AMD RyZen CPU Architecture for 2017

Cooling = clocks with Ryzen.
On AMD Spire RGB I couldn't get any reasonable stability above 3.9GHz even with 1.4V. Water AIO enabled easy 4GHz and today I will find out if not more than that.

Looks like I'm going for water AIO next then, I'm currently running 3.9GHz at 1.33 volts with the Wraith Spire :)
 
Looks like I'm going for water AIO next then, I'm currently running 3.9GHz at 1.33 volts with the Wraith Spire :)
the Wraith Spire is an all-purpose heatsink and my CPU was bundled with it, I am pretty happy with it. But for certain speeds and faster CPUs it shall run short at some point. Still, it is a very good solution to begin because it works like a charm if you don't push the CPU much.

My Ryzen 1500X is set at 3.7GHz...could it be reasonable to run it 3,9-4.0GHz with the Wraith Spire?
 
As I said in another thread I set the max voltage of my CPU to 1.25V and set it to 3,8GHz without XFR (didn't want to go to 3,9GHz though it is doable with stock heatskin, the Wraith Spire). AMD say that for Ryzen 5 a max voltage of 1.45V is doable but not recommendable because it is going to shorten the lifetime of the CPU, although 1.35V is perfectly fine for a 24 hours 365 days operation.

These are the results (not bad):




My sweet point is 3,875GHz which I also tried and that's the frequency that I have now. Temps are around 62ºC with this configuration.
 
the Wraith Spire is an all-purpose heatsink and my CPU was bundled with it, I am pretty happy with it. But for certain speeds and faster CPUs it shall run short at some point. Still, it is a very good solution to begin because it works like a charm if you don't push the CPU much.

My Ryzen 1500X is set at 3.7GHz...could it be reasonable to run it 3,9-4.0GHz with the Wraith Spire?

The Wraith Spire can provide enough cooling for up to 1.35 volts imo. It's not ideal but it's not bad either. Running prime95 at 1.33volts on a 1700 at 3.9GHz I see a maximum of 73C, If i drop down to 3.8 and 1.28 volts the temps drop to 66C. Under gaming load it sits around mid 50s.
 
Last edited:
The Wraith Spire can provide enough cooling for up to 1.35 volts imo. It's not ideal but it's not bad either. Running prime95 at 1.33volts on a 1700 at 3.9GHz I see a maximum of 73C, If i drop down to 3.8 and 1.28 volts the temps drop to 66C. Under gaming load it sits around mid 50s.
that sounds good to me, because very few software is going to use the 16 threads of your 1700.

Which program do you use to measure the temps of your 1700?

On a different note, I settled at 1.25V and 3,875GHz and I was using CPU-Z and randomly got this. The 9976 is the closest to the most powerful i7 processor -9993 benchmark-. I didn't upload the values 'cos it wasn't my intention to share them, but I got a screen capture. Thing is that at 3,8GHz I got 9700+ and also at 3,875GHz I get similar, but slightly better results, rarely achieving 9800, not to mention 9900.

cpuz.png
 
Implying there are bugs or lackthereof indicating it doesn't really have any?
 
Intel consumer chips (since Nehalem) have had a big shared LLC for all cores.
But its split into 2MB slices per-core & clocked with the core.
I've never been able to understand how that didn't produce all sorts of performance affecting variable latencies.
 
Isn't the discussion about achievable clocks and voltages and coolers off-topic? Please move it somewhere else

How is it off-topic? At the very least it shows how the architecture scales with voltage, and it also helps current Ryzen owners to fine tune their system.
 
Ok , maybe Its me that considers this CPU arch threads similar to the GPU ones. There you have a thread for the more theoretical architectural discussion and additional threads to discuss the products themselves.

For me at least, it's annoying to find replies that are relevant mostly to owners of the cpu or people intending to buy it. Its ok if I'm the only one who feels this way :)

On the contrary, steppings and errata , or news or predictions about Epyc & Threadripper, they definetly belong here in my view
 
Last edited:
But its split into 2MB slices per-core & clocked with the core.
I've never been able to understand how that didn't produce all sorts of performance affecting variable latencies.

Because that's not how it works. All cores access all slices evenly. Every cache line is mapped to a single slice depending on it's physical address and can only exist in that slice. And access to different slices have variable latencies, but the added latency is only one cycle (!) per hop, so it doesn't matter much.
 
It seems that the inter-socket bandwidth is close to the scenario I was guesstimating earlier for GMI versus xGMI. It's almost twice the 42 GB/s of one chip's GMI links. xGMI is either slower or has some extra overhead versus the on-package links.
 
ServeTheHome have released some limited and preliminary power consumption figures:
https://www.servethehome.com/amd-epyc-7601-dual-socket-early-power-consumption-observations/
We were surprised to see how low the actual power consumption is on the platform. Running an AVX2 workload we were expecting much higher power consumption but at under 500w for 128 threads, this is excellent.
...
The AMD EPYC platform is still seeing major updates to BIOS for power and performance which is why we are calling these preliminary results. At the same time, we are already seeing some impressive figures.
 
So it is actually 4* Ryzen dies on one MCM :cool:
epyc_tech_day_first_session_for_press_and_analysts_06_19_2017-page-015.jpg

Somehow I got the impression there would be a 16 core die & this would be 2* that.
 
So it is actually 4* Ryzen dies on one MCM :cool:

Somehow I got the impression there would be a 16 core die & this would be 2* that.
Nope, there's only two Zen-dies so far, Zeppelin which is used in Ryzen, Threadripper and Epyc and whateveritscalled that's used in Raven Ridge with the iGPU
 
Has AMD made any clarifications on those Epyc prices?
I mean, they don't make any sense as they are presented in the comparisons.
Are the 2P prices supposed to be per processor or total?
If per processor, why is Epyc 7601 at >$4000 when AMD states in a press release that it's priced at $2100? And it's compared to a Xeon priced at nearly $5k per CPU?
If per system, why are the lowend 7281 and 7251 ">$600" and ">$400" in both 1P and 2P comparisons?
 
Back
Top