*** AMD "Zen" thread (inc AM4/APU discussion) ***

[Silent_Scone]

Over years i found out that only real stability test is.... Producing 2 hour long video in PowerDirector.
I could pass Prime Handbrake Intel burn test at maximum. And still crash after 2 hours of rendering .

It can depend on the platform, Prime doesn't really have much of a place on the recent Intel platforms. Until Intel updated the Microcode early on for Skylake, it was possible to pass Prime for hours and fail Handbrake in minutes.

 

[harney]

Memory latency might seem high but everything else is pretty mind blowing.

Compared to my 4.2Ghz 5820k on 2400Mhz memory(it's 3000Mhz, worked at that for a while but stopped booting until mem speed was dropped ), the L2 cache of Zen literally doubles performance, L3 has 50% higher read, 3x the copy and marginally behind on write. L1 is a little slower on Zen. Read though... wow. My quad channel 2400Mhz is getting 46.9GB/s, Zen with 3200Mhz is beating it. Max theoretical numbers are once again popping up showing AMD smashing to pieces anything Intel has on efficiency. I don't have l2/l3 latency as trial version hides that, L1 read(so read might be great), also memory write/copy are hidden but there is no reason write will be monumentally higher than read, nor is it more important anyway.

At 2133Mhz Zen appears to have almost 100% bandwidth efficiency, Intel is at around 75% on Skylake/Broadwell, at higher memory speeds efficiency reduces for Zen it appears and Intel. Max theoretical on my system is 76.8GB/s and I'm getting 47GB/s, around 61% efficiency. 3200Mhz dual channel has max theoretical bandwidth of 51.2GB/s, so 47.7GB/s is still 93% efficiency.

Zen doesn't have a weak IMC< it has one of the best IMC's ever seen, with incredible bandwidth.

That bodes badly for Intel when it comes to APUs later this year. AMD has the HBM for super high margin products and Intel memory cube seems to be AWOL(and much more expensive and higher power, maybe it less use for mobile where Iris pro is aimed), but in more mainstream pricing AMD is looking like it will have 20-30% more memory bandwidth for their APUs at any given memory speed used. AMD APUs are going to absolutely destroy Intel on igpu performance.

Very impressed thx for the info

 

[humbug]

@Gibbo can you please confirm if your screw is too long or not ?

 

[balky12]

Yes, I'm not dissing on AIOs, they perform just fine for the most part, but some people do have high expectations simply because they have liquid in them, which they believe is assured to offer a far superior cooling solution, when it factually does not. It's quite obvious many people buy them for this reason. The aesthetics argument is by far the strongest, and as mentioned, where high profile RAM or case clearance height is air cooler prohibitive.

SNIP

I have a 6/7 year old Akasa Nero S Cooler sitting massively on my 4670k @ 4.0Ghz (not the best OC granted). Chip at idle sits in the low 30s, whilst under typical gaming load I've never seen it break 55.

Shame it's a PITA to install on anything...

I will most likely pick up an AIO with my next Ryzen/ Vega upgrade. But purely for the looks.

 

[Zhuni]

So which mobo is that with the screw too long. Looks like msi titanium

 

[Omega]

Which M-ATX motherboard did people go for ?

 

[Popov]

Interesting info on some mobos having issues already

http://www.tomshardware.com/news/amd-ryzen-am4-motherboard-heatsink-problem,33771.html

 

[Zeed]

It can depend on the platform, Prime doesn't really have much of a place on the recent Intel platforms. Until Intel updated the Microcode early on for Skylake, it was possible to pass Prime for hours and fail Handbrake in minutes.

Ill run everything anyway But thats the case with x99 and it was same with z77 2500k

 

[N19h7m4r3]

It can depend on the platform, Prime doesn't really have much of a place on the recent Intel platforms. Until Intel updated the Microcode early on for Skylake, it was possible to pass Prime for hours and fail Handbrake in minutes.

Over years i found out that only real stability test is.... Producing 2 hour long video in PowerDirector.
I could pass Prime Handbrake Intel burn test at maximum. And still crash after 2 hours of rendering .

I found the same. For stability I'll fire up AIDA 64, and leave it for hours. Once that passes, it's time to use Premiere Pro on a massive project. Both VBR1 and VBR2 encodes to see if it holds up.
Often a heavy real workload is what you need to see if it holds up well.

 

[TheBiznes]

@drunkenmaster
Yes very impressive L1 is faster on 6800, but level 2 and 3 are a 1/3 to a double higher on Ryzen. So much for Quad vs Dual argument meaning dual channel being weak AMD is very strong.
Unlocked info 6800 on a friends machine for com
Level 1 1043.00 Gbs R 713.51 Gbs W 1426.2Gbs C 1.1ns
Level 2 523.40 Gbs R 226.56 Gbs W 335.51Gbs C 3.2ns
Level 3 197.09 Gbs R 137.35 Gbs W 165.41Gbs C 15.8ns

 

[harney]

So which mobo is that with the screw too long. Looks like msi titanium

i do not think its the MB at fault is it..... its the backplates from the cooler companies

 

[KingDazza]

It's worth noting, the 1700x vs 6800k test used a 1080, the 1700 vs 7700k used a 1070, so that is why performance is different.

