28c Intel W-3175X Reviews; $1700 motherboard?

[Taf the Ghost]

Gamer's Nexus is currently doing an OC session.

[Taf the Ghost]

1 minute ago, huilun02 said:

Seems like Intel saw the Radeon VII launch and said "Hold my beer" 

You missed the industrial-strength chiller and 5 Ghz all-core part from Computex last year.

[williamcll]

How is the chip on stock compared to the TR 2990WX?

[Taf the Ghost]

16 minutes ago, williamcll said:

How is the chip on stock compared to the TR 2990WX?

In Windows, generally better by a little bit, or a lot in the spots that the Windows Scheduler is causing problems. Not seen any Linux benchmarks yet, but, over there, it's likely to be fairly similar in most workloads and better in the AVX2 workloads.

 

It's better, but it's a 3000USD CPU and a 2000USD Motherboard, to start with.

[RejZoR]

2990WX seems like a better deal lol.

[CarlBar]

On 1/30/2019 at 11:49 PM, yian88 said:

intel is fukkd with these products, their design model increases prices non linearly to cores, it doesnt scale well too expensive

until they get their new chiplet/chip stacking designs out AMD will  be way ahead

 

I did some maths on the yield intel was getting with a little online tool and depending on what the wafer cost is for intel on the 14+++ node they might not actually be making any money on these things. They probably are but their margins definitely aren't good. A die this big just doesn't yield well alas.

[Taf the Ghost]

6 minutes ago, CarlBar said:

 

I did some maths on the yield intel was getting with a little online tool and depending on what the wafer cost is for intel on the 14+++ node they might not actually be making any money on these things. They probably are but their margins definitely aren't good. A die this big just doesn't yield well alas.

Some of the recent discussions around 10nm has pointed to 14nm never being that great of a yielding node. The ++++++ versions actually are less dense than the original Broadwell processors, though a lot of that seemed to be due to issues with yielding GPUs. There's a reason Intel wants 10k USD for these CPUs in their Xeon form.

[CarlBar]

4 minutes ago, Taf the Ghost said:

Some of the recent discussions around 10nm has pointed to 14nm never being that great of a yielding node. The ++++++ versions actually are less dense than the original Broadwell processors, though a lot of that seemed to be due to issues with yielding GPUs. There's a reason Intel wants 10k USD for these CPUs in their Xeon form.

 

Youch. My numbers assumed a really low defect rate, (i used Adoed TV's numbers for mature nodes 0.04 defects), if it';s significantly higher then they could be losing money. It also depends of course how many lower core count Xeon's they're selling ofc. But i'd expect enterprise to be chewing a lot more top core count parts than low core count ones so there's probably a lot of poor dies being binned. And for that matter their wafer costs, since it's their own fabs they could easily be making a major saving there. 

[Taf the Ghost]

32 minutes ago, CarlBar said:

 

Youch. My numbers assumed a really low defect rate, (i used Adoed TV's numbers for mature nodes 0.04 defects), if it';s significantly higher then they could be losing money. It also depends of course how many lower core count Xeon's they're selling ofc. But i'd expect enterprise to be chewing a lot more top core count parts than low core count ones so there's probably a lot of poor dies being binned. And for that matter their wafer costs, since it's their own fabs they could easily be making a major saving there. 

My impression is that the top SKU CPUs don't sell as many as you'd think, as it's the TCO that matters most, and the more efficient processors (the Gold models in this generation for Intel) are probably the hot sellers. But Intel owning its own fabs is what still makes this node work.

[leadeater]

27 minutes ago, Taf the Ghost said:

My impression is that the top SKU CPUs don't sell as many as you'd think, as it's the TCO that matters most, and the more efficient processors (the Gold models in this generation for Intel) are probably the hot sellers. But Intel owning its own fabs is what still makes this node work.

That was the same as last generation, go Gold/E5 unless you have a good reason for Plat/E7.

[Fasauceome]

On 1/30/2019 at 5:58 PM, Tristerin said:

Its misleading by default.

