|
KillHour posted:Maybe if you're just looking at the server hardware, but once you factor in the extra switches, PDUs, software licenses, storage, hardware failures and management overhead? Assuming your workload parallelizes well, a Epyc 7551 32c@2.55ghz is going to run you ~$2250/socket while the most comparable Xeon is the Platinum 8168 24c@3.4ghz for $5,928/socket. So yeah, 40U of 7551's are going to run you $90k in processor costs while the Xeon would be $240k. Even with the potential cost overhead you're talking about this isn't even close. Maybe if you're doing blades or FX2 with high CPU density the embedded switches and shared PSUs will make it break in Intel's favor but if you're just looking for conventional U pancake servers Intel's offerings are poo poo. And if AMD can get traction I would expect to start seeing the OEMs packaging Epyc platforms in to blade or other modular chassis configs. BangersInMyKnickers fucked around with this message at 16:34 on May 28, 2019 |
#
?
May 28, 2019 16:30
|
|
|
|
| # ? May 26, 2024 20:47 |
|
|
BangersInMyKnickers posted:Assuming your workload parallelizes well, a Epyc 7551 32c@2.55ghz is going to run you ~$2250/socket while the most comparable Xeon is the Platinum 8168 24c@3.4ghz for $5,928/socket. So yeah, 40U of 7551's are going to run you $90k in processor costs while the Xeon would be $240k. Even with the potential cost overhead you're talking about this isn't even close. Maybe if you're doing blades or FX2 with high CPU density the embedded switches and shared PSUs will make it break in Intel's favor but if you're just looking for conventional U pancake servers Intel's offerings are poo poo. And if AMD can get traction I would expect to start seeing the OEMs packaging Epyc platforms in to blade or other modular chassis configs. List pricing is not relevant at all for high end Xeons. There's enormous discounting that happens at the vendor level, supported by direct deals the vendors do with Intel. I'd also say that in general the Platinums are for 8-socket configs, because you can run quad socket with Golds without paying the 8S support price premium. I still agree that Epyc is an unbelievable value and if you can run on 2S boxes, do it and do it on Epyc.
|
|
#
?
May 28, 2019 16:41
|
|
|
Twerk from Home posted:List pricing is not relevant at all for high end Xeons. There's enormous discounting that happens at the vendor level, supported by direct deals the vendors do with Intel. I'd also say that in general the Platinums are for 8-socket configs, because you can run quad socket with Golds without paying the 8S support price premium. How much are discount are you getting on those Platinum chips? Best we could squeeze Dell on was ~20% and it was still a terrible proposition.
|
|
#
?
May 28, 2019 16:52
|
|
|
BangersInMyKnickers posted:How much are discount are you getting on those Platinum chips? Best we could squeeze Dell on was ~20% and it was still a terrible proposition. The only quotes I've dealt with for >$50k per unit servers recently are a university lab working with a Supermicro reseller, but there was a good bit more discounting than 20%. They were still not a better value than Epycs, and I actually steered them to Epyc 7371 because for their HPC faster cores are preferable to more cores, and that let them more cost-effectively hit their ratio of RAM per thread. Intel really sticks it to you if you need the -M chips with more than 768GB of RAM per socket support.
|
|
#
?
May 28, 2019 16:56
|
|
|
Ice Lake products are actually launching after 5 years, it is a surreal day: https://www.anandtech.com/show/14426/when-ice-matters-dell-announces-xps-13-2in1-with-ice-lakeu This is also the first appearance of Sunny Cove. Of course, the question is now "when will they actually ship?"
|
|
#
?
May 28, 2019 17:26
|
|
|
Interesting. Everyone else is saying its a problem unless you RAID and BZ and Steve got that from several mobo vendors too. I guess it depends on what that ASUS rep meant by hot?
|
|
#
?
May 28, 2019 17:32
|
|
|
drat, AMD dropped a bomb at Computex.
|
|
#
?
May 28, 2019 17:34
|
|
|
Intel is claiming 18% IPC gain from Ice Lake which is pretty drat good if you forget about how long it's been delayed, clocks max out at 4.1GHz for now though so you can see why there's no desktop parts yet.
|
|
#
?
May 28, 2019 17:55
|
|
|
It's a damned shame that we will never get the long form article/book detailing exactly how 10mm went so wrong that it deserves.
|
|
#
?
May 28, 2019 18:08
|
|
|
DonkeyHotay posted:It's a damned shame that we will never get the long form article/book detailing exactly how 10mm went so wrong that it deserves. I'd be interested in this too. I've spent far too long wondering how I consistently manage to lose every 10mm socket I own. Something must be done!
|
|
#
?
May 28, 2019 18:14
|
|
|
DonkeyHotay posted:It's a damned shame that we will never get the long form article/book detailing exactly how 10mm went so wrong that it deserves. yeah, i'd really really like to read this too
|
|
#
?
May 28, 2019 18:29
|
|
|
KillHour posted:I'd be interested in this too. I've spent far too long wondering how I consistently manage to lose every 10mm socket I own. Something must be done! Whoops
|
|
#
?
