Skylake-X i7-7820x Real world numbers...

[TahoeDust]

Alright.  I can call my 7820x completely stable at 4.7GHz on 1.225v.  Cooling is a H115i with push/pull Noiseblocker 140mm fans.  RAM is Corsair 3200MHz running stock 16-18-18-36 timing.  Mesh was set to 3200MHz. 

 

Here are the real world numbers...  

 

In Cinebench it scored 2035cb in Multi thread and 204cb Single Core.  The hottest core was #1 hitting 72c.

 

In Geekbench it scored 33189 Multicore and 5831 Single-core.  The hottest core was #3 and #7 both hitting 67c.

 

In Realbench for 15 minute test the hottest core was #3.  It hit 78c.

 

In Prime with AVX in a 20 minute test the hottest core was #2.  It hit 88c. AVX offset was -500MHz)  Personally I consider Prime with AVX a very unrealistic usage scenarios for me, so I am OK with this.

 

AIDA Cache and Memory looks like this...

 

Just for shits and gigs here are some 3D benches with a 1080ti FTW3...

http://www.3dmark.com/fs/13034422

 

http://www.3dmark.com/spy/2015990

 

 

[manikyath]

i just wanna note.. if you're into the kind of things where this performance applies to your workload.. thats some dope AF extra pricetag over ryzen... also that wattage..

 

just out of curiousity, with recent talks about VRMs overheating, what are VRM temps like?

[TahoeDust]

19 minutes ago, manikyath said:

i just wanna note.. if you're into the kind of things where this performance applies to your workload.. thats some dope AF extra pricetag over ryzen... also that wattage..

 

just out of curiousity, with recent talks about VRMs overheating, what are VRM temps like?

I was wondering how long this thread would go until the Ryzen "Value Police" showed up...first post.

 

I'm running Prime95 again right now to get VRM temps...just for you.

[TahoeDust]

30 minutes ago, manikyath said:

just out of curiousity, with recent talks about VRMs overheating, what are VRM temps like?

After 20 minutes in Prime with AVX the VRM temps hit a terrifying 63c

 

[manikyath]

12 minutes ago, TahoeDust said:

I was wondering how long this thread would go until the Ryzen "Value Police" showed up...first post.

 

I'm running Prime95 again right now to get VRM temps...just for you.

wasnt being "value police" as much as facepalming at the price intel stuck on these parts..

 

and the VRM temps is actually honest curiousity, i cant imagine they're good seeing you're squeezing close to 300 watts trough that thing, just wondering how "not within comfort zone" they are.

 

but you really gotta admit that tripling cpu price for about 20% extra performance isnt a good idea..

 

EDIT: doubling laptop price for a functional keyboard may be worth however...

[manikyath]

1 minute ago, TahoeDust said:

hit a terrifying 63c

what the flip are you doing to your VRMs to get them that cool? there's not a single board on the market i know of that does that with p95 on stock clocks..

[TahoeDust]

3 minutes ago, manikyath said:

what the flip are you doing to your VRMs to get them that cool? there's not a single board on the market i know of that does that with p95 on stock clocks..

Not a thing...this is actually looking pretty normal.

 

The board is one of the very few that does have a heatpipe on the VRMs and also a metal backplate behind them.  It also has 8+8 power.

 

Take a look at what OC3D measured their Asus Strix board at....

 

[manikyath]

Just now, TahoeDust said:

Not a thing...maybe it is being reported incorrectly?  

 

The board is one of the very few that does have a heatpipe on the VRMs and also a metal backplate behind them.  It also has 8+8 power.

cant have enough VRMs for 300 watts can we?  (totally not a jab at the "power the world" FM2 boards)

 

as for reported incorrectly.. seems likely, may be as simple as the sensor being a bit away from the VRMs. my board has two VRM sensors (one between VRMs and cpu, and one on the other side) and one of them is usually between 10 and 20°c hotter than the other.

 

on the note of heatpipes on VRMs, which exact board is this?

[done12many2]

44 minutes ago, manikyath said:

i just wanna note.. if you're into the kind of things where this performance applies to your workload.. thats some dope AF extra pricetag over ryzen... also that wattage..

 

just out of curiousity, with recent talks about VRMs overheating, what are VRM temps like?

 

Way to shit on his post!  This is getting so typical on LTT.

 

56 minutes ago, TahoeDust said:

Alright.  I can call my 7820x completely stable at 4.7GHz on 1.225v.  Cooling is a H115i with push/pull Noiseblocker 140mm fans.  RAM is Corsair 3200MHz running stock 16-18-18-36 timing.  Mesh was set to 3200MHz. 

 

Here are the real world numbers...  

 

 

Man, congrats on the build bud.  Results look good!

 

I'm not having any VRM related issues clear up to 4.9 GHz on 10 cores and 5 GHz on 8 cores in Cinebench.  Proof that if you make a video where you test the cheapest x299 boards that you can in an open bench configuration with no air flowing over the VRMs, you can make people believe that the sky is falling.  