Hardly, it's abundantly clear throughout Steve's review that the stock 2990wx performance is the only thing included, and when power draw is highlighted, he doesn't even mention the OC power draw until the very end as an afterthought.

 

 

 

In fact, at the end of the review he says for the most part the Xeon is terrible value, highlighting the excellent price to performance advantage threadripper has. GamersNexus is one of the most accurate and thorough reviewers out there.

[Taf the Ghost]

39 minutes ago, leadeater said:

That was the same as last generation, go Gold/E5 unless you have a good reason for Plat/E7.

As much as "more cores!" is the matra, it's really about what those cores do. Every workload is different, and a lot of them are shifting towards both accelerators and a lot of memory. Only in servers would AMD be touting their ability to handle even more memory and that be a selling point. But it is.

[leadeater]

4 hours ago, Taf the Ghost said:

As much as "more cores!" is the matra, it's really about what those cores do. Every workload is different, and a lot of them are shifting towards both accelerators and a lot of memory. Only in servers would AMD be touting their ability to handle even more memory and that be a selling point. But it is.

Even then that is very easy to do since for what ever reason Intel has M sub variants that support more memory, why you would do that beats me. AMD supports what it supports, no M F or w/e else, what you see is what you get. Also none of that 1 FMA unit if Gold 5xxx and below or 2 FMA units Gold 6xxx and above, if it's hard to sell 50 different SKUs just try having 20 instead sighhhhh.

[Taf the Ghost]

1 minute ago, leadeater said:

Even then that is very easy to do since for what ever reason Intel has M sub variants that support more memory, why you would do that beats me. AMD supports what it supports, no M F or w/e else, what you see is what you get. Also none of that 1 FMA unit if Gold 5xxx and below or 2 FMA units Gold 6xxx and above, if it's hard to sell 50 different SKUs just try having 20 instead sighhhhh.

I can appreciate the Intel does semi-custom SKUs that make it to larger sales via the SIs, but some of the segmentation approach lately has been nuts. However, if we assume 14nm is a Production Node but not a great yielder, maybe up at say .25 defects per cm², it would explain why Intel went even more nuts with the segmentation. The cost to yield certain SKUs really is a lot higher. (I'm kind of wondering if 14nm doesn't like the AVX512 unit, as that seems to be the part that's been limited most in the segments. Maybe it's responding like the GPU problems on they had on the node?)

[jagdtigger]

Intel still got rekt by the TR chip... Performance is darn close, but power consumption is way less. Not to mention the price gap X'D .

[jagdtigger]

Intel still got rekt by the TR chip... Performance is darn close, but power consumption is way less. Not to mention the price gap X'D .

[strajk-]

I can see a couple reasons why I would pick this CPU over a TR:

1. I want to replace my Central Heating at home
2. Clout of having a component that has more power consumption that an African Village
3. A lot of company funds left and need to hit the threshold to not have a reduction next term
4. CPU Premine some random Shitcoin that's never going above 10 sats with a decent volume to dump
5. You tell me

[GoldenLag]

7 minutes ago, strajk- said:

I want to replace my Central Heating at home

4 vega cards would do the job better. because you could at least mine on them or fold using them. 

 

or you could use both, i wont stop you. 

[strajk-]

Just now, GoldenLag said:

4 vega cards would do the job better. because you could at least mine on them or fold using them. 

 

or you could use both, i wont stop you. 

If we're going all out why not 4 2080Ti's instead?

At least that way we have even more heat and the possibility of having a Scented VRAM Candle.

[GoldenLag]

1 minute ago, strajk- said:

If we're going all out why not 4 2080Ti's instead?

At least that way we have even more heat and the possibility of having a Scented VRAM Candle.

because you cant crossfire or SLI those together. 

 

workstation Vega cards can be 4 way crossfired. also they are hotter. 

