i7-6950X = $1750

[cecofuli]

As in the title..

Isn't too much?


LINK

[Ondra]

Well, what'cha gonna do? Buy an AMD?

[Juraj]

Still a rumour but I saw it at few places...

.. and had a good laugh :- D Intel going full retard. I never ever imagined that Intel will topple nVidia one day but they really did (nVidia actually being quite generous with the new 10xx range which is nothing else than Win-Win situation)

At that price level, there is almost zero reason not to go dual Xeon route now.

But maybe they understand their niche customers well... there are only handful of people who kept buying 980X/3960X/4960X/5960X/ and nope, it wasn't professionals. It's the ultra "I need to waste money away quad-980Ti, let's post our overclocking scores and Doom3 fps in upper hundreds" nerds. And if those were willing to pay anything and those were the prime customers, then the price is obvious result.

[agentdark45]

AMD's consumer Zen and server Opteron Zen chips really need to deliver! Intel cannot be allowed to continue price gouging the market this hard.

[Juraj]

At least it does deliver improvement :- ) 1800 points in Cinebench R15 is nice score, like really nice. I presume that's only using all-core turbo so you could potentially reach 20-25perc. more performance.

[cecofuli]

My hope is that Zen will not a new "Bulldozer"...
I remember well the hype behind the old AMD CPU and... we know the story...
Better to wait this autumn and see...
Right now, i7-6950X is unjustifiably expensive.
And, in Italy, it will be 2000 euro... I already know...  ($/Euro change + taxes)
2.000 euro ONLY for a CPU desktop.. Mauahaaaa XD

[Noah45]

$900 for 2 more cores than the i7 5960X? Gamers can have at it, not for me.

[Jann]

And this is the reason I'm starting to look at V3/V4 Xeon cpu for my upcoming X99 system. Even though this one will definitely be single CPU, I'm sure I can get way more price/performance by getting an Engineering Sample.
Now to find a balance between core count and frequency, not to loose too much of single-dual threaded applications. Anyone got their hands on any new V4 Xeons that work in regular boards with normal ddr4? :)

[agentdark45]

And this is the reason I'm starting to look at V3/V4 Xeon cpu for my upcoming X99 system. Even though this one will definitely be single CPU, I'm sure I can get way more price/performance by getting an Engineering Sample.
Now to find a balance between core count and frequency, not to loose too much of single-dual threaded applications. Anyone got their hands on any new V4 Xeons that work in regular boards with normal ddr4? :)

Managed to snag a 12 core Xeon v3 on ebay a year ago for around £550, still works beautifully! Runs around 3-3.2ghz when rendering which is quite alright in my books. It scores at just over 1800 in Cinebench.

I have spotted a few of the new crazy 22 core v4's going relatively cheap, but their stepping seems to indicate that they are very early versions and will probably have a few bugs. Worth a gamble though if the seller accepts refunds considering the retail price of those chips!

[Rhodesy]

Re the AMD Zen chip. Doesn't Corona really utilise Ebree in a big way and isn't that only Intel?

[Jann]

Managed to snag a 12 core Xeon v3 on ebay a year ago for around £550, still works beautifully! Runs around 3-3.2ghz when rendering which is quite alright in my books. It scores at just over 1800 in Cinebench.

Which model/stepping did you get? I'm looking at something with at least 1800 CB points, but high core clockspeed. That 3-3.2GHz seems good to me!

[FrostKiwi]

Re the AMD Zen chip. Doesn't Corona really utilise Ebree in a big way and isn't that only Intel?

No, embree was made by intel engineers, not exclusively for their plattform.
It utilizes SSE4.1 and/or AVX  and/or AVX2 and/or AVX512 instruction sets to optimize their code for speed. Ondra mentioned, that the Code checks for these instruction sets and automagically chooses the best way to run the code, with as many instruction set benefits as possible. AMD FX series supports all instruction sets up to AVX, thus it receives the performance benefit of these optimizations. It is unclear how much AVX 2 and the youngest AVX512 instruction sets improve performance, but it is probably very small. AMD Zen will support AVX 2 and AVX 512 aswell when they release.

Intel embree was also made (in part) to push their addin card - "Xeon Phi" which failed spectacularly, because you had to specifically write code for it to be fast, which made it unusable outside niche projects.

So, although it has the Intel label on it, since it knows the instruction set, it is going to be just as fast in terms of optimization techniques...
...Now, if AMD CPUs wouldnt suck so hard up till now that is...

[Ondra]

Embree actually runs pretty nice on AMD. Corona itself (not counting the embree part) has much bigger slowdown on AMD than Embree. We dont know why, we just develop it on i7s exclusively and choose algorithm versions that run fast on this hardware. CPU architecture is pretty universal, compared to GPU - you dont need to rewrite code for each vendor/every time there is a new card out.

[FrostKiwi]

Corona itself (not counting the embree part) has much bigger slowdown on AMD than Embree.

How did you come to this conclusion? Cinebench scores match up with Corona benchmark times with AMD and an i7, albeit error tolerance.
Source attached.
Only AMD cpu's before Bulldozer barely work normaly, due to lack of SSE 4.1 and AVX.

[Ondra]

I remember it from some previous testing - there was really bad performance of Corona on some AMD CPUs (maybe it was those old ones). So we tried disabling embree and trying out embree example renderer, and it actually worked better. But it is possible it is fixed. AMD is so rare in our segment (less than 10% of results in Corona Benchmark), that there is almost no point in optimizing for it :/

[FrostKiwi]

AMD is so rare in our segment (less than 10% of results in Corona Benchmark), that there is almost no point in optimizing for it :/

9,94%
Ohh, you are right.

1767 missing. Good thing Jitbit has me covered.
With an average AMD FX render time of 5:43, it is only 7 days and 15 minutes to artificially create a 50% Corona benchmark market share.

I'm off making money for my electricity bill.