Seems Bit-Tech broke the NDA and just dropped a review of the X299 based 10 core Skylake-E chip.
Spoiler: It's fast, it gets hot and it drinks the power.
Honestly I'm not impressed. Yeah it stomps the 1800X fairly handily in some tests (as it should it's a fricken 10 core with higher clocks and IPC) but comes awfully close in some others.
Does overclock well it seems but gets very hot in the process and the power consumption is wow.
The performance in context with price does not impress me
I will be going with threadripper I think, as a cinema4d user I am fairly certain amd's 16 c 32 thread monster will destroy that.
my expectations for threadripper / cinebench are around 160 cb's for single thread (same as normal ryzen really...) and 3k+ multithread (based on good scaling across the extra cores / threads, c4d scales excellently)
It's really cool to see competition again and I am hoping this is just the start of AMD's comeback into the cpu / gpu markets
Going to wait for an Anandtech or PcPer to do power testing, as 72 watts higher than the 6950x at stock doesn't seem quite right.
Also worth noting that they're not technically breaking an NDA since as they noted:
Please note that we have sourced multiple CPUs from multiple sources for the purpose of this review. No hardware was supplied by Intel.
It doesn't look like they signed one if that is the case. Kind of really shitty for them to put out a review early, undercutting all the guys that are going through the process the correct way.
I hope an Anandtech or PcPer does clock for clock power comparisons, considering that comparing 6950x to 5960x without equalizing for clock leaves you with a massive all-core turbo frequency delta of 700mhz
There are two things that bug me out about that measurement. One, they're using Prime95 which we know isn't really realistic. Two, I don't know how they're measuring it being that I don't know the site. Are they using a multimeter or are they using software readouts?
Goddamnt you got me all messed up, I was right the first time. They're showing a 72 watt delta between the stock 6950x and stock 7900x, both 10 core parts.
Only in some benchmarks though, which makes this very weird. Some seem to benefit from the larger L2 and smaller L3, and some seem to be cache starved without the large L3.
I punched everything into Excel, and my math is showing me that a base i9-7900X is ~21% faster than a base Ryzen 7 1800X on average. The overclocked i9-7900X is ~28% faster than the overclocked Ryzen 7 1800X on average.
So with the i9-7900X you get like 25% more performance, with 25% more cores, at an additional 100% cost compared to the Ryzen 7 1800X.
You also get 2x the memory capacity, 24 more cpu lanes, and 24 chipset lanes compared to Ryzen's max of 8, but all the chipset lanes are pcie 3.0 vs pcie 2.0 on Ryzen. Its not a fair comparison. Your comparing performance but disregarding the platform pros/cons. You have to compare to threadripper, where AMD will still more than likely win.
Intel are done as far as their current architecture is concerned. It cannot get faster IPC wise as far as I can see, it definitely cannot clock higher as I will simply melt, and adding cores is only letting keep level.
AMD have hit a clock wall but other than that have the temps easily in check, have the IPC and have the per core performance. Considering g this is the first generation of Ryzen and improvements are to follow I expect it to gain a little in the future.
So AMD are already level and putting it to Intel, and intel are tapped out performance wise.
This is a lost battle for intel, they are just shotgunning shit into the market in a desperate attempt to stay relevant and grab headlines.
What Intel do have is cash reserves. That is their only saving grace, use it to develop something new because the iCore line is finished it looks like.
So but here's what I don't get. Why is everyone freaking out over the 8 core being nerfed, when the combined total with the chipset lanes is still equal to the pcie total on x99 (except now all lanes are 3.0)? The only thing that was limited to using cpu lanes only was sli, but now that sli is capped to two cards, which can totally be done on 24 lanes no issue, who cares? Your still getting 48 lanes total to use just like on x99.
Well I mean there are disadvantages of going through the chip set rather than direct connection but tbh I think it is more principle. There is no reason why the lower end chips couldn't have the same number of lanes. They are all based on the same die.
So I think people are just angry that Intel is being stingy with it.
The RAID shit, some chips having slower memory capabilities... Just a bunch of stuff about it that rubs people the wrong way.
I do have to second that the way they're handling raiding m.2 devices is total bullshit DRM that has no place on the platform.
Let's be honest, the "stock" memory speed really doesn't mean anything. I'm pretty sure that 6 core part with a 2400mhz default instead of 2666mhz will have no issue clocking far past 3000mhz, as all intel DDR4 memory controllers have past the infancy of X99.
I bet they're doing it for yields reasons honestly. IMO I don't think that the community response has been reasonable. The total number of lanes combining pch and cpu lanes is either staying the same, or increasing, at every core count.
True but why even bother then? It almost seems like they are doing it to catch out the novice who may not know reading a spec page. It just seems deceitful to me
PCIE connectivity shouldn't be effected. Even if it has only 8 working cores the PCIE controller should still have all lanes available. Broadwell-E had a 10 core and an 8 core both on the same die both with full PCIE connectivity.
It is an artificial nerf to create segmentation and nothing more.
"Ohh this CPU supports slower RAM I want to have the fastest! Let me buy this one that is more money!"
Like I said not a huge deal really and I don't see that happening but why the hell would you even do it in the first place? Seems like a dick move because they need to take something away or segment it some how.
The 6, 8, 10 and 12 core parts all share the same die, with the same PCIE controller, yes.
