|
Risky Bisquick posted:Only shipping 10nm mobile for 2+ years is just unbelievably bad. I don’t think it’s true because it would effectively be handing AMD enormous marketshare. You're also assuming that 10nm has better performance characteristics for desktop chips, which is not a given.
|
#
?
Apr 24, 2019 20:38
|
|
|
|
| # ? Jun 12, 2024 00:53 |
|
|
Going by the client chart I don’t believe that the server chips with gen4 pcie won’t show up until 2021 though, that would be a huge gap being filled by amd rome in the meantime. I could swear there were some differences between client and server roadmaps in the past, but perhaps misremembering.
|
|
#
?
Apr 24, 2019 20:40
|
|
|
I mean, Intel is now AMD ca 2012 so ceding parts of the market to the competitor isn't unexpected 
|
|
#
?
Apr 25, 2019 00:17
|
|
I CANNOT EJACULATE WITHOUT SEEING NATIVE AMERICANS BRUTALISED!
|
Rip to any hope of a Kirby Lake-G refresh.
|
|
#
?
Apr 25, 2019 00:30
|
|
|
suck my woke dick posted:I mean, Intel is now AMD ca 2012 so ceding parts of the market to the competitor isn't unexpected This is hardly true because they remain the default option in large parts of the high-margin market. Top laptop SKUs and the 4 socket or more market are going to remain entirely Intel, and AMD isn't even competing in those spaces.
|
|
#
?
Apr 25, 2019 01:09
|
|
|
What the gently caress is intel doing when '7nm equivalent' already exist in mobile phones and devices millions of people are using daily.
|
|
#
?
Apr 25, 2019 02:08
|
|
|
It probably doesn’t scale power/tdp wise right now to match 14nm+++++ clocks at an acceptable yield. If they have ES that can do it but the yields are garbage so there is no sense trying to commercialize it yet.
|
|
#
?
Apr 25, 2019 02:14
|
|
|
Twerk from Home posted:This is hardly true because they remain the default option in large parts of the high-margin market. Top laptop SKUs and the 4 socket or more market are going to remain entirely Intel, and AMD isn't even competing in those spaces. You're going to be able to do a quad socket intel workload on a dual-socket epyc 2 in the very near future. I'm sure intel can glue some more chips together to get higher core density but they're charging so much loving money at this point is more cost effective to buy more 2U dual-socket AMD servers even if you need twice as many as 4U quad socket intels.
|
|
#
?
Apr 25, 2019 02:21
|
|
|
coke posted:What the gently caress is intel doing when '7nm equivalent' already exist in mobile phones and devices millions of people are using daily. Not 100% transferable going mobile phone processor to desktop. As you scale up die size, you end up with more complexities and are prone to errors more frequently per wafer making it expensive as hell. A Snapdragon 855 (8C/8T) has a die area of 73.27 mm²(8.48 mm × 8.64 mm) while a 9700K has a die area of 174mm² (someone correct me on this if I'm wrong)
|
|
#
?
Apr 25, 2019 02:48
|
|
|
Intel can always go to bat on price, too. Which would be a funny role reversal... AMD leading on process and performance/watt, Intel cutting prices. Or if you wanna get really wild, they are already a TSMC and Samsung customer... not for CPUs, obviously. But Intel's got the cash to make it happen if they really need. I think its more likely that they just let cut margins for a few years, though. Enterprise and laptop (and all their other business units) will easily carry the boxed CPU enthusiast market for a bit... IF they really have a solution after 10nm. Cygni fucked around with this message at 04:03 on Apr 25, 2019 |
|
#
?
Apr 25, 2019 04:01
|
|
|
Intel has a lot of yield issues with these monolithic designs, I seriously doubt they will be able to beat AMD on price with the ability to do much more favorable binning on the chiplet architecture.
|
|
#
?
Apr 25, 2019 04:14
|
|
|
Intel literally can't make enough chips to meet demand. Maybe if Zen2 clowns on them something will have to change, but right now they don't really need to do anything.
|
|
#
?
