AMD Ryzen 9 3950X

[DanceswithUnix]

... and implying Pifast is somehow representative of 'single threaded performance' as a whole is disingenuous at best.

Sadly I think any dismissal of PiFast is taken as AMD fanboy sour grapes. So I suggest you just ignore it and move on; fanboy

What I really want to see is a benchmark of Intel's Quartus VHDL compiler. That should (unsurprisingly) favour Intel chips, but historically that has in part been down to Intel using a larger L3 cache and I would love to see how a big Intel chip fares against AMD's best on what is a genuine workstation benchmark. Releasing such results is against the software terms and conditions of the compiler unless you get written permission from Intel, but I don't know if anyone has ever simply asked! The compiler has multi threaded and single threaded portions during the different phases of optimisation and layout, and a pretty big memory footprint.

[scaryjim]

I don't think I've bought and used a retail cooler for ages. Used one or two in a few builds I've done for others, but I can't say I've used a stock cooler myself for years (actually decades). Thinking back I think my last stock cooler was for the Celeron 300A (a CPU launched in the 1990s!).

Perhaps it's partly down to AMD's recent stock cooler being pretty good (and excellent compared to the "aluminium block of suck" they used to provide). That said, I can't remember the last time I didn't use the stock cooler in a new build. Perhaps that's influenced by the fact that I've built a lot of cheap, entry level machines over the years, though...

[Tabbykatze]

I assume it's the fact that you need to buy one on top of the cost of the chip, rather than getting a cooler included as part of the retail package? I don't think I've got any high quality AM4 coolers just lying around, so I'd need to budget for one if I was going to do a 3950X build (which I'm not, but....)

If you're buying a CPU at £750, you're budgeting also for a proper cooler. At these price points I don't ascribe the ridiculous notion of "oh no, I need a cooler to add to my £1500+ build".

In budget builds, I get that, but this is not a budget CPU.

Thankfully, by it not including a cooler, the numerous reviews who go "whoops Intel didn't provide a cooler, lets use a £100 cooler on it! Oh look, AMD provided an in box cooler, lets use that for our review and overclocking guide!" will not be able to play that crud.

Sadly I think any dismissal of PiFast is taken as AMD fanboy sour grapes. So I suggest you just ignore it and move on; fanboy

I think the only thing about PiFast is it is just the mathematical side of the processor being taken into account, nothing else (IIRC, someone correct me if i'm wrong). So we are focusing on a single aspect of the processor.

The limited benchmarks run and the very odd gaming benchmark choices that constantly persist are why I only use Hexus reviews for the comments.

[00oceanic]

Error on Page 4
"If you thought Ryzen 9 3900X was extremely fast, think again. An extra two cores and four threads, along with decent frequency, makes the 3950X a mainstream monster."
Should be 4 cores and 8 threads.

Great review! Seems kinda pricey, but that's what dream machines are for

[DanceswithUnix]

I think the only thing about PiFast is it is just the mathematical side of the processor being taken into account, nothing else (IIRC, someone correct me if i'm wrong). So we are focusing on a single aspect of the processor.

tldr: There are lots of ways to do maths and lots of pitfalls, and we don't actually know what pifast is doing.

Single aspect is fine. Maths performance is singled out in many benchmarks, like the 3D particle movement tests on Anandtech. If you read through that part of Dr Cutress's review it makes interesting reading and shows the sort of problems you get with doing heavy computation, he has version 1 results at the end of their review to show what badly written multi threaded code would run like as well as the version 2 which is better written to get. Stuff it, let's paste them in...

First we have the badly written version, as written for his PhD. The version 2.1 improves the scores a lot, and the order mixes around a bit. So which CPU is best? Enter the final AVX based graph, and the Xeon based i9 chips with their massive FPU leaves the rest for dust as this was actually what they were designed for.

These three programs are from the same person, performing the same algorithm, but with different implementation and giving wildly different but all equally valid outcomes. That's fine because it is explained what the code is doing so the transparency means you can decide which graph means the most to you. In a similar way if you compress some video in Handbrake which is also maths heavy, it actually tells you what instructions it has found that it can use. PiFast has none of that, it is neither a real world task like Blender or Handbrake and not explained in implementation details like 3DPM.

