So Intel released the full details for their 12+ core HEDT parts.
Interestingly it seems as though the leaks the past week or so were correct. Base clocks are generally under 3Ghz except for the 14 core which clocks higher than the 12. Assuming to keep the 12 at 140W TDP
TDPs have also been raised for the 14 core+ models to 165W. All turbo between 4.2 and 4.4. although this is two core max without AVX. Full load and how motherboard venders implement all core enhancement may change it.
Prices haven’t moved either.
12 core is launching August 28th and the rest September 25th…
Im looking forward to upgrade my trusty 5820k to a 5960x, because i dont see any need for x299 compared to x99, or to broadwell since haswell overclocks better.
Going to be interesting to see what that “165W” means in practice.
Having said that, dumping Haswell-E for Skylake-X, if you don’t “need” moar corez, is not something I would expect an informed consumer to do, given that we’re basically talking about the same IPC with unhealthily high(er) factory (over)clocks – hence the atrocious power draw – while the current lack of optimization for the new mesh arch (plus the ringbus’s advantages at <=10c) causes them to perform worse than the last gen in a number of ways.
Yeah well with SkylakeX you do get more chipset lanes.
And a dmi3.0 4x bus interface.
But yeah other then that not really that much interesting new stuff with SkylakeX.
Eh. They’re workstation chips at this point, and not that compelling ones at that. They only make sense for those doing AVX512 really. For gamers they make no sense. For workstation use, they don’t really make sense (because Epyc, Threadripper, or Xeon). Really unless AVX512 is quint essential to everything your doing, you might as well buy something else.
Yeah but the dmi bus is pci-e 3.0 4x on SkylakeX
The dmi bus on Haswell-e is only pci-e 2.0 4x.
So that is still an improvement of sorts, dual the bandwith between cpu and chipset.
IDK I wouldn’t say it is a complete fuck up. When it works god damn it stomps everything in its path but at the cost of insane heat and power consumption. Coupled that with the price, nerfed IO and performance in unoptimized workloads it becomes really hard to recommend.
Just saw apparently a leaked bench of a stock 16 core X299 part. Cinebench it pulls in around 3200. Little under at stock. So about 80 points more than Threadripper at stock. So yeah it is faster. Is it worth the extra $700? I don’t think so. Especially when I almost guarantee it is using more power.
Double the bandwidth yet still can be saturated completely by a good NVME SSD…
If you have something bad then replace it with something that is slightly better but still can’t perform it really doesn’t make much of a difference IMO. You’re at the same place essentially.
Yeah but here’s where I look at it from. If your paying $1700 for a cpu, you might as well go TR, Xeon, or Epyc. Anyone that wants one of these for anything but workstation use is out of their mind, considering the cost and considering how alternatives can do everything Skylake-X can do, cheaper, and depending on case, faster.
Alternatives to a $1700 Intel 7960XE or $2000 7980XE:
1 Epyc 7401, 24 cores @ 2.8Ghz all cores $1850
2 Epyc 7301’s, 32 Total cores @ 2.7Ghz all cores >$1600
I don’t know of any workload that would scale across a 16 core or 18 core part that wouldn’t also scale across 24 or 32 cores. I just don’t see a usecase outside of AVX512. Hell even if you go for a $1000 1950X and buy an extra $700 worth of GPU.
So what kind of ssd/raid data throughput speeds would you need to saturate x299’s chipset<>CPU connection, and what might that be relevant for? video editing? genomics / bioinformatics data set reads?
I don’t really understand this tbh. A lot of users aren’t going to use like 18 or 24 core CPU’s. A lot of desktop use cases don’t even use them.
