• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

X299 and 7980XE in 2024

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.
Interesting. That software must be highly specialized for you to hang on to that old system. I almost feel bad for the electric bill.
I don't "have" to have AVX-512, and more AVX2 performance can offset it. thinking about it, Zen 5 getting more FP execution could improve AVX2 as well as AVX-512, if they make the front end sufficiently wide.

The optimisation for my uses are, like most things, not straightforward. It doesn't scale like Cinebench, and fewer faster cores generally work better than more slower ones. Throw in clock, power efficiency, up front cost. Other complications like split CCDs on >8 core AMD, which is really bad news because the data is in the wrong place and there isn't nearly enough bandwidth to move it around.

That I haven't upgraded yet is more in part nothing has felt sufficient up to now. Sapphire Rapids would be idea as long as I don't look at the cost.

Also it isn't the sole factor into my system choice. It will get used for other things like gaming and video editing for example. There is no one offering that does it all the way I'd like. I almost bought a 7800X3D on several occasions. Great for gaming, has AVX-512, limited core count. The 7950X3D was 2nd choice, but the hybrid nature is a problem. Both sides should have 3D cache or neither to avoid problems. 7950X is one of those but I'm trading more cores for less gaming performance. Something like 14700k is a good generalist as long as you don't look too close at power usage.

At which point I get a headache, buy nothing and wait for next gen :D That's still the current plan. Basically if I had to buy a system today, any of the above would work. But none of them feel enough to buy them because I want to, as opposed to have to.
For fun. Why do some people fold? If you like doing something, you might weigh it higher as a factor. The software used is generally close enough to Prime95 that I'd use it as a well known example. Basically the math library from Prime95 can be used by other software. To many in the enthusiast space Prime95 might be known as a stress test but that is only a side effect of the actual work it was created for. People smarter than me have concluded that was the best approach.

Again, AVX-512 isn't an absolute "must have", otherwise I'd wont be able to buy Intel until... AVX10 replaces it? Still it is a strong "nice to have" and I will be keeping a close eye on Zen 5's rumoured improvements in that area.
Why do some people fold?
Because there's a point to the work. It's crunching folding proteins to find a cure for various diseases. You dont have to have a reason, per say, i was just wondering what the point was to this focus. :)

To me, prime number apps are for stress testing/benchmarks. Like is there a point to prime work? Or running super pi outside of benchmarking? How do these things apply to real world use?

Again, not trying to minimize what you're doing, just trying to understand what you REALLY gain from the methodology (f@h has a purpose, versus 'testing'). I dont understand how you apply what you learn when, it seems, the only thing you're doing with 512 is the test itself. In other words, the testing has little to do with performamce outside of 512. 512 'enabled' apps are few and far between for an overwhelming majority of users.
Last edited:
Many people do things to find what's out there. If there is a benefit from that can come later. Not in the same lifetime later. An example I like is when a French mathematician came up with Galois fields, did he do that because it had a use then? In the digital world it forms the basis of many ECC type schemes, but that was a century or two later.

Through my systems, I have found three notable prime numbers contributing towards mathematical conjectures. At most I'd get the tiniest footnote in history for it. Might not seem worth anything right now. If humanity survives long enough, maybe it will. Maybe not.

There are mathematicians and programmers looking at ways to improve computational techniques at large number computations. Make it faster, or more reliable. They're answering questions I can't even understand. Along the way they come up with new methods to do things. Over time, those techniques could spread and give benefit to other uses.

So in that sense, it isn't as targeted as folding, but it is not to say there aren't benefits from going through it.

