Return to Level1Techs.com

AMD CPU Road map *SPECULATION*

amd

#61

If we ever threw shit out, I would. We keep it until it dies. That’s just how our openstack clusters work. Actually saves us a fair bit of money in the long run.


#62

My bet is $1500-$2000 for a full fat 32 core threadripper (depending on just how aggressive AMD want to be. I believe they can make decent margin on $1500 for it). Lets split the difference and say $1750 with that air cooler.

Intel is going to get torn a new one in HEDT this year.


#63

80mb of cache, doubling from TR1. Looks like another winner but going to struggle against Intels 5ghz 28 core part.


#64

1500 for 24, 1900 for 32. Gives sales room in a couple months for Xmas.


#65

Pretty sure I saw 1500 for 32


#66

It is well established that the Intel 28 core is a Xeon 8180 that has a base clock of 2.5ghz and boost of 3.8ghz. Doesn’t seem that threatening to me against these Threadripper specs.

Will have to see what Intel pulls out of the hat for this one.


#67

Intel’s 28 core part at similar clock doesn’t currently exist.

Intels current 28 core part (not for HEDT, but enterprise server platform) at lower clocks is between $8k-13k US…

Sure. It has higher memory bandwidth than TR2. But it isn’t in the same market, and is massively more expensive… TR2 is likely to be cheaper than intel’s 18 core…


#68

Like i said i reckon that is TOTALLY doable, at a profit for AMD.

They’re selling 2700X for 350 or whatever, retail. And that is profitable.

  • 4 of those is ~$1400 at retail, including profit margin
  • only need one cooler (if they even supply one, 2700X ships with a funky cooler included in that cost)
  • one box/packaging
  • one supply chain logistics overhead
  • one support/warranty backing overhead
  • sure, there’s the substrate packaging process involved, but surely that’s under a hundred bucks

I’m not sure what AMD’s margins are, but based on the above i’m sure they’ve got a lot of wiggle room to be able to get it out for $1500 or maybe even less. Selling it for above $1500 would be making bank…


#69

Tech news has been so quiet of late. I am really looking forward to next months official announcement. Rumored above 13th Aug. Making me want to build a new machine.
I will settle for a decent priced Vega GPU and learning / tinkering with passthrough.
AMD said they want to be disruptive. Boy are they. Even Epyc, I posted in Wendels thread about how a company is looking forward to 7nm EPYC and 50% more cores next year in a 2U high density rack. Thats 48 Cores next year. Im still holding out that the non-dual socket EPYC’s could even get to 64 cores still.
What EPYC gets we get next in 7nm silicon on AM4 and TR1.


#70

Intel definitely has AMD covered on a few niche workloads at the extreme high end. But… these are few and far between and at much greater cost.

I suspect that the number of customers well served by things like the 8180 that wouldn’t get a benefit from EPYC or Threadripper at 30-50% of the cost with higher core density per rack are very few indeed.


#71

Yeah, I agree the additional cache is going to be nice.

Also, it looks like some people didn’t get the satire. :smiley:

If that’s the case, damn. That’s a really compelling price point.


#72

One thing Josh from PCPER brings up time and time again. Is that Intel has massive clean space FAB to build tons of chips. I mean look at intel’s market.
It would good for AMD to sell all the production they can. Can TSMC meet the demand for 7nm GPU and CPU production if there was a large market shift.
I have no idea on capacity to ramp up or down production. Still would be good to see AMD clawing back into a more even market.
EDIT: I guess intel dont have ton and tons of 10nm FAB yet so there is that.


#73

The beauty of AMD’s strategy right now is that they can just pump out millions of the exact same dies and use them like lego bricks to build whatever products are required. If demand changes, they can link differing numbers of them or disable/enable individual cores quite simply as required.

Intel are making how many different dies right now? Kaby Lake, Coffee Lake, Whiskey lake… Skylake X. All in different core counts (low, high, extreme core count dies), etc.


#74

And to be clear, this is beneficial because of R&D speed or what, specifically?

They’re getting there on the desktop side. I think new sales were nearly 40% AMD this last couple of years. They still have a long way to go on the enterprise, but we knew not many were going to adopt EPYC 1. EPYC 2 might be a much bigger adoption because they’ve shown solvency and that EPYC works.

That’s so true. That’s part of what fascinates me about EPYC and TR. I want to get my hands on one of those chips so badly, I’d love to play with it.


#75

Yeah, i mean for intel… let’s say that they want to build a bunch of 8180s… or rather, XCC dies.

They’re big expensive dies (both in terms of RRP, and to actually produce), that may or may not sell. Making the choice to build them is a significant risk vs. using that manufacturing run to build smaller dies. They can’t just take unsold XCC dies and stuff them in a laptop or whatever.

AMD just keep building zepplins, if market forces dictate demand for EPYC or Threadripper drops, they just sell them as Ryzens. If demand rises, they link them together.

Being relatively small dies, the yields are likely to be good, too.


#76

I think his point is that when intel does get 10nm working they have clean space to retool to 10nm and out supply AMD. Its not really R&D because intel dropped the ball there. They say next year.

If AMD gets record sales / demand. Is there a production cap on what AMD can actually build ?

Josh brings it up as AMD can’t threaten intel because they cant built fast enough. This is all server CHIP demand. The cream of profit for CPU’s in numbers and $$$ per chip.

EDIT:
When I say is there a cap. I know manufacturing, of course there is. Can AMD through manufacturing chips and selling them give Intel a big black eye in the server market or is Josh right.
I know PCPER shill a lot. Even now Ryan still is Intel / Nvidia. Josh seems to make sense and its interesting if it will take a decade to ramp up manufacturing, build more fab’s leading into 5nm. Meaning well beyond ZEN3+++.


#77

IF intel get 10nm working…

If they don’t get 10nm working in a hurry, the other fabs are going to eat their lunch.

I’m sure TSMC, Glofo and Samsung would much rather be building PC processors with the much higher margins than dinky little mobile chips…?

edit:
remains to be seen now how much of intel’s IPC advantage is actually due to fab or advanced design, and how much of it is “optimisation” along the lines of meltdown…


#78

Yer AMD did dodge the bullet a lot on vulnerabilities. Intel spinning like a top trying to say it’s everyone. AMD did make some smart engineering choices that helped.

My hairs on the back of my neck stand up a little everytime Intel gets a 90 day or 180 day researcher disclosure. AMD’s are always zero days, end of the world.


#79

I adore when people come to The Onion or Duffel Blog and go all real world in the comments, hilarious as hell.


#80

I thought they were working on breaking their exclusivity deal with GF.

If they hit production limits, that means they can start raising their prices a bit and not lose customers, because clearly people would rather have them.

I think they can. They used to, back in the Athlon days, they had the superior chips and every server I see in the recycle bin from that era is an Athlon.

I think AMD has access to the production resources to hit maybe 40-50% of new purchases. You have to remember that AMD has significantly higher yields on the XCC field than intel does, so their lower wafer counts are made up for, partially, by their high yield.