Discussion in 'Article Discussion' started by bit-tech, 6 Jun 2018.
A bit like the upcoming sequel to Top Gun - I don't need it, but I want it.
Intel: "We're number one again! 28 cores, AMD only has 16! Suck it!"
AMD: "32 cores." *drops mic*
Man, I love this return to competition.
You feel the need... the need for speed?
This is ridiculous, now there are thirty two of them!
I can only imagine this is of interest/need for high end Render or CFD workstations. Interesting that both AMD and Intel are somewhat gouging their own server chips to one up each other. At my last job we used to run CFD software on 8 Core Xeon workstations but I would definitely look at moving (if I still worked there) to Threadripper for bang for buck (ECC memory wasn't really necessary as the simulations would through errors more regularly than you would get a genuine memory error).
It was so obvious. "Here is a 16 core which just so happens to have two extra disabled clusters with 8 cores on each, but honest, 'guv, we're not making a 32 core".
It's silly, but as cool as hell iced over.
Just a little bit insane to do it on the current manufacturing process instead of waiting for the shrink in 2019...
But if they can price it competitively versus the £1700 i9-7980Xe it will be a very sexy chip for rendering.
They already have the chips. Why wait? Release another in 2019. Easy money.
I hope it’ll be priced well... the current EPYC chips are expensive, but they are awesome.
The steep performance penalty of the interconnect between the modules combined with the inevitably low clocks seriously limits the appeal of Ripper Ridiculous Edition.
And we know already they'll double the cores per module next year, so they can do 32 cores in 2019 with just two modules and it will make 2018 32 core Ripper look like a complete joke.
Of course if they are willing to put 4 fully enabled modules in a consumer cpu then the question in 2019 won't be what AMD did in 2018, but how many years it will take Intel to catch up to the 64 core Threadripper 64 (god I hope they go for that pun based name if they are crazy enough to release it).
Think about all the VMs....
They drooled themselves dry over the 8 channel memory of Epyc...
So the clock speed wars are over and we're now into the 'hold my beer!' phase of core counts.
I'm waiting for the Intel Core Neuron, with 100 billion cores.
Whenever we see proper review samples I'll show some interest. AMD has not fixed clock speed and nobody knows about overclocking potential, Intel uses a 1 HP cooling unit to get their 28-core chip to 5 GHz. And, as of yet, nobody has a clue about pricing.
Wake me up when there are samples. It's all exciting stuff but there is no substance right now.
I wouldn't expect either to have an all core turbo above 3.5ghz tbh.
And just for fun I'll throw out a predicted oc headroom using normal (full custom loop, min 360 proper thickness rad, none of this AIO stuff) watercooling:
3.8 32 core Ripper
4.2 - 4.3 28 core i9
I'd say you're looking closer to 4 - 4.2 on the TR but only, and only with a high-end custom loop.
Realistically, the i9 will be for the 1% of the 1%'ers. Let's be real. It's a $10k Xeon chip with an unlocked multi. Intel aren't going to drop the price *that* drastically, or else they'd cannibalize their own server market.
4 maybe if they bin them very carefully (or you win the silicon lottery) but if you need any extra voltage you can kiss the puny power delivery on x399 boards goodbye.
In theory, they should be capable of delivering 750w+ to the CPU ... but that all depends on how well it can be cooled.
Knock off lets say 152W for storage, ram, idle gpu, psu efficiency, then double it since it is essentially two 16 core Rippers and you get 740W oc'd (and that 1950x used as a baseline was only at 3.85ghz), so even if we give them the benefit of the doubt and assume they can handle 750W+ it would still end up very marginal.
Now of course AMD said they would bring the same improvements to TR as they did to the 2xxx Ryzen, but 2xxx Ryzen doesn't provide any too significant power savings over 1xxx Ryzen, so the same will probably apply to 2xxx TR.
TR was typically the better binned die so ran with lower voltage for same clockspeed.
Will be interesting to see how precision boost and a XFR2 work on these, as its 4 Ryzen die instead of 2 like last time so might mean we will see upto 8 threads boosting to >4.35Ghz then slowly rolling off with more cores and heat, that wouldn't be too shabby.
Thus igniting the "50 Month Core War"...
But will it run Minesweeper?
Separate names with a comma.