Discussion in 'Article Discussion' started by bit-tech, 20 Dec 2019.
Let's be honest, if you're looking at pure CPU performance, the 3950X renders most HEDT solutions null and void from a price/performance/efficiency perspective.
The article makes reference to cheap x299 boards without realizing that cheap x299 boards don't support high core counts, the new chips, or satiate the lane desires for HEDT, you have to step up to ~250 starting point for x299 to get something you can use a decent HEDT chip in, which changes the viewpoint of the article a bit IMO.
The gigabyte x299 linked is a prime example, it is very cheap indeed but it only supports 4 core chips with 16 lanes, pretty much any 2nd/3rd gen desktop ryzen chip is superior.
Unfortunately whilst the 3950x CPU is great, the platform for 3950 is not, makes it a weird purchase as it is not really a gaming CPU but it does not really have the platform to let it stretch its entry level HEDT chops, the good x570 boards with a bit more lane flexibility for example to do 8x8x8x (but of course no more lanes) cost as much if not more than x299/x399/TRX40 boards, so many of the ryzen boards offer the likes of 16x/4x 8x/8x/4x if you are lucky but often less until you step up, really weird segmentation of boards across the range.
Cascade Lake X seems to fill a good middle ground between gaming ryzen setup and TR3 HEDT setup. Particularly as TR2 ha been end of lifed so you can't buy one if you want to unless you want the left over stocks of crippled 2970/2990 chips but no one with half a clue would, you would just get TR3 for not a lot more.
I was going to jump on Cascade Lake, it would get me a CPU I would of setup to be quicker than a 3950x due to good overclocking and a decent platform, but in the end I stuck with my x399, i'll see what happens with the next cycle.
Its a shame there are no low core count TRX4 CPU's, as AMD left Intel a gap for low core count HEDT platform with many PCIe lanes.
I would pay happily €500 ~ €700 8 or 12 core TRX4 CPU, as the platform looks like it has a longer life then the C422/X299 platform.
So i think i gone get a quad core S-2066 Xeon W-2223 with a C422 Supermicro X11SRA-F, as i need the extra PCIe lanes for my SAS cards, to upgrade from my S-2011 Xeon E5-2609 / C602 Supermicro X9SRL-F.
I still gone wait for some time (3~6 months), and see if there gone be cheap TRX4 CPU's or TRX40 SP3 Epyc mobo's coming
I like to get a AMD platform, not because i like AMD that much, but just because i like Intel even a lot less.
So yes, Intel's plan worked sort-off, as AMD left a gap, at least for now.
Xeon W isn't a platform many would consider for a home PC unless they have multiple 'work' machines; the target for those parts is more the corporate/enterprise server market. AMD's answer to Xeon W would be Epyc (their Rome variant being the latest iteration).
Cascade-X is aimed squarely at home 'work' machines, on the same level as TR4 parts... 8-12 cores for this sector got left behind a cpl. generations ago.
So, there's lots here to unpick about AMD’s decisions. Firstly, I'm not sure if they're making pricing decisions more to reduce demand so that they can sell more Rome EPYCs or if they're making the same old mistakes again as they did in the 64 era.
I can understand why they might need to bump up pricing to account for the 7nm node but prices should come down fairly quickly as they make the money back off of R&D investments and as the cost premium of the 7nm node falls off (although outside competitors for capacity at TSMC seem to be keeping wafer prices high until Qualcomm and Apple move production over to the 7+nm node next year.)
I think that AMD still need to keep themselves focused on value. 50% more for 100% more cores sounds great but when the product it is replacing is already £500 that's a huge jump. It's far too high really. An acceptable jump would have been to go to £650 but with an aim to get it down to below £500 after R&D had been reconciled for this generation. Remember that they can sell the same silicon much cheaper as the do so already, with often perfectly usable silicon disabled to fulfil market segments and keep ASPs high. ASPs don't interest buyers too much, they're there to keep investors happy but I couldn't give a toss about investors but unfortunately they often have too much power to force bad decision making.
I know that that's a very consumer focused call but it does us no good for us to have an AMD that just tries to replace Intel in the price gouging department. Mainly because Intel will put the resources in place and eventually get the right decisions taken to get them back into a competitive or even a position of superiority over AMD.
If they continue price gouging they lose any good will built up with the initial Zen of releases and will quickly be dropped when they need the momentum to continue to compete. I like to point out that you can actually spend near nothing on marketing if you let word of mouth do the selling for you; which is especially true if you product is superior or at least extremely competitive with the competition. This is what's measured by net promotor scores.
