Der8auer hits 5.0GHz on i9-7900X with AIO water cooling

[PCGuy_5960]

Just now, XenosTech said:

only if you but the 6 core lol and it's still less lanes than ryzen

Nope. 6 and 8 core have 28 CPU lanes and the 10+ core have 44 CPU lanes. All of them have another 24 lanes from the chipset, so you get 52 lanes in total with a 6 or 8 Core CPU and 68 lanes in total with a 10+ core CPU both of which are more than Ryzen which has 16 CPU lanes and 8 chipset lanes (24 lanes total)

[XenosTech]

Just now, PCGuy_5960 said:

Nope. 6 and 8 core have 28 CPU lanes and the 10+ core have 44 CPU lanes. All of them have another 24 lanes from the chipset, so you get 52 lanes in total with a 6 or 8 Core CPU and 68 lanes in total with a 10+ core CPU both of which are more than Ryzen which has 16 CPU lanes and 8 chipset lanes (24 lanes total)

Quoted it too fast

[cj09beira]

Just now, djdwosk97 said:

It will depend on the specific program, but in theory it would put total cpu power of a skylake-x octacore on par with a threadripper decacore, and a skylake 12 core on par with a threadripper 15 core.

 

As I said that doesn't necessarily translate directly to performance in that some programs will benefit more from the single for performance while others just want cores and don't care too much about single threaded performance.

the thing is that amd say's their processors scale better with more cores and i wanna see if its true

[done12many2]

1 minute ago, MageTank said:

5ghz without AVX, for only an hour? Yeah... that's totally going to be stable, lol. 

 

That was me being sarcastic. 

 

Come on Der8auer, you are better than this.

 

I guess the with/without AVX argument is becoming less and less when one of the two competitors doesn't even give you the choice.  

[done12many2]

1 minute ago, cj09beira said:

the thing is that amd say's their processors scale better with more cores and i wanna see if its true

 

Scale better how?  In clockspeed or performance?

[cj09beira]

Just now, done12many2 said:

 

Scale better how?  In clockspeed or performance?

its related to how efficiently you can spread the task between threads, also how well can each core talk to io/ memory/cache 

[MageTank]

Just now, done12many2 said:

 

I guess the with/without AVX argument is becoming less and less when one of the two competitors doesn't even give you the choice.  

Hm? AMD supports AVX2 just fine, it's AVX-512 that it's missing support for: 

 

You can also turn it off on AMD as you can with Intel. 

[PCGuy_5960]

2 minutes ago, done12many2 said:

 

Scale better how?  In clockspeed or performance?

I think he meant that their Hyperthreading is better...

[done12many2]

5 minutes ago, djdwosk97 said:

It will depend on the specific program, but in theory it would put total cpu power of a skylake-x octacore on par with a threadripper decacore, and a skylake 12 core on par with a threadripper 15 core.

 

As I said that doesn't necessarily translate directly to performance in that some programs will benefit more from the single for performance while others just want cores and don't care too much about single threaded performance.

 

Neither the Intel x299 or AMD x399 will be used by 90% of the people posting here.  We are just hear to talk shit and spit ball our versions of the truth.  

[done12many2]

1 minute ago, MageTank said:

Hm? AMD supports AVX2 just fine, it's AVX-512 that it's missing support for: 

 

You can also turn it off on AMD as you can with Intel. 

 

It's crippled.  Pretty well known bud.  

[PCGuy_5960]

Just now, MageTank said:

AMD "supports" AVX2 just fine,

It does "support" it, but Ryzen's AVX performance is shit

[MageTank]

Just now, done12many2 said:

 

It's crippled.  Pretty well known bud.  

 

Just now, PCGuy_5960 said:

It does "support" it, but Ryzen's AVX performance is shit

Correction. It's AVX1 is weak. It's AVX 2 is just as strong as Intel's. Zen can do 8x 128-bit op/clock, or 8x 256-bit op/clock. Intel can do 16x 128, and 8x 256 (technically 32x 128 and 16x 256 on these new AVX-512 SKU's, but yeah...)

 

[done12many2]

4 minutes ago, MageTank said:

 

Correction. It's AVX1 is weak. It's AVX 2 is just as strong as Intel's. Zen can do 8x 128-bit op/clock, or 8x 256-bit op/clock. Intel can do 16x 128, and 8x 256 (technically 32x 128 and 16x 256 on these new AVX-512 SKU's, but yeah...)

 

 

Actually, AVX2 is weaker as well.  Run some Handbrake with H.265 encoding and you'll see what I mean.  HWBOTx265 benchmark shows the same thing.  Almost half as fast as Intel's AVX2.  Actually, it's not that bad, but it's worse for sure. 

 

Thanks for the breakdown though. 

