Users with Intel Xeon W-3175X.

[sedy25]

N

4.5ghz all core? So what you're saying is that your overclocked TR was not stable?

No what I'm saying AMD user, is run 4.5GHz in AVX256 workloads and let me know how stable that architecture really is with the lack of 256 bit data paths and registers.

 

[Vince]

N

No what I'm saying AMD user, is run 4.5GHz in AVX256 workloads and let me know how stable that architecture really is with the lack of 256 bit data paths and registers.

I'm also an intel user. So run 4.5ghz on a single or on all cores? One is an overclock, one is stock. What is it?

What is it with people these days, yes as a community we should question unsubstantiated claims, if what you're saying is that it's not stable on a given overclock then fine. But if you are saying that running stock it crashes on avx256 then clearly you have/had a problem that the rest of the world don't.

 

[sedy25]

I'm also an intel user. So run 4.5ghz on a single or on all cores? One is an overclock, one is stock. What is it?

All cores obviously.

 

[Vince]

All cores obviously.

Then no more needs to be said. Your overclocked TR rig was not stable, it really is as simple as that. The 3175x is a cool (not literally) chip though so enjoy fiddling with it

 

[sedy25]

I

Then no more needs to be said. Your overclocked TR rig was not stable, it really is as simple as that. The 3175x is a cool (not literally) chip though so enjoy fiddling with it

It crashed at stock too, the IMC is trash, bandwidth is trash, L1 cache bandwidth is trash, gaming performance is trash, AVX256 performance sucks, non-existent AVX512, less efficient SIMD instruction sets, incomplete AVX256, need I go on? Or should I post a picture of the W-3175X beating the 3960X with 4 cores disabled at stock at lower frequencies?

I'll stick with Intel, not this new garbage.

 

[Vince]

I

It crashed at stock too, the IMC is trash, bandwidth is trash, L1 cache bandwidth is trash, gaming performance is trash, AVX256 performance sucks, non-existent AVX512, less efficient SIMD instruction sets, incomplete AVX256, need I go on? Or should I post a picture of the W-3175X beating the 3960X with 4 cores disabled at stock at lower frequencies?

I'll stick with Intel, not this new garbage.

Mate you do you. Enjoy, if it's the right cpu choice for your workloads then nobody on here is going to argue that, me included, people will however call out BS when they smell it. I'm sure if all chips were crashing stock on AVX that it would be widely publicised, it isn't. I'm not saying it's a better or worse chip not at all but am calling out claims that can't be substantiated in terms of platform stability.

 

[Deleted member 66701]

Run 4.5ghz in prime95 small fft

Lol what? No wonder your TR was crashing!

 

[Deleted member 66701]

I

It crashed at stock too, the IMC is trash, bandwidth is trash, L1 cache bandwidth is trash, gaming performance is trash, AVX256 performance sucks, non-existent AVX512, less efficient SIMD instruction sets, incomplete AVX256, need I go on? Or should I post a picture of the W-3175X beating the 3960X with 4 cores disabled at stock at lower frequencies?

I'll stick with Intel, not this new garbage.

Yeah, now you're just trolling.

@Vince Do you still have the 1950x I sold you?

 

[Vince]

Yeah, now you're just trolling.

@Vince Do you still have the 1950x I sold you?

I have a 1950x but not one you sold me - I was silly and bought one on launch day way back when At the moment I have pretty much one of everything. I have all gens of ryzen, including 4000 (renoir - has an awesome imc!) and all the TR's, well one of each series and multiple of each series of EPYC. I also still have some intel stuff (mainly xeons) around that I play with from time to time. I just pulled 100 8th/9th gen CPU's from my estate as well so currently sitting on absolute stacks of intel chips.

To be fair, even 1950x isn't trash at gaming There literally isn't a game it can't play at more than acceptable frame rates with everything maxed on that system @ 3440x1440, this is also using NUMA (distributed) memory as other workloads I use benefit and i'm too lazy to reboot.

On that same SP3 socket:



I swear to god if I hadn't moved them into a co-lo they would never have been rebooted since deployment I am sure I was on more than 300 days uptime before the move. All 3 hosts are currently on 123 days

 

[Deleted member 66701]

I have a 1950x but not one you sold me - I was silly and bought one on launch day way back when At the moment I have pretty much one of everything. I have all gens of ryzen, including 4000 (renoir - has an awesome imc!) and all the TR's, well one of each series and multiple of each series of EPYC. I also still have some intel stuff (mainly xeons) around that I play with from time to time. I just pulled 100 8th/9th gen CPU's from my estate as well so currently sitting on absolute stacks of intel chips.

To be fair, even 1950x isn't trash at gaming There literally isn't a game it can't play at more than acceptable frame rates with everything maxed on that system @ 3440x1440, this is also using NUMA (distributed) memory as other workloads I use benefit and i'm too lazy to reboot.

On that same SP3 socket:



I swear to god if I hadn't moved them into a co-lo they would never have been rebooted since deployment I am sure I was on more than 300 days uptime before the move. All 3 hosts are currently on 123 days

Ah, must have been someone else on here that I sold it to.

Yeah, my 2950x at uni is on 325 days uptime atm.

I've gone 1950x > 2950x > 3960x and not had a problem with any of them. I will definitely be going 5000 series TR when they launch.

 

[Vince]

Ah, must have been someone else on here that I sold it to.

Yeah, my 2950x at uni is on 325 days uptime atm.

I've gone 1950x > 2950x > 3960x and not had a problem with any of them. I will definitely be going 5000 series TR when they launch.

