-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Canard mentions second gen' EPYC, if true, nice!
- Thread starter Fishbait
- Start date
This is impressive stuff from AMD !!!
I wonder what intel would do to try to counter it ?
Pay firms not to sell AMD stuff?
CPU's APU's and GPU's.
@Rroff They could do an MCM of their LCC dies since they can get those to low TDPs, but their current strategy seems to be using monolithic dies.
They have tech like EMIB so they'll probably venture into MCMs again with next gen Xeons if AMD is pushing these kinds of core counts with next gen Epycs.
I'm still waiting to see what kind of impact Epyc had in data center, maybe we'll find out when Q4 earnings reports are out.
They have tech like EMIB so they'll probably venture into MCMs again with next gen Xeons if AMD is pushing these kinds of core counts with next gen Epycs.
I'm still waiting to see what kind of impact Epyc had in data center, maybe we'll find out when Q4 earnings reports are out.
Within the context of the link you were referring to at the time - it has to have been someone putting 1 and 1 together and making 3 - literally the next day after the information referenced AMD reaffirmed commitment to GF for the bulk of their production and at the time they weren't one of the initial group of customers testing TSMC's 7nm. As I said they might make use of TSMC for future products but the already big order bit doesn't mesh with any other information.
I never linked anything. The contract is for APU's CPU's and GPU's on TSMC 7nm in Q1. Maybe that includes these chips.
Last edited:
Fair enough it was CAT-THE-FIFTH who linked to it and because I have them on ignore I only saw it through your referencing their post and didn't realise at the time you were quoting someone.
I wasn't quoting anyone. I said AMD had signed a big deal with TSMC for them to produce 7NM chips. You said that wasn't the case and few people provide links that referred to the deal in one way or another.
Anyway, regardless of what factory makes this chip it will be very interesting and probably the best in its markets.
CPU performance is really starting to boom now AMD have low cost fabs at their finger tips.
Last edited:
https://forums.overclockers.co.uk/posts/31073219
At some point before you edited or something that post showed a link that is actually what Cat posted or something which at the time is why I thought you linked to it (the forum tries to remove content posted by people on your ignore list but it doesn't always get it right which can be confusing especially when it looks like people are talking to themselves LOL).
Its the only mention anywhere of a big contract with TSMC and was essentially denied by AMD's conference IIRC the very next day.
Last edited:
Yeah Intel put back the 7nm process and went with 14nm/10nm so AMD will be on a smaller process before Intel in the first time since... well maybe forever or at least since AMD sold the Desden factory.
GloFo's 7LP 7LPE won't actually be smaller than Intel's 10nm. Samsung/TSMC's "7"nm would be somewhat on par with Intel's 10nm, but by the time TSMC ramp that up, Intel will be on 10nm+ which should have improved density.
For example Intel touts its 10nm of reaching 100.8 million transistors per mm2, while TSMC and Samsung "10"nm reach around 48 million transistors per mm2, which is closer to Intel's 14nm.
For example Intel touts its 10nm of reaching 100.8 million transistors per mm2, while TSMC and Samsung "10"nm reach around 48 million transistors per mm2, which is closer to Intel's 14nm.
Last edited:
There is no consistency with any of it any more - you have to look at the gate and fin details and what kind of clock speeds they are actually achieving to get any real idea.
True, we need more specifics to make a proper comparison, but we're basically in the wild west when it comes to marketing nodes nowadays.
GlobalFoundries especially is stretching it with their marketing, seems they're calling it "7nm Leading Performance"...
Edit: Made a mistake earlier, 7LPE is Samsung's not GloFo's.
GlobalFoundries especially is stretching it with their marketing, seems they're calling it "7nm Leading Performance"...
Edit: Made a mistake earlier, 7LPE is Samsung's not GloFo's.
Last edited:
Who cares what it's calle or how its marketed. All that matters is cost and performance. I'll buy a 250nm chip if the price is right.
The issue is that based on what we know so far (albeit it's only specifics regarding CPP/MMP/SRAM), it's likely that GloFo's 7LP won't be as dense as Intel's 10nm, but I do agree that it doesn't matter much for the end user as long as they get a chip with good performance at a reasonable price.
Soldato
- Joined
- 14 Apr 2014
- Posts
- 2,602
- Location
- East Sussex
This is all very very nice to have out there and would love this to be true - claims seam reasonable, but a few things it's got me thinking about:
- This is probably indicative of what we can expect from Zen2 it terms of 3200 stated support and cache per core size etc, as Amd have the common Zen building block that all Ryzen, TR and Epyc CPUs share
- Must be what? 3 years away? Epyc has barely hit the channel yet, APU and mobile on the current gen tech probably aren't going to finish hitting the market for at least 6 more months either, probably longer! And once that lots is done we'll have another round of that with Zen+
- I guess this just be using the current spare empty space on the current Epyc CPUs etc, rather then them doubling core count and cache size using the same size real estate as the current chips (which have the dummy dies in place already infact) - their process won't have shrunk enough to do that in even 5 years, so that decision probably makes 3200 memory support an essential - as there'll be a more complex / infinity fabric with a greater number of inter connects, they'll need more memory speed just to match current performance for cross ccx memory access.
- And that's another thing, server memory speed! Right now the fastest ecc and or registered dimms available are 2667mhz - and availability is not great, that's going to need to change quite a bit in quite a quick time frame for us to able afford server grade memory at those speeds!
- This is probably indicative of what we can expect from Zen2 it terms of 3200 stated support and cache per core size etc, as Amd have the common Zen building block that all Ryzen, TR and Epyc CPUs share
- Must be what? 3 years away? Epyc has barely hit the channel yet, APU and mobile on the current gen tech probably aren't going to finish hitting the market for at least 6 more months either, probably longer! And once that lots is done we'll have another round of that with Zen+
- I guess this just be using the current spare empty space on the current Epyc CPUs etc, rather then them doubling core count and cache size using the same size real estate as the current chips (which have the dummy dies in place already infact) - their process won't have shrunk enough to do that in even 5 years, so that decision probably makes 3200 memory support an essential - as there'll be a more complex / infinity fabric with a greater number of inter connects, they'll need more memory speed just to match current performance for cross ccx memory access.
- And that's another thing, server memory speed! Right now the fastest ecc and or registered dimms available are 2667mhz - and availability is not great, that's going to need to change quite a bit in quite a quick time frame for us to able afford server grade memory at those speeds!
Soldato
- Joined
- 18 May 2010
- Posts
- 23,283
- Location
- London
Hopefully this theoretically mean second gen Ryzen might double cores! 
A 2700 16/32 for £300? Yes please. Even £400 would be a good price.
We can dream!
A 2700 16/32 for £300? Yes please. Even £400 would be a good price.
We can dream!
Well we're getting Zen+ some time around CES and AMD have already kind of qualified 4000Mhz memory for the desktop.
APU's will be out before the end of the year.
ECC speeds will follow market demand although when we're talking about 4 and 8 memory channels you have to ask how much performance would be gained and the benefit of registered memory is capacity.