The real important thing is a stock 1700(which is unlocked) gets extremely close to a chip running around 1GHz faster than it, and by all accounts the 1700 clocks pretty well, it can close a lot of that gap. Sure the 7700k can also overclock but it overclocks a smaller percentage and after you hit a certain point gains will diminish significantly and the 7700k is almost there already.

The 1700 only seems to be running 3.2Ghz or so with multiple threads running, because it's a 65W chip. The 1700 65W chip is literally beating the 91W(iirc) 7700k in some of those games already, even if you overclock enough to be using 91W you'd be beating it in almost everything.

I will actually give you credit though, of all people you are saying a 1800x will not beat but smash a 7700k. I think a overclocked 1700 will beat the 7700k in most games now(newer releases, older stuff maybe not so much, but also not too relevant).

So, you're saying that an i7 7700k running a 1070 beat a 6800k and a 1700x running a 1080?? Really?

 

[Silent_Scone]

Memory latency might seem high but everything else is pretty mind blowing.

Compared to my 4.2Ghz 5820k on 2400Mhz memory(it's 3000Mhz, worked at that for a while but stopped booting until mem speed was dropped ), the L2 cache of Zen literally doubles performance, L3 has 50% higher read, 3x the copy and marginally behind on write. L1 is a little slower on Zen. Read though... wow. My quad channel 2400Mhz is getting 46.9GB/s, Zen with 3200Mhz is beating it. Max theoretical numbers are once again popping up showing AMD smashing to pieces anything Intel has on efficiency. I don't have l2/l3 latency as trial version hides that, L1 read(so read might be great), also memory write/copy are hidden but there is no reason write will be monumentally higher than read, nor is it more important anyway.

At 2133Mhz Zen appears to have almost 100% bandwidth efficiency, Intel is at around 75% on Skylake/Broadwell, at higher memory speeds efficiency reduces for Zen it appears and Intel. Max theoretical on my system is 76.8GB/s and I'm getting 47GB/s, around 61% efficiency. 3200Mhz dual channel has max theoretical bandwidth of 51.2GB/s, so 47.7GB/s is still 93% efficiency.

Zen doesn't have a weak IMC< it has one of the best IMC's ever seen, with incredible bandwidth.

That bodes badly for Intel when it comes to APUs later this year. AMD has the HBM for super high margin products and Intel memory cube seems to be AWOL(and much more expensive and higher power, maybe it less use for mobile where Iris pro is aimed), but in more mainstream pricing AMD is looking like it will have 20-30% more memory bandwidth for their APUs at any given memory speed used. AMD APUs are going to absolutely destroy Intel on igpu performance.

3200Mhz with tight sub timings on a 5960x (write speed is tied to uncore speed). 3000Mhz probably didn't work well for you as that ratio is still pretty barfed even now. The strongest ratio on Haswell is 3200

 

[Nutella33]

i do not think its the MB at fault is it..... its the backplates from the cooler companies

beQuiet Silent loop and Alphacool Eisbaer effected

According to the cooler thread so far.

 

[iHATEMacs]

I just phoned two different companies (NOT OcUK) and both are having trouble with 1800X not being in stock (well one guy said he received 4 today when he was supposed to receive 200! ). The 1700 and 1700X are of no issue and have been in stock for nearly a week.

All stock coming from multiple suppliers and none of them can deliver on time, which points me to AMD having supply troubles. I've bought mine locally so I can pick it up in person on day of release (instead of waiting for delivery on 3rd) and apparently it's not gonna be here until Friday now anyway!!!
Also the Crosshair mobo hasn't arrived on time either.

Anyone else heard anything similar?

 

[Techen]

I just phoned two different companies (NOT OcUK) and both are having trouble with 1800X not being in stock (well one guy said he received 4 today when he was supposed to receive 200! ). The 1700 and 1700X are of no issue and have been in stock for nearly a week.

All stock coming from multiple suppliers and none of them can deliver on time, which points me to AMD having supply troubles. I've bought mine locally so I can pick it up in person on day of release (instead of waiting for delivery on 3rd) and apparently it's not gonna be here until Friday now anyway!!!
Also the Crosshair mobo hasn't arrived on time either.

Anyone else heard anything similar?

Am going to be thankful as i expect my asrock to be delayed. So now am forced to read the reviews and gives me a chance to change my mind

 

[Popov]

i do not think its the MB at fault is it..... its the backplates from the cooler companies

The Noctua AM4 kit uses the backplate already supplied with the motherboard (you just replace the bracket on the top of the board with the Noctua kit) so if other heatsinks do the same then the fault lies with the backplate supplied by the motherboard manufacturer.

 

[Yaayuh!]

The problem is, just looks like Tom's trolling. No info on cooler or motherboard/s but just says that their "could"/"may be" an issue. Great work...

 

[Silent_Scone]

The problem is, just looks like Tom's trolling. No info on cooler or motherboard/s but just says that their "could"/"may be" an issue. Great work...

Who's the author on it? Crashman?

 

[gavinh87]

Wanting to pull the trigger on either a 1700 or 1700x. Want to order now whilst still in stock and I know mobo stocks are not great. Where are the leaks when you need them?