geekhack
geekhack Community => Off Topic => Topic started by: tp4tissue on Thu, 24 October 2024, 11:36:12
-
[attachimg=2]
Ok, so the big reveal is out.
Tp4 must emphasize, It's not awful.
But, probably very very few people will buy it. Which makes it a huge markdown on the balance sheet. These slower chips are also more expensive to produce than AMD's FASTER counterparts.
Gaming performance slightly slower than 14900k, which is already much slower than 7800x3d, and the upcoming 9800x3d (+18% vs 7800x3d).
And there could be a 9800x3d upgrade late cycle w/ new server class i/o die, which will be anywhere from +5 to +10% from 9800x3d vanilla
In productivity workloads, 285k is on par. The chip relies again on the E-cores, which gets it over the hump. In some workloads it has power efficiency nearly that of 7x and 9x amd chips.
Intel still has alot of rough spots going forward. This 285k chip (built by TSMC), is adequate in demonstrating Intel's technical competency, however, they DO NOT have a money maker product. If they are still 3-5 years away from their own TSMC competitor, it's going to be extremely icy roads.
Server side, AI cloud, AMD's Turin has no competition. You can E-core all day, but I think here the ship has sailed, Data centers can not wait, they need chips NOW.
Intel's Fabs are a very heavy burden right now. Under-utilization is equivalent to setting money on fire. And you can't exactly just turn these things off.
Intel could in theory sell the fabs, take the weight off, let the Stock rebound, but TO WHOM. Who can afford something this expensive besides a Nation State.
Who could even be trusted to continue development of leading edge Node.
So BEST CASE, the INTC ticker stabilizes for a year or 2, Tops.
Worse, we're looking at a 5-10 year come back.
[attachimg=1]
-
If history repeats...
The first gen after an architecture change is always a bit of a dud and the first refresh will likely close the gap with AMD or exceed it in most cases but 3rd will very likely beat them quite handily.
However,
This post-Moore's law and in most instances the hardware is outrunning the software so if your system is working fine, stick with it as real world performance, good or bad, isn't really that big of a deal despite how the charts may look.
-
From a PC gaming perspective, what is the point of any of these increases? I see every cycle that new CPU is 20-40% faster at X and Y. But does your average PC user notice or benefit from these increases?
-
Esp0rts yes, casual gam3r, no.
-
You pretty much only need enough CPU to push whatever GPU you have and most people are not rocking a 4090.
Frankly, that's one of the few things that can really take advantage of these processors, we're still trying to get heavy multi-threading to work at doing anything other than pumping raw data.
This is why you don't actually need 12, 16, or 32 cores, almost nothing you do can really take advantage of it. Servers, CAD, and virtual machines can (if you load up enough of them), Photoshop can (under certain loads), but notice these are all static or perform multiple operations by default. Things that require time sensitive stuff such as a video editor cannot. Same for gaming, the actual game isn't using a ton of of processor cores except to feed the GPU. It's slowly getting better but really for 99% of what you do, 4 cores was/is actually peak, you get more performance from newer cpu's from the clock speed and optimizations than you do from core/thread count but benchmarks won't show you that. You do get some real world advantage from more, but beyond 4 or so you very quickly start getting diminishing returns. Which is why Intel E cores actually work pretty decent, there's a lot of little things going on in the background that don't really require a top speed performance core most of the time (streaming radio, video, uploading video, running the network and peripherals, running a anti-virus, etc) leaving your P cores to do the heavy lifting as needed. That being said, I still hate e-cores in desktops.
Do I want 32 cores, YES!
Could I actually take full advantage of it... Yes, but only for about 10-20 minutes once or twice a year.
Oh, side note, with a big cpu like this, you're actually going to be SSD bound in some of these operations and no, PCIE 5X will not fix the issue, the real problem for them is I/O. Just because you can flow 5x the previous amount doesn't mean you can gather, collect and prioritize it fast enough to flood that bus. This is part of why servers use raid, it's not just about bandwidth and these newer/faster drives will have less I/O than an older raid system even if you have more bandwidth because each generation slaps larger storage chips onto the same number of chip controllers. This is why an older 1Tb SSD can (but not always) have more I/O and bandwidth than a newer 1 TB ssd, it has more controllers and chips doing the work (less parts is also why the newer one is cheaper). Remember what I said earlier, more cores are really only good for dumping raw data, well that's all a drive does so therefore more memory chips/controllers the better.
-
Do I want 32 cores, YES!
Could I actually take full advantage of it... Yes, but only for about 10-20 minutes once or twice a year.
Llann is up to no good, what could someone possibly do with that many cores.
-
Out here still running my 'ol i7 3770K
Even having it OC'd to 4.5GHz I have noticed it struggling a bit the last 3-4 years or so. DDR3 probably doesn't help either, even with 64Gb
My rig is going on 13 years old now :'(
-
Delid, you can hit 4.8 w/ more tweaking.