tamz_msc
Diamond Member
- Jan 5, 2017
- 3,821
- 3,642
- 136
Intel clearly used solder as recently as Broadwell-E, and obviously it isn't any bigger than Skylake-X. So clearly it isn't about the size of the die.
Of course it was all about the money. Intel is run by bean counters lately.
All those >10 year old high end at their time, soldered chips (high end P4/PDs, E6xxx, E8xxx, Q9xxx, i7 9xx/8xx) are running just fine out there without a single problem. skt1156 Nehalem in particular, Lynnfield, is 296mm², those certainly fit the bill for "big chip, too hard to solder, *insert BS excuse here*" and are running fine... and will continue to do so until all those motherboards finally break down.
Oh well. I guess if you want to push these to the limit after spending a crapton of money you'll have to delid to reap the 20-30°C savings and those extra few MHz out of it. You're forced to do it! It's a lot of cores that need quick heat transfer here, a single one becoming a hotspot will shoot any decent overclocking attempt down. As der8auer says in the video, it's also a risky thing to do because of the glue/sealant being so close to the SMD caps on package, you could rip some off if glue got over them during IHS placement.
What a shame. At least the guys who make delid tools are happy going forward, you can't do those with a razor blade.
The sensible thing to do here would be to use TIM on the non K parts and solder on K/HEDT parts that you know will be used out of spec at some point of their useful life...
Last edited:
Stability isn't the issue. TIM is also less of an issue in Xeons. No OC and they can be cooled by server fans (eg loud something you never want in your PC). Also compared to consumer PC(workstations servers are usually updated much more often hence the huge xeon flood on ebay. (if you need a multicore cpu for transcoding or your personal business usually cheapest to buy up xeons from ebay.). meaning longevity isn't a huge concern with xeons.
You can. But I could easily have the same thing at 20-30 lower CPU temps if it used solder. Which means less power use and hence even less heat. Also for the 7700k it's somewhat ok. It's a consumer chip. HEDT on the other hand with higher platform cost and generally being the "luxury" platform, it's annoying as it obviously is a cost-reducing measure at cost of us consumers. If a $10k dollar toyota uses all cheap plastic you get what you pay for but if the $70k lexus uses the same shitty plastic everywhere I would kind of be unimpressed and not buy it.
Intel's real customers are laptop/tablet manufacturers and datacenter operators who could not possibly care less about this drama. As for why they do not solder their "extreme" SKUs, where do you think those come from? Desktop chips are just overclocked laptop chips, and HEDT chips are just overclocked server chips. They come from the same wafers, the same dies, and in fact, likely are the exact same batch of packaged CPUs until the final SKU determination is made. There is no way Intel will create a separate assembly line for something with non-existent sales (relative to total revenue).
None of this makes the lack of solder on HEDT any better, but Intel is doing this to save far more than $1. In fact, if their old soldering process destroyed some percentage of dies during assembly, the cost of soldering would be proportional to the final retail price. In that case, going to TIM would always make sense for Intel, regardless of how "luxury" the product is.
Last edited:
https://rog.asus.com/forum/showthread.php?93632-Late-June&p=653445#post653445
Raja@ASUS
The 18 core CPUs are not scheduled until next year. Won't have them for a while. Either way, unless you're using the rig for rendering or encoding to make a living, no need.
Raja@ASUS
The 18 core CPUs are not scheduled until next year. Won't have them for a while. Either way, unless you're using the rig for rendering or encoding to make a living, no need.
Core i9-7980XE is basically a modified Xeon Gold 6150, which is already shipping to select customers for quite some time and will be broadly available real soon - no reason to expect it only in 2018 - and doesn't match what I heard either. As you for your ''Skylake-X has no AVX-512'' thread (based on dubious SiSoftware engineer sample scores), I can now confirm AVX-512 is fully enabled on Skylake-X. The chips just scale back frequency in AVX-512 mode.
Last edited:
- Jun 10, 2004
- 14,330
- 4,918
- 136
Looks like I'll have my Threadripper system built before my Skylake-X system then...
what do you mean with fully enabled, two 512-bit FMA per core ? i.e. twice throughput with AVX-512 code vs. AVX2 code ? and what is your source for this ?