May 28, 2019 18:51
|
|
|
DonkeyHotay posted:It's a damned shame that we will never get the long form article/book detailing exactly how 10mm went so wrong that it deserves. Yeah, CPUs the size of football fields turned out to be less than ideal, who knew!
|
|
#
?
May 28, 2019 18:52
|
|
|
Twerk from Home posted:The only quotes I've dealt with for >$50k per unit servers recently are a university lab working with a Supermicro reseller, but there was a good bit more discounting than 20%. They were still not a better value than Epycs, and I actually steered them to Epyc 7371 because for their HPC faster cores are preferable to more cores, and that let them more cost-effectively hit their ratio of RAM per thread. If you’re paying for 128gb dimms or optane you can find a few thousand extra bucks per socket. But yeah if you’re not getting better than 20% off on 8 series you’re not buying enough or your vendor isn’t getting it done.
|
|
#
?
May 28, 2019 19:13
|
|
|
PCjr sidecar posted:If you’re paying for 128gb dimms or optane you can find a few thousand extra bucks per socket. This attitude only works when AMD isn't competitive, and when you're buying a decent number of machines then a few thousand extra bucks per socket starts to really add up. Epyc supporting 16 DIMMs per socket also lets you get further with 64GB RDIMMs or LDRIMMs, you have to reach for the 128GB DIMMs earlier with Intel.
|
|
#
?
May 28, 2019 19:20
|
|
|
https://www.youtube.com/watch?v=YeyAG-CMKCs This little hardware demo from Intel looks interesting, and ASUS is already in on something like it, since they just announced a remarkably-similar (but missing the second hinge) ASUS Zenbook Pro Duo. I do think that the Intel demo would be more usable, though, and I gotta say, as I have moved over to ultrawide monitors, I've started stacking dual monitors vertically instead of horizontally too, so I think it's a good move.
|
|
#
?
May 29, 2019 06:54
|
|
|
The first table in this article indicates to me that there's no reason to get RAM faster than DDR4 at 2666MHz if you have an Intel i7-9700k or an i9-9900k, because that's the most that they'll utilize. Is that true? edit: vvv That's good to know! Thank you. surf rock fucked around with this message at 14:20 on Jun 5, 2019 |
|
#
?
Jun 5, 2019 13:33
|
|
|
Absolutely not. That's the fastest ram officially supported, but Intel platforms can go far, far beyond that with absolutely no stability compromises or anything. What ram speed you should get depends heavily on exactly what you intend to do with the CPU and how much you're willing to gamble on future trends mirroring past ones, but for some applications (open world games most notably), there is excellent scaling to the fastest ram you can get and presumably beyond. Generally speaking, the fastest you can go without paying a big price premium is worth it. The current sweet spot price wise tends to be around 3600, with going beyond that driving the price up quite a bit but possibly still providing worthwhile performance increases for the cost, depending on your specific goals.
|
|
#
?
Jun 5, 2019 13:41
|
|
|
Yeah, Intel usually caps the official support for different RAM standards in their memory controllers at whatever the official top JEDEC speed is for that standard. E.g. DDR stopped at 400MHz, DDR2 at 800, DDR3 at 1600. Sometimes even lower - if you look up a 2500K on ark.intel.com, the highest "officially supported" bin is 1333MHz. It can run at 2133+ all day though, at least as long as the RAM can handle it. For realistic maximum speeds, you'll be better off checking the motherboard specs than the processor specs. Even though the processor contains the memory controller these days, maximum stable speeds are often defined by the quality and length of traces on the motherboard.
|
|
#
?
Jun 5, 2019 15:49
|
|
|
 "Reddit engineer posted:Found this chip in my bosses office. I'd pay good money for one of these.
|
|
#
?
Jun 13, 2019 14:51
|
|
|
Manufacturing plant guy "I was just doing what I was told" That's pretty loving awesome. Hopefully some made it into the hands of collectors.
|
|
#
?
Jun 14, 2019 05:52
|
|
|
Vanagoon posted:Manufacturing plant guy "I was just doing what I was told" The customer literally asked for that exact thing. The hell do you want me to do? Tell the customer "No" to his face when he asks for something?
|
|
#
?
Jun 14, 2019 06:42
|
|
|
So I thought I'd finally ran out of Z77 boards, and bought a new AMD 2600 setup. Well, I found one that still worked that I'd written off, probably because my modular PSU had a cabling issue that was preventing it from POSTing. Back up at 4.5Ghz, in a lovely low-end Z77 board with like 5 phases. Still slays the stock 2600 in single core and 2D perf, but it needs 130W to do it so I guess I'm going to put it on marketplace with a spare GTX770 and send it on its way. Was tempted to give it a viking's funeral but the drat thing wouldn't post above 4.8 and I feel like running it OC without the HSF is just a lovely way to die. loving Soldier On, Sandy Bridge. 
|
|
#
?