I'm not too sure what I am going to do tbh but I have certainly deployed enough EPYC/TR to know that the platform is sound. My desktop at home could do with a refresh (1950x, 64gb 3466, Radeon 7) but then in reality most projects i'm running right now I still have plenty of power for in this rig and can always offload anything heavy to EPYC/Xeon estates, im old now so overclocking is in my past for the most part, the gains aren't worth the stability issues especially in business where availability is key. On a platform which is intended for out and out stability TR/Xeon-W overclocking for me is not an option and I would hazard a guess that most businesses don't want to be running "overclocked" intel chips in their estate, it doesn't make sense when stability/availability is king.

Ill probably go TR 5000 for my home workstation unless something magic happens on Intels side of the fence.

 

[Deleted member 66701]

OP's Intel fanboy bias confirmed:-

https://www.overclockers.co.uk/forums/posts/34119934/
https://www.overclockers.co.uk/forums/posts/34119671/

 

[Vince]

OP's Intel fanboy bias confirmed:-

https://www.overclockers.co.uk/forums/posts/34119934/
https://www.overclockers.co.uk/forums/posts/34119671/

His bias to me doesn't matter, many of us have a preference and im the sort of guy that can see both sides of the coin. It's unsubstantiated and frankly wrong claims that get me, weirdly backed up by 8pack in this instance?! I'd love to know what this actually means backed up by something more than a throw away comment:

You're correct in saying the platform is more stable than trx40 but at a financial cost.

I also corrected the You're just because...If somebody can prove categorically that the Xeon-w platform is "more stable" than TR for any given workload at stock settings then ill give it some time. Otherwise my opinion is unwavering and I will continue to believe that this is just words that are frankly opinion and nothing more.

 

[d_brennen]

What a noddy thread

 

[Vince]

What a noddy thread

Thread won't go anywhere, nobody will come along and substantiate any of these claims. 8pack likely wont come back to back up the throw away comment with any real data and I'm fairly confident nobody else is going to come along and try and prove such a claim. Meanwhile loads of people out there are happily running avx256 workloads on TR so really what we have is somebody who clocked 3970x to 4.5ghz all core and couldn't get it stable. All the other trash talk is just that.

At this point all you are left with is an overclocked (45%) £2,250 chip running a chiller besting a £1,500 chip in a particular workload. Everything else in here on stability, gaming performance, cache layout etc is all junk.

 

[tommoT.Striker]

curious thread...

What can I say, I bought the TR 3960X CPU especially for it's AVX256 performance. Love it and the TRX40 platform. Looking forward to some 4th Gen TR single core boost too, when the time comes.

Me thinks people try and run a threadripper at low voltage at 4.4Ghz+ then complain that it's an unstable platform...

 

[koooowweeee]

I also corrected the You're just because...If somebody can prove categorically that the Xeon-w platform is "more stable" than TR for any given workload at stock settings then ill give it some time. Otherwise my opinion is unwavering and I will continue to believe that this is just words that are frankly opinion and nothing more.

As he's a overclocker I'm going to guess that he's talking about when both are overclocked. Neither AMD or intel are going to release a unstable at stock CPU

 

[Vince]

As he's a overclocker I'm going to guess that he's talking about when both are overclocked. Neither AMD or intel are going to release a unstable at stock CPU

Overclocking and running outside of intended spec what do you expect? But he did clarify that it was junk and unstable at stock and that it was also junk as it couldn't do all core 4.5ghz avx256. It doesn't really matter anyway as in the end it's not really reasoned debate more of an obscure thread on a product that is niche to the extreme, if however, all you want to do is overclock the balls out of it and win records then that's something all together different, for it's intended stock use it's a hard product to shift imo. The fact is for most on here £1500 on a threadripper is bonkers money and £2,250 on any cpu is crazy talk. Away from all that though that 4.6ghz overclock is pretty cool even if it did take a nuke and a chiller.


For it's actual intended use buying these cpu's outside of instruction specific workloads such as AVX 512 in a corporate setting is a different story entirely.

 

[stuie]

His bias to me doesn't matter, many of us have a preference and im the sort of guy that can see both sides of the coin. It's unsubstantiated and frankly wrong claims that get me, weirdly backed up by 8pack in this instance?! I'd love to know what this actually means backed up by something more than a throw away comment:



I also corrected the You're just because...If somebody can prove categorically that the Xeon-w platform is "more stable" than TR for any given workload at stock settings then ill give it some time. Otherwise my opinion is unwavering and I will continue to believe that this is just words that are frankly opinion and nothing more.

In fairness to 8packs comment, I too would expect xeon to be more stable than TR but then I'd also expect it to be more stable than an I-series cpu too, I'd also expect epyc to be more stable than TR and I series. Server grade cpu's with eec ram and boards should be more stable. Thats not to say i'd expect a TR rig to be crashing all over the place just that theres a reason server grade chips are more expensive and its not cause they are shiny

 

[Vince]

In fairness to 8packs comment, I too would expect xeon to be more stable than TR but then I'd also expect it to be more stable than an I-series cpu too, I'd also expect epyc to be more stable than TR and I series. Server grade cpu's with eec ram and boards should be more stable. Thats not to say i'd expect a TR rig to be crashing all over the place just that theres a reason server grade chips are more expensive and its not cause they are shiny

So both TR and also that Xeon-W both have ECC support are both workstation grade chips, the xeon-w is not a server grade chip and you wont find them in the server room. The W literally stands for workstation, the server version of the chip would likely be the Xeon Platinum 8176. I still don't understand the metrics based around your comment, if everything supports ECC on a server grade socket with server grade memory we just end up back to this expectation but once again backed up by zero metrics... So ill ask again what are the metrics? where is any proof that the Xeon platform is more stable than the TRX40?

Let me save you the effort, it isn't and the metrics don't exist. However, i'm open to somebody coming along and trying to prove this.