Atari2600
Golden Member
- Nov 22, 2016
- 1,409
- 1,655
- 136
I wonder how the shills and fanbois will spin this now.
Trolling and Threadcrapping are not allowed
Markfw
Anandtech Moderator
Last edited by a moderator:
AVX-512 is fully enabled on Skylake-X？
Skylake-X 6C AVX512? http://ranker.sisoftware.net/show_r...d5e3daeadeeadbfd8fb282a4c1a499a98ffcc1f9&l=en
Multi-Media Integer 599.30Mpix/s
Multi-Media Long-int 189.23Mpix/s
Multi-Media Quad-int 2203kpix/s
Multi-Media Single-float 546.11Mpix/s
Multi-Media Double-float 326.96Mpix/s
Multi-Media Quad-float 10157kpix/s
Skylake 4C AVX2 http://ranker.sisoftware.net/show_r...d5e3d5e2d1e7d2f486bb8badc8ad90a086f5c8f0&l=en
Multi-Media Integer 525.14Mpix/s
Multi-Media Long-int 190.14Mpix/s
Multi-Media Quad-int 2115kpix/s
Multi-Media Single-float 466.73Mpix/s
Multi-Media Double-float 272.81Mpix/s
Multi-Media Quad-float 10500kpix/s
Skylake-SP 56C AVX-512 Xeon Platinum 8180： http://ranker.sisoftware.net/show_r...d5e3dbe3d1e0d6f082bf8fa9cca994a482f1ccfc&l=en
Multi-Media Integer 6437.86Mpix/s
Multi-Media Long-int 2299.06Mpix/s
Multi-Media Quad-int 22385kpix/s
Multi-Media Single-float 7268.05Mpix/s
Multi-Media Double-float 4593.02Mpix/s
Multi-Media Quad-float 105285kpix/s
Broadwell-EP 44C AVX2 Xeon E5-2696 v4 ： http://ranker.sisoftware.net/show_r...d5e3d4e6d2e6d3f587ba8aacc9ac91a187f4c9f9&l=en
Multi-Media Integer 3024.58Mpix/s
Multi-Media Long-int 995.63Mpix/s
Multi-Media Quad-int 15626kpix/s
Multi-Media Single-float 3104.20Mpix/s
Multi-Media Double-float 1746.95Mpix/s
Multi-Media Quad-float 61812kpix/s
Underlined: that's no news. I'd be surprised if it were for people on this forum.
Bolded: Then explain why did Intel have both assembly lines for decades in the past, even when accounting for destroyed dies during the soldering process. Both for small and big dies.
You could take a low end chip on any generation, delid it and you'd see TIM. Celeron E1xxx, Core 2 E2xxx, E4xxx, E5xxx, E7xxx for example. E6xxx, E8xxx, Q9xxx parts were soldered. Same for old high end Pentium 4/Ds. Low end Sandy Bridge parts vs K parts... etc.
Hell, have a look at a cheap i3 530, with all its dual package, 32nm CPU + 45nm GPU+MC nonsense
Solder on the CPU, a thermal pad on the GPU/MC. Makes sense, those little things could be clocked to the moon, and so could Sandy a generation later.
A quick google search will return pictures for any of those TIM'd parts as well.
The bean counting and nonsense at the higher end of the spectrum started with Ivy Bridge and has now moved all the way over to HEDT parts. Unless there's some external issue Intel can't control that it doesn't make sense anymore for them to keep soldering chips, there's no way to defend this. All in all it's a downgrade for the end user, be it average joe or server/big guys, and no % of those cost savings were ever passed down to the consumer. Ivy costed the same as Sandy while using TIM. Haswell then drove prices up to $250 and $350 for the i5k and i7k parts, still using TIM, and it's stayed that way ever since in part thanks to AMD's incompetence during that time. Again, 296mm² Lynnfield is soldered and this qualifies as a HEDT/server big die and so do all of Sandy, Ivy, Haswell and Broadwell HEDT/server versions, invalidating der8auer's lenghy explanation on why Intel stopped soldering stuff. It's simple, it's bean counting and it must make a lot of sense for them at this point. A shame.