 

[manikyath]

Just now, done12many2 said:

 

Way to shit on his post!  This is getting so typical on LTT.

yes. because out of experience with people who will pay 5 figures for this kind of performance, they all somehow choose dual-socket boards over the 200 watt monsters intel is deciding to create here..

 

there wouldnt be much purpose to everyone just saying "good job", without giving things any further thought.

--

just also wanna note, i'm equally negative about people buying R7 1800x's. there's a place where paying "exponentially" more for extra performance makes sense, and there's a place where you're just tossing money out the window.

[TahoeDust]

9 minutes ago, manikyath said:

cant have enough VRMs for 300 watts can we?  (totally not a jab at the "power the world" FM2 boards)

 

as for reported incorrectly.. seems likely, may be as simple as the sensor being a bit away from the VRMs. my board has two VRM sensors (one between VRMs and cpu, and one on the other side) and one of them is usually between 10 and 20°c hotter than the other.

 

on the note of heatpipes on VRMs, which exact board is this?

 

It is a Gigabyte Aorus Gaming 7. 

 

Take a look at the video in my edited post above.  I am pretty sure the temps are being recorded pretty accurately.

[manikyath]

10 minutes ago, done12many2 said:

 

I'm not having any VRM related issues clear up to 4.9 GHz on 10 cores and 5 GHz on 8 cores in Cinebench.  Proof that if you make a video where you test the cheapest x299 boards that you can in an open bench configuration with no air flowing over the VRMs, you can make people believe that the sky is falling.  

exactly why i was asking OP about his temps

[Drak3]

7 minutes ago, done12many2 said:

 

Way to shit on his post!  This is getting so typical on LTT.

I need to balance this out, quick!

 

 

 

Bruh, 7980XE and 4 Quadro P5000 are the true value. Uncompromised performance is true value!

[manikyath]

1 minute ago, TahoeDust said:

 

It is a Gigabyte Aorus Gaming 7. 

 

Take a look at the video in my edited post above.  I am pretty sure the temps are being recorded pretty accurately.

have you done any extended tests on this? not specificly P95 because it *is* kind of a worst case scenario, but lets say having 7zip benchmark run for upwards of an hour or more? that design of VRM cooling has the tendency to heat up over time if not given some direct airflow. although i do expect that to go fairly quickly with 250+ watts

i stuck a fan on my asus z97 board with a very similar design because keeping it under moderate cpu load for "more than the average workday" did have some rather disappointing results on VRMs, despite having my case act as a windtunnel.

 

but if those temps are actually what the VRMs will stick at for an extended time, gigabyte may have regained some trust here after not knowing their own bios post codes...

[MageTank]

@TahoeDust Surprisingly great results for an AIO, I'll eat my words now. Even with a -500 offset, 4.2ghz of Prime95 with 8 threads is considerably hot, especially considering this is unsoldered unlike the X99 SKU's we saw. Was that a custom workload at a specific FFT length? Or a normal blend? Smaller FFT lengths run considerably hotter, so if that is indeed small, it's even more impressive than I initially gave it credit for.

 

4.7 at 1.22 is actually on par with what I expected, given the results I've seen from @done12many2. As usual, my only real gripe with these bigger chips is the memory latency. Copious amounts of bandwidth, but poor latency scaling across the increased channel count. If you have the free time, I'd love to help tame that latency a little. Wouldn't take much more than tRFC and tREFI tweaking to get it to around 60ns. If you have time to actually overclock beyond 3200, then 50-55ns would be feasible. From what I hear, memory overclocking has done quite a bit for done1's benching results, which tells me these high core-count chips love that extra bandwidth/lower latency. 

[VagabondWraith]

2 hours ago, TahoeDust said:

*snip*

It seems RAM and mesh doesn't do much? ? I scored 2053 at 4.8 with RAM at 2666 and Mesh at 2.4. I'm wondering if higher RAM and Mesh speeds even do anything... I was going to up the RAM and cache but seems like there's no benefit. 

 

EDIT: maybe some benefit as I scored 2012 at 4.7 with everything else at default speeds.

 

Anyways, very nice scores! Nice to see you got things stable and these chips are NOT the surface of the sun people make them out to be. I was hitting 71 at 4.8, 1.28v with the package at 75.

Edited July 4, 2017 by wkdpaul
massive quote

[manikyath]

2 minutes ago, VagabondWraith said:

these chips are NOT the surface of the sun

well... however you spin it, 250 watts is 250 watts, its a matter of using propper cooling to not have your own miniature fusion reaction

[VagabondWraith]

Just now, manikyath said:

well... however you spin it, 250 watts is 250 watts, its a matter of using propper cooling to not have your own miniature fusion reaction

Well sure, my 1080 Ti pulls 320 OC but it has the performance to back it up.  

