New The PC Build Thread

[Neoprod]

I have a sneaking suspicion we may be in for some minor improvements when we get the new chipset. But overall a fairly lacklustre release thus far.

Oh there's a good chance of BIOS updates improving things - there were tons of AGESA updates for the 3000 and 5000 series CPUs. Lots to do with memory support but also boost behaviour - not saying that will happen here but it's soething they've changed before post-launch.

 

[Fulcrum29]

Oh there's a good chance of BIOS updates improving things - there were tons of AGESA updates for the 3000 and 5000 series CPUs. Lots to do with memory support but also boost behaviour - not saying that will happen here but it's soething they've changed before post-launch.

I am not buying a board until the 800 series drops. The go-to-market date keeps on moving up... Ugh. At least by that time we may have memory with its own client clock driver, but it will be expensive, so most likely not within my price range. It is still to be seen how it will behave with current gen CPUs.

Phoronix had an interesting benchmark.

AMD Ryzen 7 9700X Performance With DDR5-8000 Using Corsair Vengeance 2x16GB DDR5-8000 DIMMs Review - Phoronix

www.phoronix.com

Latencies do have an impact, but see the numbers in the benchmarks. Theoretically, memory modules with a CCD should reduce latencies. Time will tell... but this could be the biggest game changer in this decade.

G.Skill also announced DDR-6000 CL28 modules.

 

[Neoprod]

I am not buying a board until the 800 series drops. The go-to-market date keeps on moving up... Ugh. At least by that time we may have memory with its own client clock driver, but it will be expensive, so most likely not within my price range. It is still to be seen how it will behave with current gen CPUs.

Phoronix had an interesting benchmark.

AMD Ryzen 7 9700X Performance With DDR5-8000 Using Corsair Vengeance 2x16GB DDR5-8000 DIMMs Review - Phoronix

www.phoronix.com

Latencies do have an impact, but see the numbers in the benchmarks. Theoretically, memory modules with a CCD should reduce latencies. Time will tell... but this could be the biggest game changer in this decade.

G.Skill also announced DDR-6000 CL28 modules.

That looks cool but also kind of normal. I've done some tests before with memory scaling on my system and gotten similar results -





That's about 19% better read, 20% better write, 18% better copy, 12% better latency. That's been the differential between green stick memory and memory with racing stripes since forever really. Which kind of leads on to my point.

Real world that increased bandwidth and reduced latency hasn't mattered much. It's like a couple percent here and there outside of synthetic benchmarks with some applications (especially hevc \ x264 encoding and rendering) not caring at all. I'd be interested to see if that changes and why but my gut tells me it won't because code isn't written with the expectation that that memory bandwidth is available.

Memory overclocking is interesting but it's not rewarding from a daily use point of view - just kinda neat to see "number go up"

 

[Fulcrum29]

https://twitter.com/x/status/1821659169352773801

Fokkit. RGB OFF!

 

[Fulcrum29]

That looks cool but also kind of normal. I've done some tests before with memory scaling on my system and gotten similar results -

View attachment 1749893

View attachment 1749895

That's about 19% better read, 20% better write, 18% better copy, 12% better latency. That's been the differential between green stick memory and memory with racing stripes since forever really. Which kind of leads on to my point.

Real world that increased bandwidth and reduced latency hasn't mattered much. It's like a couple percent here and there outside of synthetic benchmarks with some applications (especially hevc \ x264 encoding and rendering) not caring at all. I'd be interested to see if that changes and why but my gut tells me it won't because code isn't written with the expectation that that memory bandwidth is available.

Memory overclocking is interesting but it's not rewarding from a daily use point of view - just kinda neat to see "number go up"

Sure, but it is still dependent on the interconnect, and I am making a correction to my previous post. The memory clock driver is not a CCD, but a CKD. My brain isn't always working. This will be paired with the registered clock driver (RCD). As technology advances, more onboarding is done on modules, increasing latencies. For normal consumers it is not a big deal since we have small workloads compared to compute nodes; however, we still have interconnect between logic blocks. A CKD will reduce the compute task by improving the signal strength by re-timing (the signal) and distributing the clock sent by the processor, essentially managing the clock to eliminate noise. Fewer errors. Jitter mitigated. You could get much better timings without reducing the clock because the syncing is done on chip, allowing improved interconnectivity. Frequencies be damned.