It is correct that AVX-512 is basically a way to do certain instructions faster than without. AVX2 is the fallback path supported since Haswell, and even older paths exist. Both do the same basic things at the end of the day, just one is much faster. Adoption always takes time. If only Intel didn't drop it in mainstream since 12th gen it might be more popular now. AMD picking up the ball could help too, especially due to their presence at datacentre/supercomputer level, where Intel still support it too.
That makes sense... like a folder finding the most powerful/efficent card(?) or wanting to know/testing f@h performance. So, you're running Primegrid or something to help find prime numbers? Just say that!!!
Does it matter if it is PrimeGrid? While it is the project of choice at the moment, I have done others in the past as well as kicking off some independent searches. That was more "filling in the blanks" to provide data which may help in the study of the distribution of prime numbers in certain formats.
The application doesn't matter, no, lol. The point I was getting at was testing something applicable for what you're doing. As many times as I've asked/we've danced around it, this Is the first time I recall you actually mentioning WHY.
I'm pretty sure I said multiple times it is Prime95 adjacent.
You have... just don't think you've ever said it's a distributed platform-type work before. I think several of us got the vibe it was like a stress test program or......no clue, lol. Nothering ever struck me as a straightforward answer for w/e reason, lol. Anyway, thanks for explaining things/humoring me. :soda:
Please take this as someone who understands the part BUT not the whole argument. IIRC the Intel 13th and 14th Gen can run AVX-512 IF you disable the E-Cores + a setting in the BIOS (?). The AMD x3D has the additional cache to improve performance does this not help in Prime? I have also read where AMD has Large memory (64gb sticks) working on desktop boards. <- AMD has also stated that Future chips will also be AM4 compatible. Intel is a 2 chip per socket so the only the 15/16 chips will work on the z890/z990 (?)
IIRC the Intel 13th and 14th Gen can run AVX-512 IF you disable the E-Cores + a setting in the BIOS (?).
Some very early 12th gen has it enabled. Intel hard disabled AVX-512 after that. I'm not aware of anyone getting it working on 13/14 gen but I haven't looked either. Also Intel further disabled it in microcode, so if you ever update bios or Windows, it'll take over and disable it too. It will be painful as you need an early 12th gen CPU, a compatible mobo on old bios, and an old version of Windows. I only want to try it for testing purposes but as you can see it is a lot of pain to even get there, so this might never happen.

Or I could get the workstation CPUs if they ever make it cheap in the future.

The AMD x3D has the additional cache to improve performance does this not help in Prime?
For tasks that fit within the cache, which I think is all the ones I'm interested in, it means the ram doesn't matter any more so I can buy any cheap stuff. Like on Broadwell with its 128MB cache I could run single channel 1333 and not be held back. It becomes core limited. The problem for me is that above 8 cores you have multiple CCX and I have to treat them as logically separate CPUs.

AMD has also stated that Future chips will also be AM4 compatible. Intel is a 2 chip per socket so the only the 15/16 chips will work on the z890/z990 (?)
You mean AM5? AM4 is way past EOL outside of cheap/disposable systems, and the only "new" CPUs they're putting out appear to be the lowest bins they're clearing out. I'm really not bothered by generational support by mobo since I tend to build systems and retire older ones. Platforms also improve over time so I don't want to have a leading edge CPU on an ancient mobo. With my life situation at the moment I'm unlikely to get many test systems like I used to. I definitely want one new system as my main desktop, and optionally a new one for my TV gaming system. So I could get one each of Arrow Lake and Zen 5 if either are worth it. It also depends on timing since Zen 5 X3D might be very late.
To be clear, my earlier comment was specifically about a use case where the cache is either enough or not enough. Where it is enough, the ram really doesn't matter at all. For other use cases, ram can still have an impact.
Thank You for taking the time to explain my comments :cool: I was going to mention the Intel Server/AMD Pro chips BUT they are WAY to expensive compared to the x299 when it was new :-(
I will say I never had any of these issues on my EVGA X299 FTW k Mobo just like another poster.
But I was a idiot and trusted my water block to hang on a loop hose and it smacked my EVGA X299 FTW k mobo socket and destroyed it.
So, I went to an Asus x299 Tuf Mark 2 mobo and this happened if i keep things under 4.6 GB I narrowed it down more now then no channels fail.
Once I Go above 4.6ghz I need to initialize all of them 1 at a time it's so broken and thankfully I can run like 4 cores at 4.6 and all the rest at 4.5 GHZ and it's all good I can also push the ram speeds up doing that without errors the Mem controller with 8 Sticks just feels like its on the edge.