[manikyath]

1 minute ago, VagabondWraith said:

Well sure, my 1080 Ti pulls 320 OC but it has the performance to back it up.  

exactly

[done12many2]

19 minutes ago, MageTank said:

@TahoeDust Surprisingly great results for an AIO, I'll eat my words now. Even with a -500 offset, 4.2ghz of Prime95 with 8 threads is considerably hot, especially considering this is unsoldered unlike the X99 SKU's we saw. Was that a custom workload at a specific FFT length? Or a normal blend? Smaller FFT lengths run considerably hotter, so if that is indeed small, it's even more impressive than I initially gave it credit for.

 

I think once we start delidding these things, we're going to see that the seal/TIM job from the factory is a little better than what we saw in the 7700k.  Obviously, solder would have been best, but it looks like they at least reduced the die/IHS gapping previously observed in Kaby Lake.

 

Quote

4.7 at 1.22 is actually on par with what I expected, given the results I've seen from @done12many2. As usual, my only real gripe with these bigger chips is the memory latency. Copious amounts of bandwidth, but poor latency scaling across the increased channel count. If you have the free time, I'd love to help tame that latency a little. Wouldn't take much more than tRFC and tREFI tweaking to get it to around 60ns. If you have time to actually overclock beyond 3200, then 50-55ns would be feasible. From what I hear, memory overclocking has done quite a bit for done1's benching results, which tells me these high core-count chips love that extra bandwidth/lower latency. 

 

Sooo much room here to play with.  The latency / bandwidth on my 3600 kit paired with my 7900X weren't terrible with XMP set, but they damn sure got better once I started messing around with stuff.  Currently I'm at 101 GB/s + Read, 102 GB/s + Write, and ~85 GB/s Copy with latency at 53 to 55ns.  Latency is definitely not anywhere near z270 territory, but it's great for quad channel.  That BANDWIDTH though!  

[manikyath]

Just now, done12many2 said:

 

I think once we start delidding these things, we're going to see that the seal/TIM job from the factory is a little better than what we saw in the 7700k.  Obviously, solder would have been best, but it looks like they at least reduced the die/IHS gapping previously observed in Kaby Lake.

i seem to vaguely remember someone already delidded one, i dont remember specifics or results tho. been too busy with work and following the engineering insanity/madness that is threadripper to really bother with extreme OC on intel enthousiast platform.

[done12many2]

Just now, manikyath said:

i seem to vaguely remember someone already delidded one, i dont remember specifics or results tho. been too busy with work and following the engineering insanity/madness that is threadripper to really bother with extreme OC on intel enthousiast platform.

 

Yes, der8bauer dilidded a few of them already. 

 

I'm tracking Threadripper as well as I plan to build one of those also.  

[manikyath]

Just now, done12many2 said:

 

Yes, der8bauer dilidded a few of them already. 

 

I'm tracking Threadripper as well as I plan to build one of those also.  

i'm mostly keeping an eye on threadripper because if it is affordable enough one of our clients at work could literally replace their entire VM farm with two of those, mirrored, as failover  (said VM farm is aging a bit, needs replacement soon-ish either way)

[MageTank]

1 minute ago, manikyath said:

i seem to vaguely remember someone already delidded one, i dont remember specifics or results tho. been too busy with work and following the engineering insanity/madness that is threadripper to really bother with extreme OC on intel enthousiast platform.

I question your repeated need to mention threadripper in a thread that has nothing to do with threadripper. Even going out of your way to quote someone about delidding, just to bring up threadripper. I am certain that is going to wear thin on the nerves of others, especially given it has zero relevance here. 

[TahoeDust]

25 minutes ago, MageTank said:

@TahoeDust Surprisingly great results for an AIO, I'll eat my words now. Even with a -500 offset, 4.2ghz of Prime95 with 8 threads is considerably hot, especially considering this is unsoldered unlike the X99 SKU's we saw. Was that a custom workload at a specific FFT length? Or a normal blend? Smaller FFT lengths run considerably hotter, so if that is indeed small, it's even more impressive than I initially gave it credit for.

 

4.7 at 1.22 is actually on par with what I expected, given the results I've seen from @done12many2. As usual, my only real gripe with these bigger chips is the memory latency. Copious amounts of bandwidth, but poor latency scaling across the increased channel count. If you have the free time, I'd love to help tame that latency a little. Wouldn't take much more than tRFC and tREFI tweaking to get it to around 60ns. If you have time to actually overclock beyond 3200, then 50-55ns would be feasible. From what I hear, memory overclocking has done quite a bit for done1's benching results, which tells me these high core-count chips love that extra bandwidth/lower latency. 

Thanks man.  It is running Prime95 normal blend.  I'll give small TFT's a run soonish.

 

I would love some help dialing in the RAM overclock...it is not something I have a lot of experience with.  I'll shoot you a PM when I have time to mess with it.