8 and 6 core variants. So Intel are finally letting go of more of their precious cores for desktop systems.
I was told the other day that Far Cry 4 is it? will not even run unless you have at least an I3 with 4 threads total. So we are making progress, it's just terribly slow.
AMD has already announced that the first Zen products will be enthusiast-centric, then server and mainstream to follow.
Oh OK that's cool then at least I won't be tempted to buy a very expensive server board just to try one without waiting for ages
From what I heard, this 32-core will be an Opteron (though, that doesn't mean it couldn't be marketed toward workstations).
Anyway, I think focusing on enthusiasts is a wise move. Sure, the server market is where the real money is, but it is very difficult to convince IT admins to switch to a different platform if they don't like the results of the mainstream counterparts. It isn't really fair, because very often mainstream parts often have very little resemblance to how server software performs. Many of AMD's current Opteron series are actually pretty good, but nobody would know because they jump to the conclusion that the architecture is bad.
I suppose the same could be said about PPC, where the desktop parts were distinctly behind Intel's, but PPC servers were and still are relevant.
The server guys will be reading the Spec results and their specific analyst reports. Server are increasingly scenario specific. It's a different kettle of fish.
A high-core count chip will really hurt Intel's carefully crafted economy though.
Is this statement directed specifically at the server market? If so, how will this affect the PC market both for small business and the average home or consumer PC user? There are few programs that require or benefit from a "high" core count [over 4 to 8] for that market, especially PC gaming where high performance per core is more important than lower performance per core and a high core count [16 to 32 cores for example]. Of course, "high power" consumer programs would benefit from GPU usage, especially if Nvidia would open source their software and as yet that has not happened. That is why bit-coin mining was so successful with AMD GPUs, AMD makes their GPU software public.
I suppose that was a rather long comment, my thought process began to run away with me even though it is 12:30 a.m. here. Time for bed.
More the PC market actually. Intel puts $ on adding 2 virtual cores (pentium to i3), then $ again to replace them with physical ones (i3 to i5). For AMD to come in and go 2-4 more cores than Intel at the same price point for within ~5% same performance/power the market would go nuts. It totally doesn't matter if programs require this many cores - people will want it. The more = better is such a strong selling point. In the US the i7-K series is ALWAYS the bigger seller versus i5-k, despite the fact those extra cores are only virtual. In smartphones the Chinese market demands 8-core phones at a minimum now. Those are big markets with big preconceptions.
Server market is based on real world performance, power consumption and overall ROI. You have to play the total cost of ownership rules, which is why Intel is building value there by throwing cores at Xeon chips but not consumer.
Separate names with a comma.