I'm no engineer, but I have done some reading on the standards as published by JEDEC, and though more applicable to compute nodes, and as you pointed out, "some applications", quicker data access, and its acceleration, could positively change the relationship between the CPU and RAM. The question is, how positive?

Potentially, cache could also be more optimally handled. Process synchronisation will be improved. Overall reduced voltage. The CPU not having to send multiple clocks to the DIMM (or SO-DIMM) will also reduce the CPU's power consumption and reduce pin usage. The CPU package could be made more compact as a result.

Generally CAMM is the better RAM improvement, in the current, and CAMM could also have a CKD, however CAMM unlike SO-DIMM will reduce the traces between the module and the CPU. Again, higher speeds, reduced power consumption, errors mitigated.

PCI-e is also evolving. Improved retimers (and switches) using less power, reducing noise. There could be huge gains made in this decade. Moving towards ecosystem synchronous harmony.

Yeah, programming to leverage the above, optimally, will be needed. CAMM is already adopted, channel loss between the CPU and that endpoint will be reduced. How this will translate into gains, only reviews will tell, but I do think that CKD will show promise. Basically everything on the PCI-e roadmap is addressed by the CKD at a RAM level.

 

[kolaval]

Interesting to see how laptops are competitively priced now compared to desktops.

Yep, I'm never going back to desktop.
Main thing to check the laptop for:
GPU watts, more is better up to about 100w.
You don't want e.g. the gigabyte G5 2023 model with a 4060 at 65w.

 

[Lupus]

Interesting to see how laptops are competitively priced now compared to desktops. I remember back when laptops had way worse performance for the same price. You can pick up an RTX 4070 laptop for around the same price once adding in SSD, PSU and screen costs to the desktop (desktop will be over R30k when adding those) and the performance is only around 10% worse for the laptop.

eg. A RTX 4070 WootBook with 1TB SSD and 32GB RAM will cost R30.5k. Very similar price to a desktop of equal performance when adding in all the extras (screen, keyboard too). Desktop might even come out more expensive which is insane.

WootBook Ultra Y16 AMD Ryzen 7 8845HS 5.10GHz Boost 8-Core 16" WQXGA (2560x1600) 240Hz NVIDIA GeForce RTX 4070 8GB GDDR6 Black Laptop

Buy WootBook Ultra Y16 AMD Ryzen 7 8845HS 5.10GHz Boost 8-Core 16" WQXGA (2560x1600) 240Hz NVIDIA GeForce RTX 4070 8GB GDDR6 Black Laptop at Wootware with fast shipping & superb service.

www.wootware.co.za

But the 4070 on the laptop isn't the same as the one on the desktop as it's power constrained

 

[Neoprod]

But the 4070 on the laptop isn't the same as the one on the desktop as it's power constrained

The laptop version also has less of everything (1280 fewer shaders, 40 fewer texture mapping units, 16 fewer raster pipelines) than the desktop part. Less of everything operating at a lower clock speed - it's probably more like a desktop 3060 or maybe 3060 Ti.

 

[Lupus]

The laptop version also has less of everything (1280 fewer shaders, 40 fewer texture mapping units, 16 fewer raster pipelines) than the desktop part. Less of everything operating at a lower clock speed - it's probably more like a desktop 3060 or maybe 3060 Ti.

So it would be more prudent to get a 4060 laptop as that's almost the desktop variant

 

[Neoprod]

So it would be more prudent to get a 4060 laptop as that's almost the desktop variant

I guess...I only checked on the 4070 because the other post seemed impossible to me. I'll take your word for it that the mobile and desktop 4060 are the same apart from power limits.

Then again the desktop 4060 is a POS

 

[Lupus]

I guess...I only checked on the 4070 because the other post seemed impossible to me. I'll take your word for it that the mobile and desktop 4060 are the same apart from power limits.

Then again the desktop 4060 is a POS

Actually the 4060 isn't that bad, it got a bad rap on release but it's really not as awful as originally said.
But due to its low power consumption most laptops almost have full blown 4060s in them plus they are not badly priced, found them for 21k

 

[Neoprod]

Actually the 4060 isn't that bad, it got a bad rap on release but it's really not as awful as originally said.
But due to its low power consumption most laptops almost have full blown 4060s in them plus they are not badly priced, found them for 21k

It's slower than a 3060 Ti and that's not supposed to happen - I can see why people were annoyed.