So this question should have been very simple to answer. AMD's pricing for the new TR processors should have been about £1100 for the 24 core processor and the 16 core 3950x should have been £650 with a 16 core TR at around the £750 to £850 Mark. Instead the pressure value is slightly reduced on Intel when AMD need to keep that on full.
I'd like to think that Zen2 Threadripper parts will settle at 75% of their release price by the time they reach 'saturation' levels, so ~£1000 & £1400-&-change... leaving the 'saturation £2000' bracket open for the 3990X.
Not for the file server market, for a file server is actual a dual core 2.5GHz already plenty to run ZFS, tho they still need plenty of PCIe lanes.
I meant for home 'work machines'; video editing, etc for content creators can typically be handled by consumer 8-16 core parts.
What the W-2223 will offer over the similar consumer parts is reliability in a 24/7, 100% load scenario... that's what it's designed for, whereas a typical Threadripper or Core 10-series LGA2066 part isn't.
For me its the cheapest part that fulfill all my needs, cheap (€350), reliable, fast enough and with 48 lanes, it has 16 more lanes then i need for my purpose.
Tho i would still prefer a cheap TRX4 CPU on the more modern TRX40 platform.
AMD have killed the HEDT space. The clue is in the H.... Its game over, the end, call it what you want...
For lesser (marginal) users you have other options... This is not where the serious money is.. AMD know this and won't give (TBH a s***).... The profits are in the high end no compromise space...
For genuine HEDT customers there is only one destination.
Well done AMD, let see if you can pull of a similar remarkable turnaround in the GPU space... I doubt it but I think you might surprise me!
The margin is in the high end.
The money is in the budget consumer.
As @yuusou says, surely the serious money is in the high-volume sales of lower-end kit, as even if it is lower margin, sheer volume means that it generates far more profit than HEDT, which has tiny sales in comparison.
AGree... I was just referring to the HEDT space, where I think its clear AMD is the winner on all serious metrics... Intel is a sideshow..
LOL, Tel that to the HP's, Dell's and lenovo's in the world, in the consumer range AMD is king now, in the much bigger corporate (enterprise) market, Intel still rules by a 1:10 ratio, sure Intel got a bloody nose for now, but they can buy them self's out of trouble, like they always do, and yes this is a problem money and some time can fix, and Intel has still plenty of it!
It's literately a Goliath versus David story, but with the missing biblical heroes context.
Net income US$21.0 billion (2018) 62x
Total assets US$127.9 billion (2018) 28x
Total equity US$74.5 billion (2018) 59x
Net income US$337 million (2018)
Total assets US$4.556 billion (2018)
Total equity US$1.266 billion (2018)
Will 2019 show much better results for AMD, sure, maybe Intel's net income will drop from 62x even to as low as 40x, it will hurt, but it's far from fatal, and they still can do a lot with that 130B in cash they have, and easily buy them self's out of trouble again, and even if it takes 2~4 years, they still have plenty of time to come up with there new chiplet ''Conroe II'' design, to save Intel's day.
And Intel in the past has shown that they don't mind playing dirty, so for now Intel is going to give bulk buyers huge discounts, to be on par with AMD's price/performance ratio, just like they did with the K8, and then come up with competitive solutions, and by then there foundry issues should be solved, and they will put the knife in again, and keep stabbing AMD till they will be working in the margins again, and we will have to buy overpriced Intel products again!
HEDT is not a homogenous "just throw in more cores" market.
Beyond home-build not-sure-what-I'll-do-with-it-so-make-it-do-everything workstations (those theoretical "I'm going to be playing and streaming a AAA game while the same box is compositing my After Effects project and rendering a scene in Blender!" users), professional High End Desktops are generally going to run one application or a specific suite of applications. If a CPU is faster in general means little if it is not faster in the tasks it actually needs to do (and not just "it's faster in After Effects in a canned benchmark", but "it's faster with this specific CODEC and operation that we use daily"). Then there's system TCO to take into account: it may make more sense to buy the up-front more expensive CPU if it's going to provide workflow improvements in the long run that are worth more than the initial premium. Or it may make sense to use an apples-to-apples 'slower' CPU if it means you save enough to buy sufficient RAM to fit your entire dataset into memory at once rather than paging (giving an overall dramatic task speedup).
Separate names with a comma.