[hobobobo]

20 minutes ago, strajk- said:

I can see a couple reasons why I would pick this CPU over a TR:

1. I want to replace my Central Heating at home
2. Clout of having a component that has more power consumption that an African Village
3. A lot of company funds left and need to hit the threshold to not have a reduction next term
4. CPU Premine some random Shitcoin that's never going above 10 sats with a decent volume to dump
5. You tell me

Got dirt cheap power at my county home, pondered doing central heating with gpus, but keeping a loop of this size clean is a nightmare + balancing components power to get ~40c (in winter) constant at that loopsize is a nightmare of its own, even with sensors scripts etc. Also gotta add a "grounded" ofshoot to the loop for the summer, if you want to use it as cooling or at least not toasting you alive while its +30 outside. The idea is fascinating but a goddamn nightmare to implement and without better then 3g interwebs couldnt figure any reason to do it other then it would look cool as balls to have central heating room full of copper pipes, psus and pcbs on risers

 

 

You can also use 2 of them for a posh dining expirience, where the food is beeing cooked on 1,5kw of computational power busy figuring out all the little bits to make your experiense sublime

[Taf the Ghost]

33 minutes ago, hobobobo said:

Got dirt cheap power at my county home, pondered doing central heating with gpus, but keeping a loop of this size clean is a nightmare + balancing components power to get ~40c (in winter) constant at that loopsize is a nightmare of its own, even with sensors scripts etc. Also gotta add a "grounded" ofshoot to the loop for the summer, if you want to use it as cooling or at least not toasting you alive while its +30 outside. The idea is fascinating but a goddamn nightmare to implement and without better then 3g interwebs couldnt figure any reason to do it other then it would look cool as balls to have central heating room full of copper pipes, psus and pcbs on risers

 

 

You can also use 2 of them for a posh dining expirience, where the food is beeing cooked on 1,5kw of computational power busy figuring out all the little bits to make your experiense sublime

A GPU-based heating system would need double heat transfer, a lot like how a nuclear reactor works. The big problem with old oil or water radiator systems is eventually they became really inefficient and cleaning them normally required replacing everything.

[Taf the Ghost]

37 minutes ago, hobobobo said:

Got dirt cheap power at my county home, pondered doing central heating with gpus, but keeping a loop of this size clean is a nightmare + balancing components power to get ~40c (in winter) constant at that loopsize is a nightmare of its own, even with sensors scripts etc. Also gotta add a "grounded" ofshoot to the loop for the summer, if you want to use it as cooling or at least not toasting you alive while its +30 outside. The idea is fascinating but a goddamn nightmare to implement and without better then 3g interwebs couldnt figure any reason to do it other then it would look cool as balls to have central heating room full of copper pipes, psus and pcbs on risers

 

 

You can also use 2 of them for a posh dining expirience, where the food is beeing cooked on 1,5kw of computational power busy figuring out all the little bits to make your experiense sublime

@AlexTheGreatish

 

@hobobobo points to an interesting idea to a Space Heater v2 build for LTT.  You could also burn through all of that extra mineral oil LMG has around.  Loop the GPU + CPU into 2x 480mm external rads, and have those in a mineral oil bath (probably want to be able to seal that) which is the transfer medium into large, standing radiators. 

[Ashleyyyy]

1 hour ago, strajk- said:

I can see a couple reasons why I would pick this CPU over a TR:

1. I want to replace my Central Heating at home

I know that is a joke, but I don't know why people keep making such jokes. 

 

I ran a dual X5650 rig with a 780Ti and 290X (780Ti for Linux host and 290X for Windows VM) in it for a bit. the gpu's were only used one at a time of course, but we are still talking about 600 watts when gaming using the 290X. it warmed up my room by a few degrees after an evening of playing but not nearly enough to heat a house. 

[Taf the Ghost]

Just now, firelighter487 said:

I know that is a joke, but I don't know why people keep making such jokes. 

 

I ran a dual X5650 rig with a 780Ti and 290X (780Ti for Linux host and 290X for Windows VM) in it for a bit. the gpu's were only used one at a time of course, but we are still talking about 600 watts when gaming using the 290X. it warmed up my room by a few degrees after an evening of playing but not nearly enough to heat a house. 

Large radiator systems don't out put all that much wattage. What they do is keep a large heat sink (the fluid) at a temperature the meets your needs.