AMD - Zen chitchat

[watercooled]

Has anyone measured "heatsink scaling" yet? Thermals seem to be increasingly important as time goes on.

I was hinting at the same in my last post. IIRC this was kinda a thing with the boost feature on Nvidia GPUs too, where testing them on an open-air testbed would allow them to sustain higher clocks than they would in a normal PC case.

Assuming the throttling starts at a fixed point on these CPUs, you could get an idea of whether a reviewed CPU is likely impacted based on max temperature readings if they're available. Like power readings though, it's important to know what application those temps were measured with as it really does make a difference.

I imagine this has a lot to do with Intel's TIM - you need to go to great lengths to make a substantial difference to temps when you're fighting against a high resistance path built in to the CPU. I think I've moaned about this before, but it's so bad that fan speed makes little difference on my cooler (Hyper 212) and the motherboards stock fan profiles are pointlessly loud - if the heatsink is still stone cold with the fan at 800rpm then ramping to to 2000rpm will achieve nothing but annoyance. And mine is only a 65W TDP 7700!

FWIW here's some numbers from running y-cruncher (heavy AVX load):

I'm not too convinced about that uncore power (I suspect it could be erroneous as it tends to drop under load and idles around the 10W mark) but even looking at IA core power, it's exceeding TDP. Anyway, temp spikes to high 80's in a matter of seconds, long before the heatsink will have began to warm up. I'm quite uncomfortable with the thought of a higher-clocked 6-core version i.e. 8700k!!

[kalniel]

87W package isn't too shabby, though I guess Intel are starting to use a loser TDP definition by that point.

My (new) old westmere hits it's rated 130W exactly (despite OC to 4ghz), and it's the northbridge temps that are high (remaining at 40-50C even when rest of processor is idle and cooler) - guessing this is roughly equivalent to uncore power in yours.

Perhaps westmere still has decent TIM?

[CAT-THE-FIFTH]

This is what my 69W TDP Xeon E3 1230 V2 performs like,with a Corsair H40 running a single fan,in a mini-ITX system(however,the case side was off as I was doing some maintenance on the system).

[badass]

I've deliberately disabled turbo on my i5-3570k because no matter what cooler is on it, the cores would hit 75-80 degrees. I'm sure it's more stable now - however it's left without reboots for weeks at a time and runs very demanding workloads occasionally. This may be my imagination of course.

The main point is that Intel take the mick cheaping out on expensive CPU's.

[Xlucine]

Perhaps westmere still has decent TIM?

Wasn't westmere soldered? IIRC haswell was the first TIM'd enthusiast processors (hence the updated devils canyon models with claimed better TIM, once people started noticing)

[kalniel]

Wasn't westmere soldered? IIRC haswell was the first TIM'd enthusiast processors (hence the updated devils canyon models with claimed better TIM, once people started noticing)

Yeah, that's what I meant

[kompukare]

CB did yet another article on Ryzen2 this time memory performance:
https://www.computerbase.de/2018-04/...peicher-tuning
At default memory speeds, the Intel i7-8700K is 11% faster for their game selection:

But with DDR4 3466 with optimised timings, that difference shrinks to 5%

Now that's mainly because the Intel doesn't benefit so much from the faster memory, but I guess this also shows where AMD have to concentrate on with Zen2: IMC and cache. Guess 7nm should allow them to add more cache but improving the IMC is probably way harder.

Wasn't westmere soldered? IIRC haswell was the first TIM'd enthusiast processors (hence the updated devils canyon models with claimed better TIM, once people started noticing)

Pretty sure it was Ivy Bridge and the Wiki confirms this:
https://en.wikipedia.org/wiki/Ivy_Br...n_overclocking
Since both Ivy Bridge and Haswell were 22nm. it makes sense that they would have timed the change from solder with a node shrink (this assumes that Intel had a genuine engineering reason to no longer use solder - although since their HEDT platform stuck with solder until Broadwell-E cost/stinginess still seems like the most plausible reason).

[watercooled]

87W package isn't too shabby, though I guess Intel are starting to use a loser TDP definition by that point.

My (new) old westmere hits it's rated 130W exactly (despite OC to 4ghz), and it's the northbridge temps that are high (remaining at 40-50C even when rest of processor is idle and cooler) - guessing this is roughly equivalent to uncore power in yours.

Current Intel CPUs seemingly aren't power limited to their TDP, especially under AVX workloads.

My point about uncore power is it stays around the 10W mark even when idle (and sometimes drops under load, weirdly) however an i3 6100 system shows more expected uncore results around a couple of watts, and less when idle. At first I thought it might have been something to do with PCIe link to the GPU, but it does seem a tad high even for that? I've still not pulled the GPU out to test my original theory but for now I just assume it's possibly incorrect.

But yeah, that TIM is inexcusably rubbish. Intel more-or-less admitted it with Devil's Canyon, a 'special' CPU which basically just had a less-rubbish thermal path and even managed to make it a marketing point to boot!

[CAT-THE-FIFTH]

Ivy Bridge used TIM as well.

[watercooled]

Ivy bridge was the first of the iSomething CPUs not to use solder IIRC?

[Corky34]

Now that's mainly because the Intel doesn't benefit so much from the faster memory, but I guess this also shows where AMD have to concentrate on with Zen2: IMC and cache. Guess 7nm should allow them to add more cache but improving the IMC is probably way harder.

Going on what Anandtech's testing shows on cache latency they probably won't be able to get much more out of the cache, there's probably some small gains they could get from the IMC but i think their slower there because of the way Zen is designed.

[CAT-THE-FIFTH]

Going on what Anandtech's testing shows on cache latency they probably won't be able to get much more out of the cache, there's probably some small gains they could get from the IMC but i think their slower there because of the way Zen is designed.

Games seem to show a larger IPC uplift of between 5% to 7% and the DAW benchmarks showed big improvements too,so it makes me wonder whether the next target is reducing memory latency:

https://techreport.com/r.x/2018_04_1...memlatency.png

[Corky34]

True but I'm unsure how much more they could get out of memory latency, with Zen the cores in each CCX don't have direct access to RAM, the memory controllers are not integrated into the cores themselves like they are with Intel, they're attached to the infinity fabric's (IF) scalable data fabric (SDF), it adds an extra step compared to Intel.

That's not to say one is better or worse than the other as it's horses for course but it seems AMD have picked most of the low hanging fruit in terms of RAM latency already.

[Xlucine]

True but I'm unsure how much more they could get out of memory latency, with Zen the cores in each CCX don't have direct access to RAM, the memory controllers are not integrated into the cores themselves like they are with Intel, they're attached to the infinity fabric's (IF) scalable data fabric (SDF), it adds an extra step compared to Intel.

That's not to say one is better or worse than the other as it's horses for course but it seems AMD have picked most of the low hanging fruit in terms of RAM latency already.

Err, what? That's news to me

[Blaineoliver]

https://en.wikichip.org/wiki/amd/infinity_fabric

It does seem like the memory goes via the infinity fabric, but then that makes sense if you think about the fact memory speed affects gaming more as per CCX traffic is sped up by faster ram.

[Biscuit]

Ivy bridge was the first of the iSomething CPUs not to use solder IIRC?

My 3570K can confirm this, all these years later and the TIM inside it is clearly dried up, hits high temps crazy fast.