Apr 25, 2019 04:30
|
|
|
BangersInMyKnickers posted:Intel has a lot of yield issues with these monolithic designs, I seriously doubt they will be able to beat AMD on price with the ability to do much more favorable binning on the chiplet architecture. I mean, their client computing division (which is laptop+OEM desktop+boxed CPU) posted a $3.6 billion dollar profit last quarter with 37% margins, and TSMC's 7nm is not a cheap process at all even at AMD volumes. They've got plenty of room to play with price.
|
|
#
?
Apr 25, 2019 04:54
|
|
|
Cygni posted:Intel can always go to bat on price, too. Intel haaaates competing on price and they have to keep ASP's up to some extent to keep their fabs and the R&D they do funded. That all goes double for the server market which is where they get a nice chunk of revenue from and where they look to be facing stiff competition from AMD when Rome finally becomes widely available. Cygni posted:Or if you wanna get really wild, they are already a TSMC and Samsung customer... not for CPUs, obviously. At that point I'd think it'd become questionable if they should even stay in the fab business. Or at least to try and stay on the cutting edge in the fab business. There were some rumors of that sort of thing months and months ago. They were essentially laughed off by most everyone, including me, but if that leaked roadmap is mostly accurate then I have to wonder if maybe its less crazy then it seemed at first? There have been some rumors of delays to be expected with Intel's 7nm process (which to be fair will probably be true with everyone else's more advanced "5nm" processes too) but nothing else so far I believe. No real hints so far that its expected to be the shitshow that Intel's 10nm has turned out to be.
|
|
#
?
Apr 25, 2019 05:57
|
|
|
Risky Bisquick posted:It probably doesn’t scale power/tdp wise right now to match 14nm+++++ clocks at an acceptable yield. If they have ES that can do it but the yields are garbage so there is no sense trying to commercialize it yet. Yup, I vividly remember a chart that showed refined 14++ ahead of the first two generations of 10nm. 10nm seems to be a failure and Intel keeps refining 14nm which moves the goalposts, so this doesn’t come as a surprise. I think Intel will do fine compared to AMD in the performance characteristics that I personally care about (high singlethread performance, acceptable multithread performance, low latencies). Granted I expect Ryzen to annihilate them in the high core count department and if Apple rolls out a 5nm A14 to go up against a mobile 8C 14nm+++ CPU it won’t be pretty either. eames fucked around with this message at 08:36 on Apr 25, 2019 |
|
#
?
Apr 25, 2019 07:34
|
|
|
PC LOAD LETTER posted:the 7nm process they're working on will run into similar issues and be just as uncompetitive. They've been developing 7nm and what comes after 7nm in parallel. Intel does a lot of business. They aren't just a CPU company. I don't even think other fabs have the capacity, given their other customers, to fab all of what intel makes. No one really knows if it will harm their business longterm, but they should bounce back by 2022 or so. I figured it'd be 2021, but that roadmap seems to hint 2022. I'm a mild AMD fan, but I really don't see the doom and gloom unless they have problems with 7nm too. Intel's 10nm was real ambitious and does some things the competition's equivalent process doesn't. These things should carry forward to future nodes, and the nature of EUV really benefits how Intel does things. Intel has a whole lot of safety in the server market. Even if AMD releases a better product. AMD's strategy there is just to sell to the biggest of big. OEM market is where AMD might start cutting in provided there aren't anti-competitive practices in place like in the past. The problem there is with public perception of Intel and AMD. Intel might retain a massive share in that market solely based on their brand. The enthusiast market is already getting massive amounts of share stolen by AMD. Sales at many places are 50:50 between Intel and AMD CPUs. Or even favoring AMD. The value proposition of ryzen is great at the low-mid end, and not everyone wants or can afford to drop the money on a 9900k. Intel also has no competition for threadripper, but that's a very small market. Khorne fucked around with this message at 13:10 on Apr 25, 2019 |
|
#
?
Apr 25, 2019 12:23
|
|
|
So I'm just gonna point out first of all the post you're replying to was pretty clearly speculative and not a "this is guaranteed to all play out this way" type post there Khorne.Khorne posted:Intel 7nm uses EUV. There wasn't any way Intel was ever going to have 10nm with EUV now, in 2017, in 2018, or much less even in 2015 (yes 2015) when it was initially planned to be out. EUV is one of those things that has been delayed for everyone for a long time and Intel planned their 10nm to not use specifically because they knew EUV wouldn't be ready in the needed time frame. Khorne posted:They've been developing 7nm and what comes after 7nm in parallel... but they should bounce back by 2022 or so. Their future or success isn't guaranteed and I see no reason to act like it is even likely when they're making a series of fairly expensive and nasty mistakes. Khorne posted:but I really don't see the doom and gloom unless they have problems with 7nm too. Khorne posted:Intel's 10nm was real ambitious and does some things the competition's equivalent process doesn't.
|
|
#
?