Anyway, and steering out of this discussion, people have become used to delidding chips since Ivy, and that's a non issue going forward for the mainstream socket, you can do it with a razor blade and some patience. Now for HEDT, If I'd ever were to buy into one of those systems I'd gladly buy the tool and delid for the 20-30°C temperature drop as the reward is worth the risk in my eyes, but sure I'd be swearing while doing it.
Gotta sell all those expensive dies to the server guys first where they can charge what they actually want for those, not to leave money on the table just to one up Threadripper with 2 more cores and keep the throne.
Atari2600
Golden Member
- Nov 22, 2016
- 1,409
- 1,655
- 136
+1
Basically for the foreseeable its a bullet point on a powerpoint - existing solely so someone blue can stand up in front of an audience, dismiss any UHEDT lead that ThreadRipper gives AMD (if it does at all) and claim the best thing since sliced bread is coming down the track.
Does it (the 18C variant) have the full Intel EmergencyEdition moniker?
Manufacturing "improvement" is a continuous process. Just because Intel hadn't moved to TIM on all product lines yet, doesn't mean it can't happen. Perhaps Intel was finally confident enough that TIM would meet the thermal demands of server chips. Perhaps the engineer in charge of packaging finally had time to investigate and qualify this change. Perhaps the volume of server chips increased and justified a move away from solder (labor). Perhaps they decided to delay SKU determination to improve supply chain agility. Perhaps they did it just to spite you.
EDIT: Sad that the 12-18C SKUs are vaporware. AMD looks set to introduce "ThreadRipper" this year and destroy Intel yet again on the desktop.
Last edited:
LTC8K6
Lifer
- Mar 10, 2004
- 28,520
- 1,575
- 126
Does that version of Sandra , 22.20, fully support SL-X processors and AVX-512?
http://www.sisoftware.eu/2017/05/23/sandra-2016-sp4-and-2017-rtmc/
http://www.sisoftware.eu/download-lite/
Latest version is 24.22 RTMc, so I'm guessing no.
http://www.sisoftware.eu/download-lite/
Latest version is 24.22 RTMc, so I'm guessing no.
According to DigiTimes 12C Skylake-X (Core i9-7920X) launches in August, and 14C/16C/18C models are based on the same MCC die. From Intel's fact sheet: ''Scalability with 10, 8, 6, and 4 cores (18, 16, 14, and 12 cores coming soon!)''
Last edited:
AdamK47
Lifer
- Oct 9, 1999
- 15,230
- 2,849
- 126
That August release date was speculated in early April. Intel has not officially said anything about it. It could be August or it could be next year. Nobody really knows for sure. That's the frustrating part about all this. I need exact dates!
Maybe I'll just go ahead and build a 16 core Ryzen Threadripper system. That's getting more and more tempting by the day.
Maybe I'll just go ahead and build a 16 core Ryzen Threadripper system. That's getting more and more tempting by the day.
Precisely, for small and big dies. They never ran 2 different lines for the same die, and it does not make sense for them to do it either, since apparently SKL Xeons are fine with TIM.
More likely, we're going to see a market opportunity for Silicon Lottery Special Delidding Warranty. Might be a better seller with $2000 SKUs floating around.Maybe the 18c Extreme Edition will be soldered.
It does not surprise me that Intel is trimming costs as the pricing has come down. Many of us speculated Intel would not be budging on price with this current round of chips. I am happy to be wrong. They have reacted with pricing and in a big way. Last time around an 8 core HEDT cost you $1,000. This time around it costs you $600. Is that worth giving up solder? Probably to some...
Something has to give with public companies and profits. Imho the Ryzan 8 core will be bested by the Skylake 8 core. So they found a way to lower the price and still stay ahead of the competition.
Something has to give with public companies and profits. Imho the Ryzan 8 core will be bested by the Skylake 8 core. So they found a way to lower the price and still stay ahead of the competition.
TRENDING THREADS
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 10K
-
Discussion Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)- Started by Vattila
- Replies: 13K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 7K
Facebook
Twitter