Heres the thing though, are you an enthusiast because you spent money or ar eyou an enthusiast because you really dig the tech and know how to use it?
The argument that the market is splitting up might be true, however we are about to go from just chiplet setups to straight up server tech across the board. Yeah 5XXX might not be the hotspot immediately, but 4D and 5D are only just dropping and will have a huge huge huge impact because of cache changes. You’ll have seen a massive swapout happen by the time the LGA socket comes out.
The 5950X already throws punches with the 3XXX TR chips from feature level to TDP to straight up horsepower, and we’re about to see everything get lapped and smoothed out. I think its a waiting period, not a “I have money gimme new” period, as often most enthusiasts like to enforce, most of the time on their own periods.
From my perspective, it’s hard to argue against the fact that HEDT just doesn’t seem to have a purpose for the most part.
My 5950x + 64Gb 3600cl14 ram trades is about as much as I can picture any non-profressional needing/wanting. The only true restriction is the lack of PCIe lanes, but PCIe 4 and soon to be 5 solves that problem; we can just start spliting lanes without bottlenecking, if need be.
HEDT is fun, but the only reason it existed was because of the stagnation in the market and Intel saw an easy niche to fill with marked up, overclocked server-esque parts. With AMD (and Apple) pushing the industry to chiplet designs and higher core counts, there’s little reason to differentiate the markets anymore.
Even ECC memory protections is become part of the DDR standard.
We may see larger price gaps between SKUs such as 12900KS vs 12900K/12700K, but that’s about it. I’d actually appreciate fewer motherboard skus as that would help increase QC and feature value without nickle and diming every little thing across the stack
Valid points all.
For me, and again this is my situation only, HEDT fills the gap between the regular Ryzen line and EPYC. I work with massive data sets and it’s not uncommon for me to be running a pair of VM’s for software testing, be converting several videos and have a AAA game running at the same time. Work does not subsidize my home lab or PC so it comes out of my pocket. HEDT TR was a good compromise in that I could get a number crunching monster for a reasonable amount. The Ryzen line is great, no argument, but I don’t think even a 5950x would have enough raw horsepower to handle my present load (or future workload increases for the next five years). That why I was looking at the 32 core as an upgrade. The TR PRO line does, but at a premium cost and that may be out of reach, I’ll have to see. Quad channel HEDT was my sweet spot for budget and performance. DDR4 RAM is cheap and accessible and there would be no Worry Induced Financial Economies (WIFE) implications, if you get me.
It looks like scalpers are snatching up the few remaining TR 3970x regardless. Prices have increased by $400 in the last 24H and no major retailer is carrying them. So I’ll just have to hold fire until Q2 and see what the damage will be.
I have been doing exactly that (EPYC 24 core/128GBram):
Main VM running OSX for my day to day work related activities (8 threads/32GB)
Gaming VM (12 Threads/16GB)
Win10/11 Work Utility (Excel/Visio) VM
Multiple Linux VMs to run tests/devel/compiles/whatever
The Windows VM has an RTX3070TI passed through, the only issues I had were with finding the proper USB controller to pass through without issues, and the Windows dedicated USB sound card (initially used a Zoom U-44, lots of grief, reverted to a Roland UA-25, have been a happy camper since) and can run Apex Legends at 160fps with High settings, I limit it at 100 so that the fANS do not spin up too much and less hot air is coming out of the case …
Budget for a 2nd hand 32 core ROME 7F32, Asrock Rack ROMED8-2T and 256GB RAM is way below 2.5K, more like 3.5K new … if you don’t want to wait …
Are there statistics to quantify how much, if at all, the on-die error correction in DDR5 has improved improved the overall error rate? The only reason on-die correction is being used after all, is because DDR5 manufacturing is causing enough errors that it is untenable not to use it.
The Ian Cutress video,
notes that doing the ECC calculation on-die does not protect data in-transit from corruption, only at-rest.
What I find additionally concerning, is that on-die cannot be SECDED, it has no way, as far as I know, of communicating to the CPU that either a single error has been corrected or that a double error has been detected.
On-die correction is more complexity and less insight into what is actually going on; this is the exact opposite of what anyone would want.