Apr 25, 2019 13:25
|
|
|
PC LOAD LETTER posted:So I'm just gonna point out first of all the post you're replying to was pretty clearly speculative and not a "this is guaranteed to all play out this way" type post there Khorne. quote:EUV is pretty hard to do, harder to do than even the ambitious goals that Intel seems to have set for 10nm, and only relatively recently has it become possible to do on a volume process. [...] quote:I'm not a process expert but everything I've heard about Intel's 10nm (at least as far as what Intel's 10nm was supposed to achieve, so far the real world results are falling short of those goals) vs TSMC's "7nm" is they were supposed to trade blows when it came to transistor density, potential clockspeeds, and all while at comparable power and heat envelopes and that they essentially were considered about equal over all. I didn't quote the rest because I agree with what you said. Khorne fucked around with this message at 13:52 on Apr 25, 2019 |
|
#
?
Apr 25, 2019 13:44
|
|
|
Khorne posted:I know 10nm is not designed around EUV. Khorne posted:Intel's 10nm uses cobalt in the lower two layers. I don't believe TSMC's processes uses cobalt. Cobalt should be a reasonable advantage going forward, but it may be part of the reason for 10nm's failing.
|
|
#
?
Apr 25, 2019 14:01
|
|
|
PC LOAD LETTER posted:Than I'm not really sure why you were bringing EUV into the discussion at all. *shrugs* Maybe I misunderstood your a point your previous comment was trying to make. 2022 is around when Intel 7nm should hit real-world production, and this roadmap seems to align with that. Khorne fucked around with this message at 14:14 on Apr 25, 2019 |
|
#
?
Apr 25, 2019 14:09
|
|
|
I read an analysis on Reddit (so mountain of salt) that Intel's fastest path to 10nm/7nm might ironically be to buy out GloFo to recycle the work they put into their cancelled 7nm node. Reportedly it was shaping up very well and the only reason it didn't go into production was GloFo's owners (some Middle Eastern investment fund) didn't want to pony up the cash to build more fabs so soon after getting the 14nm line going, a problem Intel wouldn't have.
Mr.Radar fucked around with this message at 15:23 on Apr 25, 2019 |
|
#
?
Apr 25, 2019 15:00
|
|
|
ItBreathes posted:Intel literally can't make enough chips to meet demand. Maybe if Zen2 clowns on them something will have to change, but right now they don't really need to do anything. they can't make enough chips to meet demand because an entire product line that was intended to meet demand failed so suddenly they don't have as many working fabs as they expected Mr.Radar posted:I read an analysis on Reddit (so mountain of salt) that Intel's fastest path to 10nm/7nm might ironically be too buy out GloFo to recycle the work they put into their cancelled 7nm node. Reportedly it was shaping up very well and the only reason it didn't go into production was GloFo's owners (some Middle Eastern investment fund) didn't want to pony up the cash to build more fabs so soon after getting the 14nm line going, a problem Intel wouldn't have. i would expect intel would have huge antitrust issues trying to buy out any major competitor and even if the US DOJ signed off on it i sorta doubt the EU would evilweasel fucked around with this message at 15:18 on Apr 25, 2019 |
|
#
?
Apr 25, 2019 15:15
|
|
|
evilweasel posted:they can't make enough chips to meet demand because an entire product line that was intended to meet demand failed so suddenly they don't have as many working fabs as they expected True, but the point is that even with that mild disaster, they're posing healthy profits and have plenty of pricing slack to play with if they need to, because even without the new production line, their current offerings are highly desirable and selling like hotcakes.
|
|
#
?