(edit: My best guess btw is that PiFast is using early SSE instructions given it seems to have been written for the Pentium 3, and also seems heavy on memory accesses, but it could be using 8087 fpu instructions for all we know).

[Tabbykatze]

You're absolutely right, it's this not knowing problem that catches a lot out.

Hexus also have very few data points in their reviews so it makes it hard to use it as good source. It's a source nevertheless just not one I consider to be a premium one.

[watercooled]

Sadly I think any dismissal of PiFast is taken as AMD fanboy sour grapes. So I suggest you just ignore it and move on; fanboy

What I really want to see is a benchmark of Intel's Quartus VHDL compiler. That should (unsurprisingly) favour Intel chips, but historically that has in part been down to Intel using a larger L3 cache and I would love to see how a big Intel chip fares against AMD's best on what is a genuine workstation benchmark. Releasing such results is against the software terms and conditions of the compiler unless you get written permission from Intel, but I don't know if anyone has ever simply asked! The compiler has multi threaded and single threaded portions during the different phases of optimisation and layout, and a pretty big memory footprint.

I would find a proper analysis of Pifast's code and performance genuinely interesting. It's just, as it stands it's just a black box with seemingly strange performance characteristics which makes it kinda pointless as a relevant benchmark. Even if what you actually want to do is calculate Pi, Pifast is horrifically slow when compared against at least one alternative, even when you artificially restrict the alternative to a single thread.

[DanceswithUnix]

I would find a proper analysis of Pifast's code and performance genuinely interesting. It's just, as it stands it's just a black box with seemingly strange performance characteristics which makes it kinda pointless as a relevant benchmark. Even if what you actually want to do is calculate Pi, Pifast is horrifically slow when compared against at least one alternative, even when you artificially restrict the alternative to a single thread.

lol, yeah. I did briefly consider breaking out the debugging tools to do some analysis on it, and then realised the fact it was pants even at calculating Pi meant I really didn't care enough to invest time

[OOSeven]

How is needing a high-quality cooler a bad? xD

I mean, I guess you need something to put in a bad, lol!

I think its the fact that it's NEEDED, rather just being a nice optional upgrade. Although yeah they are clutching at straws with that one

[DevDrake]

Thats a monster.... aaaaahhhhh it has 4 times more cores and threads than my current setup!!!

[Tattysnuc]

Might be time to dust off my old vapochill unit

[DanceswithUnix]

Might be time to dust off my old vapochill unit

Respect for even having an old vapochill unit!

They would seem just the ticket if it still works, though I suspect this could kick out more heat stock than a CPU from 2003 would kick out overclocked. Perhaps *that* is what the 65W Eco mode is for?

[scaryjim]

... I suspect this could kick out more heat stock than a CPU from 2003 would kick out overclocked. ...

Pentium 4 hit 92W TDP by the end of 2003 (Extreme Edition 3.2GHz). I imagine that'd reach 105W+ with a mild overclock.

By mid-2004 they reached 115W TDP @ stock.

[Tabbykatze]

Pentium 4 hit 92W TDP by the end of 2003 (Extreme Edition 3.2GHz). I imagine that'd reach 105W+ with a mild overclock.

By mid-2004 they reached 115W TDP @ stock.

And people moan about 105w TDPs!!!

[scaryjim]

And people moan about 105w TDPs!!!

If you think that those 115W TDPs were for single core hyperthreaded processors running at around 3.6GHz, it throws a lot of light on just how much progress has been made in the last 15 years...!

[kalniel]

If you think that those 115W TDPs were for single core hyperthreaded processors running at around 3.6GHz, it throws a lot of light on just how much progress has been made in the last 15 years...!

But also.. not like for like. There has been amazing progress, but the TDP 'rating' has vastly changed as well. Intel's 95W is actually 210W TDP (both intel figures) for PL2 state - a sort of rolling window of power consumption which is meant to be more or less temporary but in practise is set to infinite by most BIOSes.
https://www.anandtech.com/show/13400...600k-review/21