[MageTank]

3 minutes ago, done12many2 said:

 

Actually, AVX2 is weaker as well.  Run some Handbrake with H.265 encoding and you'll see what I mean.  HWBOTx265 benchmark shows the same thing.  Almost half as fast as Intel's AVX2.

 

Thanks for the breakdown though. 

I've seen the benches (and ran them myself). I know I am sounding like a broken record at this point, but it's Ryzens IMC. AVX is extremely dependent on memory bandwidth/latency. Ryzen's IMC is incredibly weak, with lower bandwidth (at identical clock speeds to Intel) and abysmal latency. You can test this yourself by loading JEDEC SPD profiles, vs your XMP. Look at Linpack's FLOP's improve, or Prime95's duration to complete an FFT length decrease. It scales quite well, as memory is a heavy bottleneck for AVX at the moment. As much as I even hate to admit it, it was the one thing Patrick was absolutely correct on, lol.

 

EDIT: You also know this, but i'll warn others before they test this. AVX + fast memory = even hotter CPU. Be very careful when overclocking your ram to test AVX stress tests, as it won't just be a couple degrees difference. I've seen upwards of 8-12C difference when tinkering with tertiary timings. Specific timings to be mindful about are tRDWR_SG, DG, DD, and DR. This timing has a dramatic impact on AVX related thermals. The tighter it is, the hotter you run. The looser it is, the cooler you run, at the cost of AVX performance. If AMD gave us access to those specific tertiary registers, I'd bet my entire rig that I could make Ryzen's AVX2 match Intels. We've seen it's layout, we know it's capable of the exact same amount of op/clock cycles, we just can't do anything about it with that extremely crippled IMC locking us down.

[done12many2]

1 minute ago, MageTank said:

I've seen the benches (and ran them myself). I know I am sounding like a broken record at this point, but it's Ryzens IMC. AVX is extremely dependent on memory bandwidth/latency. Ryzen's IMC is incredibly weak, with lower bandwidth (at identical clock speeds to Intel) and abysmal latency. You can test this yourself by loading JEDEC SPD profiles, vs your XMP. Look at Linpack's FLOP's improve, or Prime95's duration to complete an FFT length decrease. It scales quite well, as memory is a heavy bottleneck for AVX at the moment. As much as I even hate to admit it, it was the one thing Patrick was absolutely correct on, lol.

 

Thanks for sharing that.  Sounds like any other task, memory dependent to feed the fabric.  I'm doing a Threadripper build to so you better be right.  

 

Either way, I'm going to enjoy some AVX-512, but I bet it's going to get nice and toasty.  Anything that runs that efficiently is going to get hot.   

[cj09beira]

3 minutes ago, MageTank said:

I've seen the benches (and ran them myself). I know I am sounding like a broken record at this point, but it's Ryzens IMC. AVX is extremely dependent on memory bandwidth/latency. Ryzen's IMC is incredibly weak, with lower bandwidth (at identical clock speeds to Intel) and abysmal latency. You can test this yourself by loading JEDEC SPD profiles, vs your XMP. Look at Linpack's FLOP's improve, or Prime95's duration to complete an FFT length decrease. It scales quite well, as memory is a heavy bottleneck for AVX at the moment. As much as I even hate to admit it, it was the one thing Patrick was absolutely correct on, lol.

did you try with the new agesa ? (if its out for your motherboard)

[done12many2]

1 minute ago, cj09beira said:

did you try with the new agesa ? (if its out for your motherboard)

 

Wait what?  @MageTank you did a Ryzen build already?

[MageTank]

2 minutes ago, done12many2 said:

 

Thanks for sharing that.  Sounds like any other task, memory dependent to feed the fabric.  I'm doing a Threadripper build to so you better be right.  

 

Either way, I'm going to enjoy some AVX-512, but I bet it's going to get nice and toasty.  Anything that runs that efficiently is going to get hot.   

Preliminary tests seem to show that I am, but I won't say for certain until I know it's true. I can say that you do see a dramatic difference in AVX speed just by loading an overclocked XMP over stock JEDEC. Again, you can test this right now with Linpack (AVX, not MKL version, no need to burn your house down). 

 

2 minutes ago, cj09beira said:

did you try with the new agesa ? (if its out for your motherboard)

It's not my system, am currently using a friends Ryzen test bench, but I'll ask for his permission to flash to it to test. He isn't like me, and doesn't care about overclocking, as he just wants a cheap editing rig. 

 

Just now, done12many2 said:

 

Wait what?  @MageTank you did a Ryzen build already?

Nah, making one for a friend of mine. He also didn't pick a good set of ram. Fighting these IC's has been the bane of my existence. Talk about brute force to make 29xx happen, lol.