Apr 25, 2019 15:23
|
|
|
Mr.Radar posted:I read an analysis on Reddit (so mountain of salt) that Intel's fastest path to 10nm/7nm might ironically be to buy out GloFo to recycle the work they put into their cancelled 7nm node. Reportedly it was shaping up very well and the only reason it didn't go into production was GloFo's owners (some Middle Eastern investment fund) didn't want to pony up the cash to build more fabs so soon after getting the 14nm line going, a problem Intel wouldn't have. If it was shaping up well, why wouldn't they have sold the tech if they didn't plan on developing it themselves?
|
|
#
?
Apr 25, 2019 15:24
|
|
|
Gyrotica posted:If it was shaping up well, why wouldn't they have sold the tech if they didn't plan on developing it themselves? Presumably because there were only 4 companies persuing bleeding-edge nodes (Intel, GloFo, TSMC, and Samsung) and the other three already had their own equivalent nodes under R&D. That said, this was just an interesting idea I came across so I don't know how feasible this would actually be.
|
|
#
?
Apr 25, 2019 15:34
|
|
|
I would have to imagine that there'd also be substantial entanglement of their 7nm tech and IP with other tech and IP that they're actively using/developing and would rather not sell out to a competitor, especially if they're not under financial duress, which they aren't. Amusingly, they did just sell off a 65/45nm fab to ON Semiconductor.
|
|
#
?
Apr 25, 2019 16:16
|
|
|
|
|
#
?
Apr 25, 2019 18:43
|
|
|
Pretty sure that wish was already granted anyways *monkey's paw itching to curl again*
|
|
#
?
Apr 25, 2019 21:27
|
|
|
German 3dcenter.org has Ice Lake/Comet Lake leaks, original source was a deleted posting in a taiwanese pc forum.  They speculate that Ice Lake will come with a noticeable IPC boost and/or much improved GPU performance as else it wouldn't be much faster than existing mobile chips. No 6 cores in sight and they're very far from reaching frequency/core parity with current 14nm desktop chips, confirming most suspected two years ago. https://translate.google.com/transl...-2019&sandbox=1 vvv yes, ninja edited that in before i saw this posting. eames fucked around with this message at 08:17 on Apr 27, 2019 |
|
#
?
Apr 27, 2019 08:08
|
|
|
It probably won't be much faster than existing chips. 14nm wasn't either, at first. That's part of why they aren't bothering to release Ice Lake on desktop, and why they weren't going to do Broadwell on desktop either, until people badgered them into socketing some HQ laptop chips. The first node gen after a shrink these days is much more interesting to mobile and server chips that care about efficiency.
|
|
#
?
Apr 27, 2019 08:14
|
|
|
Paul MaudDib posted:It probably won't be much faster than existing chips. 14nm wasn't either, at first. So Zen 3 should be more interesting than incoming Zen 2?
|
|
#
?
Apr 27, 2019 13:13
|
|
|
Der8auer said that while he can't comment on what has changed, he can tell that his R0 samples are overclocking to 5.1-5.4 "24/7 stable" and 5.4-5.7 GHz "benchmark stable" using watercooling. He has a proven track record so it's safe to expect that the new stepping will overclock better than what Intel sells now.
|
|
#
?
Apr 27, 2019 19:28
|
|
|
Sounds like soldered heatspreaders.
|
|
#
?
Apr 27, 2019 20:41
|
|
|
Khorne posted:I strongly disagree with this. Intel 7nm uses EUV. Part of the problem with their 10nm process is they went for overly ambitious dimensions, pitch in particular is supposedly around the theoretical limit for the process, without EUV. Lots of 10nm choices were ambitious. One thing I've always been curious about is the reliability side, which is understandably one of the least publicized things. Overclocking not withstanding, people don't really expect their PCs to fail because some key IC has failed (its usually power-related or a indirect fault caused by solder / mechanical issues). Qualcomm has made some Snapdragons available to embedded lifecycles, but generally speaking I feel like Qualcomm / mobile phones guys don't give a gently caress if latent failures start occurring in 3-4 years because the consumer tolerates it / its a rapid lifecycle compared to PCs where some companies will (arguably unfortunately) run stuff for many many years.
|
|
#
?
