Return to

Intel Xeon W-3175X


Holycrap… that PcPer was awful, confusing and leaves a lot more to be desired… sorta like this processor.

OK… i kid, its not awful, but is confusing and does leave a lot more to be desired… like market positioning, 5ghz stock, value, use case.


What I do not even find funny anymore is the “255W TDP”, I know is it not power draw. However:

The thing is constantly beaten by chips specialized in single-core and multi-core workloads, not a lot of people need a CPU that can do both (poorly).
Honestly, the benchmarks are benchmarks in how poorly desktopsoftware is writen.

And then there is this:

1.5ms frame time (avg.) for the W-3175X, 1.4ms frame time (avg.) for the 9900K. Without an O-scope and damn fast sensor, you will never ever notice that difference.


That second one lists the W3175x twice, one as “Intel spec” and the other with no note what is changed. Does the article say? I know it is unlocked but did they actually try OC it?


It was run faster than stock clocks yes. I believe the ASUS board they used ran faster than official intel spec out of the box.


They enabled the ASUS MultiCore Enhancement. That disables power limits and turbo clock limits and turbo duration limits. So effectively it lets the CPU run at the max. turbo speed, even if all cores are loaded.
Effectively it is a moderate OC, usually very stable.


So, stock it’s not even faster than TR2 because it’s not even close to the 5GHz Intel was showing on stage, it is way more money, the boards (both of them… rofl) are dumb and will cost an arm and a leg, the whole thing is a hog when it comes to power consumption and now Asus is taking the piss with MCE again.

Yup, that looks like the total train wreck I expected it to be. :+1:


Right now the only reason I can see for this CPU is the massive AVX and AVX-512 performance. If someone has an application that can utilize AVX-512 (MKL, X265, etc. or they are running code they develop for themselves) then SKL-X blows current AMD offering out the window.
This will of course get remedied with Zen 2, coming in 2019, with 2x core count and 2x floating point width. (native 256-bit wide, current parts are 128-wide only)


Would be nice to see AVX-accelerators in Intel GPU offerings. Would make this CPU more pointless than it allready is, but would be nice for certain workloads.


I would wager that at avx512 an existing i9 will maybe even be faster. Because avx512 gets HOT and this thing is thermally taking the piss already on regular workloads.


That Cinebench 4 benchmark looks like the real train wreck to me.

W-3175X > i9 9980XE > TR 2950X > i9 9900K > W-3175X (Intel Specced) > TR 2990WX > R7 2700X > EPYC 7601.

WTF is this. That would be like if a benchmark shows GTX 1060 > RTX 2080 > RX 560 > RTX 2070 > Radeon 7 > RTX 2080 Ti > RX 570, it makes no sense


Looks fine to me.

Blender on Anandtech is fine too:

Edit: Oh TweakTown, you cheeky fuckers (ordered by a D20):


Will be interesting if the CPU or MB do sell retail.

And how many systems sell from System Integrators.

The windows kernel on AMD is an issue. I wonder how the TR 3xxx series with an IO die will do. Lisa talked about EPYC’s IO chip will balance work loads across cores. Perhaps that is the fix needed ?


i think in the next 5yrs we will see a more heterogeneous architecture from both AMD and Intel, (Nvidia with an ARM implementation) it’s the next step in Computing IMO


Intel past 6000 series is a meme. Don’t waste your money.


Actually that is ordered by cpu cost.

Stupid, but tweaktown, so to be expected


I have my doubts. They have already tried that with the xeon phi accelerator cards (ex-larrabee), in fact AVX-512 came from those. (kinda, not entirely)
And that project failed in pretty much every way conceivable. They did not get a usable GPU, and the accelerator cards never really gained any significant traction. I wager this is partly due to Intel’s reluctance to put a load of SW work into getting the OpenCL interface running.
Seriously, pretty much the only way to use those stupid things is to buy the Intel compilers and have them generate code for the cards. And Intel has axed all support for those accelerator cards at this point, their latest compilers have absolutely zero support for them.
So anyone who had trusted Intel, and bought into the xeon phi accelerator card promise got burned, stuck on like ubuntu 16.04 LTS with an older Intel compiler that still has support for the cards. Same goes for developers who have gone to the substantial effort of porting their code to the cards, pretty much a dead branch at this point.


If your code scales well with threads the 28 cores will give a good boost, even if the clocks are lower. Same for AVX-512.
But of course if you adhere to Intel spec, then speedup will be limited. This CPU really needs the power limit lifted (a la MCE) to shine. Of course you need some hardcore liquid cooling that can handle 400+ Watts.