Jun 14, 2019 06:43
|
|
|
Samsung to fabricate Intel CPUs: https://www.techpowerup.com/256613/intel-turns-to-samsung-in-order-to-resolve-cpu-shortage-on-the-14-nm-process I would have expected them to outsource chipset fabrication but the article explicitly mentions 2021 Rocket Lake CPUs.
|
|
#
?
Jun 18, 2019 10:03
|
|
|
I haven't been able to keep up with all of Intel's lakes. Is Rocket Lake supposed to be some kind of low cost chip only for Pentiums, or is Intel seriously going to rely on 14 nm in 2021? By then I reckon there's a decent chance AMD will launch Zen4 on the 5 nm node TSMC is launching next year.
|
|
#
?
Jun 18, 2019 11:08
|
|
|
eames posted:Samsung to fabricate Intel CPUs: TechPowerUp posted:Samsung will begin mass production of 14 nm Intel CPUs in the fourth quarter of 2020 That's a pretty long in the tooth process node there, I guess in the intervening year it won't be worth it to port over to Samsung's newer nodes.
|
|
#
?
Jun 18, 2019 13:49
|
|
|
Drakhoran posted:is Intel seriously going to rely on 14 nm in 2021? Their 7nm is supposed to come on line in 2021 "at the earliest" according to Intel as of early 2019. If they're expecting delays and/or difficulty getting their 7nm process to ramp production going to Samsung would make some sense too for that time frame.
|
|
#
?
Jun 18, 2019 15:49
|
|
|
PC LOAD LETTER posted:If Intel's 10nm stays a shitshow throughout its expected life (which going by this announcement it seems to be a given at this point) than there really isn't anything else Intel can do to improve chip production in the mean time since all their available fabs are already busy. The alternative is to let AMD get all that market share by default. yeah i think intel has given up their 10nm and saying hello to their own 7nm.
|
|
#
?
Jun 19, 2019 15:05
|
|
|
Toms is reporting that the rumors were wrong. Samsung wont be making Intel consumer CPUs, but chipsets and PHYs and other stuff that Intel has also outsourced to TSMC on occasion. https://www.tomshardware.com/news/intel-samsung-14nm-rocket-lake-cpu-chipset,39678.html
|
|
#
?
Jun 19, 2019 16:47
|
|
|
I can see them shift all non core silicon (cpu/Gpu) to TSMC or Samsung. 10nm and Brian krzanich hosed the company in more ways than one. Did anyone publically state the reason for the delays? Rumor is that the transition to cobalt hosed them but I don't have a good source for that.
|
|
#
?
Jun 19, 2019 21:17
|
|
|
Malcolm XML posted:I can see them shift all non core silicon (cpu/Gpu) to TSMC or Samsung. 10nm and Brian krzanich hosed the company in more ways than one. Curious as to what Krzanich did to Intel that hosed them. I've never heard this side of it!
|
|
#
?
Jun 19, 2019 22:47
|
|
|
wargames posted:yeah i think intel has given up their 10nm and saying hello to their own 7nm.
|
|
#
?
Jun 19, 2019 23:07
|
|
|
iospace posted:Curious as to what Krzanich did to Intel that hosed them. I've never heard this side of it! I don't know the full story but I know that he tried to decrease R&D spending which is probably not a wise idea when you're an industry in the midst of this. 
|
|
#
?
Jun 20, 2019 00:03
|
|
|
The journalist I know who best documented the failure of Krzanich and TMG was Ashraf Eassa and he has subsequently been hired by Intel  . Charlie Demerjian at Semi Accurate has some info on this as well.
|
|
#
?
Jun 20, 2019 00:07
|
|
|
I haven't followed the news on that, does Intel have a CEO again?
|
|
#
?
Jun 20, 2019 01:40
|
|
|
MaxxBot posted:Charlie Demerjian at Semi Accurate has some info on this as well. oh yeah, unvarnished truth, for sure
|
|
#
?
Jun 20, 2019 01:57
|
|
|
Yes, Bob Swan was named CEO early this year after having been interim CEO since Krzanich left. Swan has been CFO of a number of tech companies prior to joining Intel including Northrop Grumman, HP Enterprise Services and Ebay. He joined Intel, as CFO, less than 2 years before taking over as interim CEO. I'm sure Wall Street has full confidence in his ability to account his way out of any trouble Intel may encounter.
|
|
#
?
Jun 20, 2019 02:02
|
|
|
JawnV6 posted:oh yeah, unvarnished truth, for sure I mean if you think he's a bad source feel free to explain why.
|
|
#
?
Jun 20, 2019 02:22
|
|
|
MaxxBot posted:I mean if you think he's a bad source feel free to explain why. JawnV6 posted:If we're suddenly giving credence to analysts, reminder that Apple laptops are switching to ARM by the end of the year.
|
|
#
?
Jun 20, 2019 02:27
|
|
|
|
| # ? May 26, 2024 20:47 |
|
|
Still 1000x more accurate than official Intel slides
|
|
#
?
Jun 20, 2019 02:38
|
|