Apr 27, 2019 21:58
|
|
|
eames posted:Der8auer said that while he can't comment on what has changed, he can tell that his R0 samples are overclocking to 5.1-5.4 "24/7 stable" and 5.4-5.7 GHz "benchmark stable" using watercooling. You mean on Coffee Lake? The new stepping that Tom's was getting abnormally good results on? Link?
|
|
#
?
Apr 27, 2019 22:16
|
|
|
Otakufag posted:So Zen 3 should be more interesting than incoming Zen 2? Zen3 will clock higher as 7nm/5nm matures and AMD cleans up their 7nm layouts but it probably won't be as big an absolute jump as Zen2. GF 14nm is more like an Intel 22nm, so this jump is like AMD going from 22nm to 10nm in one go, plus their first major uarch revision. It's going to be a big one. Not sure what Zen3 will have in the pipeline but I can't realistically see it being as big a jump the second time. (on the other hand, if Intel is effectively standing still in the desktop market in the meantime, then Zen3 could be the one that finally puts Zen over the top of Coffee Lake, which is exciting too) They do pretty much need Zen2's clocks to hit very close to 5 GHz if they're going to keep up with or beat Intel. Even if they make a more optimistic 15% IPC bump, if they don't get to at least 4.7 GHz they're behind Intel. If their IPC bump is more like 10% then they have to get all the way to 5 GHz. They're pretty much not going to come out significantly ahead of Intel single-thread performance even in an ideal scenario, at least this time around. They will probably get close but still slot in behind. (on the other hand it will probably offer a lot more cores for the money, and if they're offering 50% more cores for the dollar, that 5% loss to Intel single-threaded doesn't matter very much. Right now they are losing by 20% just in clock speed, so being 5% behind in total would be a lot more palatable.) Of course the billion-dollar question is how the IO die changes latencies. Zen's not that far behind in IPC but sometimes those latencies just tank performance. If the IO die ends up increasing latencies a lot, gaming performance might not match the synthetic benchmarks they've shown to date. And at those extreme clockrates, cooling becomes problematic as well. In a 5 GHz scenario they are spending all the gains from their node shrink on increased performance, so power consumption stays the same. But the CCX size is cut in half, so thermal density doubles. It'd be like Intel making the jump from Sandy Bridge thermal density to Coffee Lake thermal density in one go. It's going to be very difficult to cool 100W+ on a chiplet that small. Paul MaudDib fucked around with this message at 22:57 on Apr 27, 2019 |
|
#
?
Apr 27, 2019 22:44
|
|
|
https://wccftech.com/amd-ryzen-3000-15-percent-ipc-uplift-4-5-ghz-x570-pcie-gen-4-40-lanes/ I find it very hard to believe AMD will do 40 lanes on anything other than the highest high-end SKU. There's no way they'd self-sabotage the low-end Threadripper market that badly. Then again, this is WCCF we're talking about - a website that never met a rumor it wouldn't treat as fact for clickbait's sake.
|
|
#
?
Apr 27, 2019 23:11
|
|
|
BIG HEADLINE posted:https://wccftech.com/amd-ryzen-3000-15-percent-ipc-uplift-4-5-ghz-x570-pcie-gen-4-40-lanes/ Never mind that, consumer motherboard makers will never do 2x16 and 1 x8 gen4 slots on a board.. 1 x16 I could see nice and close to the CPU but the rest, fuggedaboutit. Unless the AMD PHYs are amazeballs!
|
|
#
?
Apr 27, 2019 23:15
|
|
|
priznat posted:Never mind that, consumer motherboard makers will never do 2x16 and 1 x8 gen4 slots on a board.. 1 x16 I could see nice and close to the CPU but the rest, fuggedaboutit. Unless it's going to be 1x PCIe 4.0 x16, 2x PCIe 3.0 x8, and 2x PCIe 3.0 x4.
|
|
#
?
Apr 27, 2019 23:37
|
|
|
|
| # ? Jun 12, 2024 00:53 |
|
|
BIG HEADLINE posted:Unless it's going to be 1x PCIe 4.0 x16, 2x PCIe 3.0 x8, and 2x PCIe 3.0 x4. That'd be reasonable from a board standpoint.. If they do have all 40 lanes coming from the CPU they'd most likely all be gen4 but the bios could configure some to go max gen3 if the links are borderline..
|
|
#
?
Apr 28, 2019 00:00
|
|