Then again, AMD just prices their stuff a little below Nvidia's so it's not like there are better options generation for generation.

Meh.

 

[Lupus]

It's slower than a 3060 Ti and that's not supposed to happen - I can see why people were annoyed.

Then again, AMD just prices their stuff a little below Nvidia's so it's not like there are better options generation for generation.

Meh.

The 90w difference in power consumption is probably why the 4060 is slower. Plus the 4060 does have some newer features, it's like AMD now coming out with the newer cpus that arent much better but use a lot less power.

 

[UrBaN963]

The 4060 is a terrible product. New release being slower than the old one is always bad. Deliver that product at an inflated price and it's even worse.

 

[Lupus]

The 4060 is a terrible product. New release being slower than the old one is always bad. Deliver that product at an inflated price and it's even worse.

The power consumption is far better on the 4060 vs the 3060ti, the 4060 is still roughly 20% faster than the 3060 it replaced at 65w less power draw.
Factor in power consumption and the 4060 isn't terrible, brings better gaming to mobile platforms as well.

 

[tetrasect]

The power consumption is far better on the 4060 vs the 3060ti, the 4060 is still roughly 20% faster than the 3060 it replaced at 65w less power draw.
Factor in power consumption and the 4060 isn't terrible, brings better gaming to mobile platforms as well.

So, only good for laptops.

Laptops are still as crap as they always used to be, sorry to break it to you.

Overpriced, annoying having to look down at the screen which you can never watch a movie on with another person and they cause all kinds of health problems from having to twist your limbs in weird ways...

Upgrade to PC MASTER RACE and bathe in the luxury of a 32 inch 4k monitor with wireless keyboard and mouse and a good set of speakers that will make your laptop look like a fisherprice toy.

 

[Lupus]

So, only good for laptops.

Laptops are still as crap as they always used to be, sorry to break it to you.

Overpriced, annoying having to look down at the screen which you can never watch a movie on with another person and they cause all kinds of health problems from having to twist your limbs in weird ways...

Upgrade to PC MASTER RACE and bathe in the luxury of a 32 inch 4k monitor with wireless keyboard and mouse and a good set of speakers that will make your laptop look like a fisherprice toy.

Why would I want to watch a movie on a laptop, am I a student? Movies will be on the OLED, plus you do know you can plug a laptop into an external monitor

 

[tetrasect]

Also the efficiency gains themselves are not as impressive if you compare the 9700x for example to the 8-core cpu with the same 65W TDP in the previous gen which would be the 7700 not the 7700x. It's more efficient, yes but performance per watt isn't as big as a leap as comparing to the very inefficient 7700x looks. Basically, the 9700x should be called the 9700 - maybe that's the typo

 

[tetrasect]

So the Intel microcode is out and it looks like it is causing some people's windows activations to fail because windows sees it as a hardware change.

 

[tetrasect]

Intel 14th-gen stability BIOS update obliterates multicore performance with 23% loss in some benchmarks​

PC Guide tested two different heavily multithreaded benchmarks immediately before and immediately after the BIOS update and found consistent performance degradation of 22–23% in each.

In Cinebench, the unlocked Core i9's score dropped a whopping 9,123 points from 39,783 to 30,660 points — or 22.9%. The publication's Blender render test also saw significant performance losses of 22%. The consistency of the performance loss in the multicore benchmarks — one synthetic and the other more realistic — indicates there is some significance to the results.

Fortunately, there was no apparent effect to gaming performance, with the CPU scoring identical FPS in Counter Strike 2, Cyberpunk 2077, and Days Gone before and after the update.

Interestingly, during both the Blender and Cinebench multicore tests, per-core CPU voltage was slightly higher after the update, meanwhile per-core CPU clocks and CPU temperatures were significantly lower after the update. Looking at the average CPU temperatures and abnormally low clock speeds during the testing, it's reasonable to conclude that something about Intel's Thermal Velocity Boost wasn't working properly during the testing.

https://www.notebookcheck.net/Intel-14th-gen-stability-BIOS-update-obliterates-multicore-performance-with-23-loss-in-some-benchmarks.873898.0.html