R5 3600 review leak linked to at TPU

I wonder what speed memory they used as the latency and memory speeds seem really bad... Also compared to many other leaks, the performance seems MUCH lower than I'd expect. Guessing they used slow ram for this?
 
The mem speeds listed on the QVL Lists for the Gigabyte boards are (imho) shockingly low considering we've been told about 4000MHz+ works fine for nearly the last 6 months.
 
chameleoneel said:
I'm guessing the memory issues are all early BIOS problems.
Click to expand...

Yeah, last month for grins I tossed a pair of Corsair 32gb lpx into a b450 1200 combo after bios updates and XMP worked fine.

Beat on it for a weekend and it was totally stable at XMP. I feel like I wasted a bunch of benching time on a box I use for a Kubernetes lab.

I threw a 1070 into it and played Civ6 and Blackout ad nauseum until I couldn't deal with low framerate anymore.
 
when are the real reviews going to be out. I'm down to go buy this at micro center on sunday but I need some reviews.
 
chameleoneel said:
I'm guessing the memory issues are all early BIOS problems.
Click to expand...

Like every benchmark ever is showing low write speeds. Either that's a feature (tm) or it's something like you say, a BIOS issue that affects everything.

I've heard that the infinity fabric speed will be a VERY big factor, and this OC will be as important as core speeds for some workloads.

It'll be an interesting week next week. These definitely don't seem to be the kind of chip you throw a big heatsink on, crank the clock speeds and are good. But on the other hand, it looks like a lazy enthusiast will get 90% of the way there with a HS + fast ram + enable PBO and be done.
 
Concillian said:
Like every benchmark ever is showing low write speeds. Either that's a feature (tm) or it's something like you say, a BIOS issue that affects everything.

I've heard that the infinity fabric speed will be a VERY big factor, and this OC will be as important as core speeds for some workloads.

It'll be an interesting week next week. These definitely don't seem to be the kind of chip you throw a big heatsink on, crank the clock speeds and are good. But on the other hand, it looks like a lazy enthusiast will get 90% of the way there with a HS + fast ram + enable PBO and be done.
Click to expand...
I don't think its gonna be a big deal, at all.

This leaked review used DDR 3200, which is upper standard spec by AMD. And with the RAM speed, it trades with Skylake, embarrasses Zen 1, and only loses to Zen 1+ in multi-core. Performance will be even better, with faster RAM.

The latency and write speed issues seem like a BIOS issue to me.

And the whole point of doubling caches and improving prediction units, was to lower latency and rely less on system ram. So yeah, seems like something is wrong which will be fixed.

And it seems like you contradicted yourself there. Because with decent heatsinks included in the box and Precision boost: putting on the heatsink, turning on precision boost, and enabling XMP on your RAM is all you'll have to do be good. And enthusiast can tweak RAM timings and/or speeds from there. But otherwise, auto overclocking is so good now, hand tweaking processors is kinda done.
 
chameleoneel said:
The latency and write speed issues seem like a BIOS issue to me.
Click to expand...

Actual RAM latency should be worse, as that's how AMD designed it, but bandwidth should be ballpark or better.

chameleoneel said:
But otherwise, auto overclocking is so good now, hand tweaking processors is kinda done.
Click to expand...

This sentiment seems born out of AMD and Intel releasing CPUs that are already clocked almost to their limit. It's true now, and in a way that's nice, but we have no reason to believe that hand tweaking will stay more or less pointless.

And by 'nice', I wouldn't mind the performance versions (X, K) of a CPU to simply be, 'here's n-number of cores, they'll run as fast as they can', and faster speeds are obtained more through robust power delivery and cooling.
 
chameleoneel said:
And it seems like you contradicted yourself there. Because with decent heatsinks included in the box and Precision boost: putting on the heatsink, turning on precision boost, and enabling XMP on your RAM is all you'll have to do be good. And enthusiast can tweak RAM timings and/or speeds from there. But otherwise, auto overclocking is so good now, hand tweaking processors is kinda done.
Click to expand...

Yeah, I didn't do a good job of conveying what I meant.

So yeah a normal enthusiast just enables PBO and makes sure he's at DDR4 3600 or DDR4 3733 with a 1:1 ratio on the IF perhaps even with the stock cooler (for 3700x and up) perhaps with a bigger cooler.

But for max tweaking, it's different. Since even Nehalem, but esp. since Sandy it's been so insensitive to anything other than just CPU clocks. But max tweaking for Ryzen 3000 seems to be a lot more than that. It doesn't seem to be very thermally limited, but will want OCing of Infinity Fabric (based on a comment I saw from someone under NDA at OCUK) And Ryzen 2000 saw significant benefit in painful tweaking of sub-timings.... So there may be quite a divergence for the more serious OCer.
 
Concillian said:
Like every benchmark ever is showing low write speeds. Either that's a feature (tm) or it's something like you say, a BIOS issue that affects everything.

I've heard that the infinity fabric speed will be a VERY big factor, and this OC will be as important as core speeds for some workloads.

It'll be an interesting week next week. These definitely don't seem to be the kind of chip you throw a big heatsink on, crank the clock speeds and are good. But on the other hand, it looks like a lazy enthusiast will get 90% of the way there with a HS + fast ram + enable PBO and be done.
Click to expand...

The big unknown with Infinity Fabric is exactly why i'm waiting for benchmarks before deciding on a 3800X (single 8 core chiplet) or the 3900X (dual 6 core chiplets) I'm curious to see how the 3900x handles workloads that can fully utilize 8 cores but might have diminishing, or no returns scaling past that.
 
Concillian said:
So there may be quite a divergence for the more serious OCer.
Click to expand...

Best case is that there isn't a divergence- that getting the fastest RAM you can, using a board with great power delivery and using a good cooler would get you within say 95% of max performance out of the box.

I know people enjoy tweaking, but just being able to slap something together with a modicum of forethought and knowing that you're getting the performance that you're paying for is a win for enthusiasts too.
 
RamonGTP said:
The big unknown with Infinity Fabric is exactly why i'm waiting for benchmarks before deciding on a 3800X (single 8 core chiplet) or the 3900X (dual 6 core chiplets) I'm curious to see how the 3900x handles workloads that can fully utilize 8 cores but might have diminishing, or no returns scaling past that.
Click to expand...

I'm waiting for Puget systems to publish their tests.

They're usually pretty solid about running builds into known real world resource limits.
 
You must log in or register to reply here.
Back
Top