SerDes are tiny so space probably wouldn't be much of a concern, going on what watercooled posted there must be two per CCD, one for each CCX, and they're directly connect to, what i imagine would be, one of multiple SerDes on the I/O die.
SerDes are tiny so space probably wouldn't be much of a concern, going on what watercooled posted there must be two per CCD, one for each CCX, and they're directly connect to, what i imagine would be, one of multiple SerDes on the I/O die.
Is it just me, or does ryzen master hog CPU time? Task manager shows it taking 10-20% of the CPU when I have it open in the background
This is interesting:
Further evidence that zen2 performance does vary depending on the rest of the system, but not by enough to worry over - the cheapest A320 board going was only 100 MHz down on a top end X570 board in a stress test. The X570 board might not have had PBO et all turned on, but it's still impressive (4 cheap CPU phases with no heatsink managing a 12 core CPU without major issue - it was well above the base clock, which is meant to be the worst case scenario)
DanceswithUnix (28-07-2019),Millennium (16-10-2019)
I thought those low end A320 boards were usually only rated for 65W! That is still power enough for a 3700X though, which is pretty awesome for a cheap board.
I haven't heard anything positive about PBO yet, it seems to just make things hotter for minimal performance gain. I think that is because the plain non-overclocking Precision Boost is very impressive so there isn't much left to squeeze out.
Edit: I looked it up, that Gigabyte motherboard really does support 105W cpus. Nice.
It is indeed. My current impression is that unless you can forsee actually needing (or using) PCIe4 inside of the next 3 or less years, the extra money for x570 is generally wasted. I suspect that next year the x670 or whatever it will be called will be a lot cooler running. Probably the B650 will have PCIe4 making really decent £100 ish PCIe4 motherboards available.
I've had a quick look at PCIe ratification dates and PCIe4 was ratified June 8th 2017, PCIe5 was ratified 29th May 2019 so I don't expect PCIe5 in consumer kit until summer 2021. Probably with DDR5 and a new socket for AMD as well.
To add to that, if you've already got an x470/B450 motherboard, there's very little reason to upgrade currently even if you do buy a Ryzen 3000 series CPU.
I'm still delaying upgrading my I5-3570k as although the Ryzen 3000 series is waaaaaaaaay faster, I'm not CPU limited enough yet.
EDIT: Just thought - overclockers may disagree with me. Take all of that above as completely ignoring overclocking capability.
"In a perfect world... spammers would get caught, go to jail, and share a cell with many men who have enlarged their penises, taken Viagra and are looking for a new relationship."
I just noticed - navi CUs take up less die area than GCN CUs (compare 5700XT with Mi60/R7 - 3.9 mm^2/CU, compared to 5.2 mm^2/CU for GCN). Uncore like the fixed function decode would be expected to be constant in size, and so make the R7 number better; the memory controller will vary though but that ought to be 1) proportional to CU count and 2) R7 is HBM anyway, so should have that off-boarded. I predict we'll see >11 CUs in the 4000 series APUs, which gets even better with how navi beats GCN on a per-CU basis in games
"In a perfect world... spammers would get caught, go to jail, and share a cell with many men who have enlarged their penises, taken Viagra and are looking for a new relationship."
That's pretty much my opinion on the matter too TBH. Gain a bit of e-peen if anything, and often totally destroy power efficiency in the process.
Enjoyed catching up on this discussion. Didn't watch the videos though (one playing in the background for audio).
A few questions please....
What does this 'chiplet' talk refer to?
How far is the 3700x overclocking?
Has anyone done a decent IPC / increased cache size comparison of clock for clock say R7 1700 14nm vs 3600x 7nm? Say as in running them both at 4GHz per thread?
Thanks in advance for the help here. I don't need to try and look into the detail of inter die and memory controller latency reductions I feel.
hexus trust : n(baby):n(lover):n(sky)|>P(Name)>>nopes
Be Careful on the Internet! I ran and tackled a drive by mining attack today. It's not designed to do anything than provide fake texts (say!)
Chiplets are the individual die that are used in the multi die layout in a multi chip module that AMD are using. A review of the 3x00 series will have diagrams and pictures of the two chiplets that make up a 3700X for example.
3700X for me doesn't seem even worth attempting overclocking. That tends to lock all cores at the overclocked frequency which is sometimes a tiny win, losing the stock power efficiency which I consider a big win. My chip can boost a single core or two up to those sorts of frequencies anyway, so for my usage where I want high clocks (code linking for example or lightly threaded games) the stock chips works fine for me.
As for IPC, is it very workload dependent. There are two big changes over earlier ryzen chips: the floating point units are better and the L3 cache is huge. The floating point units help crank through stuff like rendering, but I wanted the L3 cache for code compiling. For my programming use this chip is the stuff of legends, laughing in the face of the Xeons I'm used to using
Millennium (16-10-2019)
DancesWithUnix I hope you are well.
I was rather hoping someone would put me off spending more money. Sadly...
hexus trust : n(baby):n(lover):n(sky)|>P(Name)>>nopes
Be Careful on the Internet! I ran and tackled a drive by mining attack today. It's not designed to do anything than provide fake texts (say!)
DanceswithUnix (16-10-2019)
The performance of my Ryzen 5 3600 is great for the money, but I'm still a bit unsure about the voltage and temps... They always seem a bit too high, especially seeing as it's a 65W chip and at the lower end of the Ryzen 3000 series.
Voltages reach 1.450V when the cores hit 4.2GHz, and the voltages usually hover around 1.375V under load when all cores are being used (usually at 4.05GHz with all cores in use). I'm seeing temps of around 77C when gaming and that's also with a Noctua NH-U12A cooler which is one of the best, if not the best air cooler out there! Case cooling is also good. The motherboard I'm using is the Asrock X470 Taichi with the latest BIOS available (3.60).
VRM temps, chipset temps, drive temps, GPU temps and RAM temps are all fine. Even the side panel of the case gets pretty warm behind the CPU area after a bit of gaming (currently playing The Outer Worlds).
That seems a little high, maybe 3-5 degrees more than I would expect. Nothing to be too concerned about (After all it works like an electric heater in the winter right?) I wonder what thermal compound you used? You could perhaps have another stab at that with something different. However, having typed that, 1.375v on a 7nm chip seems a fair chunk of electrons to be throwing at it. It's just possible you ended up with a Ryzen that's on the hot side of normal.
Take the side panel off and see what the max gaming temp is and report back if you don't mind. I have a 1700 and even at 4ghz and 1.38v it wouldn't probably hit 77c gaming with an NH-D15. The D15 should be max 1-2 degrees cooler than your already excellent NH-U12A unit (I looked it up). I would expect, with both my CPU fans fitted, 72-4 degrees. I have neither gamed nor overclocked in some months though so I can't be sure, and we haven't accounted for ambient.
The 580 should produce less heat than my Vega too. I think there may be room for improvement in your 77 degrees.
hexus trust : n(baby):n(lover):n(sky)|>P(Name)>>nopes
Be Careful on the Internet! I ran and tackled a drive by mining attack today. It's not designed to do anything than provide fake texts (say!)
There are currently 6 users browsing this thread. (0 members and 6 guests)