Thanks for that. I’m going to wait until the TR PRO pricing is released before making a decision. My 1950x is still viable for at least another year, so as long as it keeps ticking over there is no real pressure to buy anything.
Spent a week searching for TR 3970x. I found a dozen pirates wanting $1,000 over MSRP for used / open box / demo CPU and one retailer who had it listed for $200 over MSRP on their web site. I bought it successfully and received a refund notification within seconds. Got an email saying it was not in stock.
So it looks like my hunt for an upgrade is on hold. EPYC is an option, but that would require replacement of pretty much everything. Will have to take a deep dive into researching it, but it may not be financially viable.
Dunno about elsewhere but here threadripper 3xxx cpu prices are ridiculous.
could literally build a high end gaming 5950x based system AND an Epyc pre built rack
Server from dell (to cover both bases) for the cost of a high end threadripper cpu with no board or anything else.
I am in a similar place but on the slightly older intel xeon side of things. I have basically resigned myself to having a hypervisor for VM duty and a daily driver for linux desktop and a windows CAD VM.
I am excited that AMD is bringing onboard graphics across the Ryzen line moving forward. I am even more excited for Proton to get to a point where it will run Windows CAD applications so I can ditch the windows VM entirely.
With news coming from AMDs new consumer platform and their so-called x570E Extreme chipset, I’m rather disappointed from an enthusiast standpoint. We still only get 24 lanes, only difference is PCIe 5.0 instead of 4.0 for parts of the slots.
But with 24 lanes, you’re again down to Z690 level of 16x for GPU (or 2x x8 slots) and 8 lanes for the chipset which fuels all NVMe and IO. AMD really has to make these lanes 5.0 or chipset bandwidth will be a severe bottleneck again. Do you really want 2x PCIe 5.0 NVMe storage to run through chipset lanes?
And dual-channel obviously didn’t change, but DDR5-5200 will certainly give a bump to memory bandwidth but it still can’t beat TR 3000.
I love the Zen4 CPU infos, but I’m not impressed by AM5/X670 as of now. I guess Fishhawk Falls and Intel could be my salvation depending on how things turn out for that platform.
For a very long while I was expecting AMD would sooner or later add an IGP on the Ryzen line, even though we may never use one when having a desktop with a dedicated GPU it can be handy for an extra encoding/processing task just like how Intel pushed QuickSync for many years. I’ve used QuickSync a few times when my GPU was doing another task.
Even though I haven’t looked at HEDT for several years, at the moment it seems more like Intel & AMD are back to consumer desktop+laptop focused speed/core wars at the trade off of leaving HEDT in a weird unpredictable place. From my own usage if I were to use two NVMe SSDs that would put a dent into leftover PCIe lanes, it sort of forces me having to settle with SATA SSD to free up PCIe lanes.
At the moment I’m still staying neutral yet I expect Intel will jump ahead of AMD with the potential growth in usage of faster NVMe storage on the consumer side. For my own usage I may have to just use a desktop focused on NVMe then rely upon a network server to fill the gap of extra I/O or PCIe device(s)… heck its just tempting to throw a GPU into a server and live off a notebook.
As “limited” as the TR3000 series is, it’s still literally three times better than my current 1950x TR and in my mind a viable update.
I’ve given up looking for one. There are still some out there, but for ridiculous prices ($700 - $1,000 + over MSRP) from scalpers. TR 5000 Pro would seem to be the best short term bet with EPYC as a second choice. However, until pricing is released I can’t make an informed decision.
I suspect the TR PRO pricing release will immediately result in the sound of air being sucked through clenched teeth.
The Lenovo P620 website now shows TR 5000 series pricing. (Note says it ships in 4+ months!)
Price for a P620 system with TR PRO 5945WX (12/24) is $2,819. If you look at the gear installed and remove those costs (PSU, GPU, RAM, SSD, case, etc.) it all comes to about $1,000. So the CPU itself will be around $1,800.
Using the same method, the 5965WX (24 / 48) CPU will be around $4,450.
Back of the envelope calculation and I used retail pricing, so lots of leeway, but I bet I’m close.
Perhaps, but in this case, I don’t think so. I can’t buy anything at the moment so best to wait a few (more) months until AMD finally gets their thumb out or otherwise says something of substance.