[done12many2]

 

Just now, MageTank said:

Nah, making one for a friend of mine. He also didn't pick a good set of ram. Fighting these IC's has been the bane of my existence. Talk about brute force to make 29xx happen, lol.

 

That latency though.  

 

I've been following Jpmboy on OCN in the Ryzen memory thread and latency is pretty bad.  Bandwidth seems okay, but the latency on Ryzen's dual channel is way higher then the Intel's current dual channel.  It's even higher then the latency on my x99 quad channel.

 

Does latency not impact performance on Zen as it does Intel?

[cj09beira]

8 minutes ago, MageTank said:

Preliminary tests seem to show that I am, but I won't say for certain until I know it's true. I can say that you do see a dramatic difference in AVX speed just by loading an overclocked XMP over stock JEDEC. Again, you can test this right now with Linpack (AVX, not MKL version, no need to burn your house down). 

 

It's not my system, am currently using a friends Ryzen test bench, but I'll ask for his permission to flash to it to test. He isn't like me, and doesn't care about overclocking, as he just wants a cheap editing rig. 

 

Nah, making one for a friend of mine. He also didn't pick a good set of ram. Fighting these IC's has been the bane of my existence. Talk about brute force to make 29xx happen, lol.

then the new agesa should help as it unlocks all the mem timings (29 i think)

[Sauron]

1 hour ago, done12many2 said:

You do realize that he said this in the video, right?

 

We've become a community on auto pilot without listening.  

I didn't watch the video, I read the summary in the post which gave me all the info I needed to form an opinion. Nothing I said was incorrect or misleading, so why does it matter? The worst that happened is that I repeated a concept that was already (barely) touched upon by the video.

[MageTank]

3 minutes ago, done12many2 said:

 

 

That latency though.  

 

I've been following Jpmboy on OCN in the Ryzen memory thread and latency is pretty bad.  Bandwidth seems okay, but the latency on Ryzen's dual channel is way higher then the Intel's current dual channel.  It's even higher then the latency on my x99 quad channel.

 

Does latency not impact performance on Zen as it does Intel?

We don't know... because good latency has been rather impossible to achieve, lol. I've been struggling to achieve sub-50ns in Aida64, and I can't obtain it. For the record: my Intel rig is currently at 36.7ns. You yourself achieved sub-40ns I believe, which is extremely impressive. Ryzen just cannot get close at the moment.

 

1 minute ago, cj09beira said:

then the new agesa should help as it unlocks all the mem timings (29 i think)

I'll have to look into it. I imagine the C7H already has it, so I'll see what we get access to. If it's only 29 registers, I highly doubt RTL/IO-L will be unlocked. We were already missing secondary timings, and there are at least 7-9 of those to worry about. 

[cj09beira]

3 minutes ago, MageTank said:

We don't know... because good latency has been rather impossible to achieve, lol. I've been struggling to achieve sub-50ns in Aida64, and I can't obtain it. For the record: my Intel rig is currently at 36.7ns. You yourself achieved sub-40ns I believe, which is extremely impressive. Ryzen just cannot get close at the moment.

 

I'll have to look into it. I imagine the C7H already has it, so I'll see what we get access to. If it's only 29 registers, I highly doubt RTL/IO-L will be unlocked. We were already missing secondary timings, and there are at least 7-9 of those to worry about. 

just went and checked they added 26 new timings

[done12many2]

4 minutes ago, MageTank said:

We don't know... because good latency has been rather impossible to achieve, lol. I've been struggling to achieve sub-50ns in Aida64, and I can't obtain it. For the record: my Intel rig is currently at 36.7ns. You yourself achieved sub-40ns I believe, which is extremely impressive. Ryzen just cannot get close at the moment.

 

Yeah, I'm at 38ns on z270 dual channel and 47ns on x99 quad channel breaking 85 GB/s in bandwidth.    Not bad for some "3200" sticks.  

 

I bet the IMCs on Skylake-X are going to be pretty sweat considering that Intel was willing to bump the supported base speed.  

[Red Hardware]

5 hours ago, spaniel53 said:

 

He hit 4.8GHz without delidding! Only 200MHz gain after delidding. Looks like the excess adhesive issue has been resolved. Very exciting results. If these results are repeatable, people will likely be changing their tunes about Intel. I can't wait to hear how far the 6/8-core models can be pushed.

Hi

It's probably a cherry-picked processor with lowest TDP and best bins that they could find . It's unlikely that others would be able to get 4.8 or 5 without LN2 out of it.

And it doesn't change any fact about their horrible X299 platform and pricing .

Some people mention that Intel has better IPC than AMD so it has a better single core performance, the problem is that anybody who is going to buy one of these and use X299 or X399 is looking into multi core/thread workloads that can benefit from all of these extra cores so talking about single core performance doesn't make sense .