- Mar 3, 2017
- 1,623
- 5,894
- 136
ATI already proved to be worth it with the consoles. MI300X is finally the big stick to truly make the investment worth it.AMD should not have purchased ATI. AMD once owned 33% of GloFlo aka. Global Foundries and spun it off their investment. It's not like AMD wasn't making money. They made big investments that were not smart. AMD was the top of the CPU market for two years. First to 1ghz CPU's, brought 64bit processors to the market when Intel said 32bit was all consumers needed.
Xilinx no, Pensando oh yes.MI400 will prove Xilinx and Pensando to be worth it too, don't worry.
There are so many designs in flight at Intel now that even if it is officially Zen V: the Return of Conroe I don't see anything dominating like Conroe did for so long.
ATI was a great company. AMD was not well run in the late 2000's. They have largely squandered the ATI brand for the most part. Nvidia's market share today tells the story of the epic battle between ATI and Nvidia. I had the ATI 9700pro and ATI 9800pro. I also had the original Nvidia Geforce 256, Geforce 2, Geforce 3 Ti4200 and several of ATI/AMD cards after that.ATI already proved to be worth it with the consoles. MI300X is finally the big stick to truly make the investment worth it.
MI400 will prove Xilinx and Pensando to be worth it too, don't worry.
Pure-play already won, so AMD was right to spin off GF. Intel must do the same or they won't make anything competitive, not when 10x more engineers work on TSMC's ecosystem due to no COI.
They ain't gonna do anything without anchor customers which are yet to materialize.Intel is going to sell silicon to 3rd parties on their fabs
But they didn't, not with Conroe et all.
That was all pure merit.
Eh they were still aggressively bleeding share before Conroe so the kickbacks weren't helping much.The FTC judgement was in 2009-10, that was very much antitrustville. Even if they stopped the kickbacks, they kept up the compiler and foundry shenanigans
AMD had market domination starting in 2003. Processors like the Barton 2500+ destroyed P4's. Conroe arrived in 2006. AMD x2 processors did not arrive until 2007.Eh they were still aggressively bleeding share before Conroe so the kickbacks weren't helping much.
? Athlon X2's were 2005 lol.AMD x2 processors did not arrive until 2007.
There may not be Xilinx IP in it, but they have such a wide area of expertise that I'm sure they make an impact.Xilinx no, Pensando oh yes.
They're gonna stick that 800G goodness everywhere.
Time flies. So 2005. I had the 3800+ X2. I was going by memory.? Athlon X2's were 2005 lol.
I wanted Athlon based PCs back in 2001 for a firm i was working in, but the local biggest IT shop didnt have any.AMD had market domination starting in 2003. Processors like the Barton 2500+ destroyed P4's. Conroe arrived in 2006. AMD x2 processors did not arrive until 2007.
yea it's them vs Nvidia.AMD really will become a one stop shop for any form of high performance computing, ideally all on one package.
And AMD has many friends. NV has momentum and VC fed startups feeding the impending bubble.yea it's them vs Nvidia.
That one's also nice but not really a selling point.but never about FP IPC
None of that stuff is really all that relevant.Int reg, Int scheduler, FP Reg, ROB? (Currently 224/96/192/320)
I’m going to get a lot of shade thrown at me by adroc for being a terminally online skeptic™ but....So if lots of Zen4 workloads are front end bound , we have a big front end , hopefully super duper prefetch + predictors and we have 50% wider ALU / execution. On your bullish statements im now upping my YOLO prediction from 33% to 41% IPC over Zen4.
This is to make -D work.Turin gets a 33% membw bump (DDR5-4800 to DDR5-6400) with a new cIOD
Exactly 25%.Turin also has a 500W power budget which means that depending on the SKU there can be 30%+ more power available
20% is puny numbers for a bigger core + tock.but I'd still personally advocate for an attitude where any IPC increase past 20% is considered a win.
Hmmm....I have sources that indicate the structure of Zen 5 looks somewhat like this
View attachment 92437
SMT4 + CMT. Extremely smart. One thread can schedule up to 1024 bit of FMA in single cycle provided the other 4 threads are not using it.
(SMT4 Bulldozer should give away that this is a joke but just in case I added this disclaimer)
Dual decoders are the new hotness (for x64 anyway).Either hinting at 30% Ops per Cycle Improvement, or that there's some sort of dual decoders. Or both?
You say that but both Intel and AMD have only done that once in the past 20 years. Intel did it in 2006 with Conroe, AMD did it in 2017 with OG Zen. The one thing that both of those architectures have in common is that their predecessor was underwhelming making that massive IPC bump possible.20% is puny numbers for a bigger core + tock.
Even Intel can get close to that and their big core teams are like the worst in the industry.
That doesn't mean that all of the other SKUs won't receive a benefit from that membw increase.This is to make -D work.
Check any 9554 review relative to Milan and ignore the membw-shredder workloads a-la CFD.
I was using the 9654 default 360W TDP, but I grant that I don't know the power limit Turin will ship with but rather the upper limit (500W).Exactly 25%.
The back-end yeah.that it likely looks more like A14's "Firestorm".
Apple did Cyclone and Twister in 3 years (and maintained a pretty nutso CAGR afterwards for quite some time).You say that but both Intel and AMD have only done that once in the past 20 years.
It really does mean that, go to STH and look at their Genoa review.That doesn't mean that all of the other SKUs won't receive a benefit from that membw increase.
It's a 25% skt power bump and that's all you need to know about it.I was using the 9654 default 360W TDP, but I grant that I don't know the power limit Turin will ship with but rather the upper limit (500W).
Nothing but Strix aligns itself to the MS cycle.Latest statements from supply chain, said Zen5 postponed a little bit to fit into Win12's release.