Intel Core i9-9990XE: Up To 5.0 GHz, Auction Only (anandtech.com) 146
Ian Cutress, reporting for AnandTech: AnandTech has seen documents and supporting information from multiple sources that show that Intel is planning to release a new high-end desktop processor, the Core i9-9990XE. These documents show that the processors will not be sold at retail; rather they will only be sold to system integrators, and then only through a closed online auction. This new processor will be the highest numbered processor in Intel's high-end desktop line. The current top processor is the i9-9980XE, an 18 core part with a base frequency of 3.5 GHz and a turbo frequency of 4.0 GHz. The i9-9990XE, on the other hand, is not simply the 9980XE with an increase in frequency. The Core i9-9990XE will be a 14 core processor, but with a base frequency of 4.0 GHz and a turbo frequency of 5.0 GHz. This makes it a super-binned 9940X.
Difficult to find a manager with tech. ability? (Score:2)
Is that at least partly because it is difficult to find a competent manager with an understanding of technology?
Re: Difficult to find a manager with tech. ability (Score:1)
Intel doesn't need a Lisa Su.
They need a Rory Read. If they fibd a Rory Read, then he will be able to pass the company to some Lisa Su.
Hint: Rory hired Lisa.
~o~ Intel Core i9-9990XE L@@K RARE!!! ~o~ (Score:2, Funny)
nt
Finally! (Score:4, Funny)
Now I can play Crysis.
Re:Finally! (Score:4, Funny)
until windows 10 starts updating, anyway.
Better have a heavy water cooling loop (Score:1)
Otherwise you'll have Chernobyl in your bedroom.
Re: (Score:2)
Why? 255W TDP is nothing exciting. You can handle that easily with a large air cooler or a standard 240mm AIO. I have a 360mm radiator in my system happily handling the 430W TDP of my combined CPU + GPU with ease. I've only managed an 11C temperature rise in the water under synthetic stressing of the GPU or CPU, not even individual benchmarks were able to get the temperature to rise that much.
Re:A good Matlab replacement, not the next big thi (Score:2)
I was thinking of hooking a water cooling loop to my central heating radiators.
Re: (Score:2)
That may be surprisingly ineffective at actual cooling. The radiators in a house are designed to create natural draft and in order to do that you need a very high temperature differential between the water loop and the ambient air. That is what you get when you have a vertical fins running lengthwise through a radiator. Additionally most depend on convection and radiate poorly thanks to the default colour choice of off-white and the thick layer of lacquer covering the metal.
Mind you it will still work. The
Re: (Score:2)
Re: (Score:2)
It was a joke.
It shouldn't be. Enthusiasts are an inquisitive and experimental crowd. You may have made a joke but your joke has been a very serious question asked over and over again on online forums.
People do all sorts of wonderful things such as using full sized car radiators: https://www.reddit.com/r/shitt... [reddit.com]
I wish I still had the link to it but in the mid 2000s I remember seeing someone who laid copper tubing in a giant S bend under the foundation slab of his new build garage and cycled that through his CPU. Dirt an
Re: (Score:2)
Followed by the 9999XE++++ (Score:1)
Re: (Score:2)
Why not go with a Xeon? (Score:2)
This type of CPU seems like something that is more sizzle than steak. If it is so expensive and exclusive, what makes it better than say, a Xeon Platinum or even a Xeon D?
There is so much overlap that one might as well jump to a Xeon if an i7 just isn't up to the task anyway, unless this is a laptop or mobile machine.
Re: (Score:3)
Re: (Score:2)
Normal humans don't want their massive filesystems to be corruption-free. :rolleyes:
Thanks, AMD, btw. BIOS vendors, get your shit together.
Re: (Score:2)
Normal humans don't want their massive filesystems to be corruption-free. :rolleyes:
Normal humans practically never experience corruption on their filesystem. If you want to get people to care about something they first need to be affected by it.
Re: (Score:2)
How do you know? That last "everything went batshit" reboot might have been caused by DRAM corruption. It might have been caused by buggy software, or even cosmic rays. Who knows how many "mystery" crashes might have been averted by use of ECC memory and bus structure. No one knows, that's part of the problem, n00b end-lusers are blamed, system gets re-imaged, and all is right with the world for another quarter. Rinse and repeat.
Re: (Score:2)
That last "everything went batshit" reboot might have been caused by DRAM corruption.
Interestingly not only did you just move the goalposts you made them as wide as the entire frigging field. I never said DRAM doesn't get corrupted, I said "filesystem". Your memory is chock full of data so in the incredibly remote chance that you actually suffer from an error that ECC detects you have an additionally incredibly remote chance that you happen to corrupt data in flight on the way to being written on the disk. The vast majority of these already incredibly rare events will as you say cause a "ba
Re: (Score:2)
Geez. I didn't expect the Spanish Inquisition!
No One Expects The Spanish Inquisition!
Our Chief Weapons are Surprise, Fear, Etc., Etc. ...
No goalpost movement here, bud. Relax!
Seriously, user "DigiShaman" started talking about ECC.
Then, user "bill_mcgonigle" conflated lack of ECC memory with filesystem corruption. I thought that's where we were at. I just went with it.
Interesting anecdote about your experience with ECC memory, though. No doubt, ECC there to handle the one-in-a-billion case of a more tha
Re: (Score:2)
No goalpost movement here, bud. Relax!
Sure there was. You came in and wanted to play a different sport than others were playing. I was replying about file system corruption and specifically talking about file system corruption and you went and changed the scenario to one which not only has many orders of magnitude different likelihood but also orders of magnitude different consequences.
But back on point, even DigiSharman's original assertion contradicts itself. If you want "high-performance" then you don't want ECC RAM. The entire topic of ECC
Re: (Score:3)
Re: (Score:3)
Intel prices this stuff carefully. You do hit a point where it's cheaper to build a 2-socket server system for the needed performance. I did all the math very carefully before ending up with a 28-core 2-socket Xeon system for video transcoding.
I rather suspect that if this CPU were priced normally, it would mess up the very careful price curve for Xeon, and end up cheaper for some loads that force you to Xeon today. Can't have that.
BTW, the lower per-core clock is all about thermal management, which is f
Re: (Score:2)
You do hit a point where it's cheaper to build a 2-socket server system for the needed performance
What needed performance? Show us a Xeon with the single core performance of this chip.
Re: (Score:3)
Tools for jobs. Single core (and not GPU-bound) is a very narrow problem space these days.
Re: (Score:2)
What is a $2000+ processor if not a tool for a very frigging specific job?
Re: (Score:2)
Status symbol. Nothing else, really.
Well, I guess that is a job, now that I think about it.
Re: (Score:2)
Well played :)
Re: (Score:3)
Re: (Score:2)
I prefer more cores as well, just because more tasks can run unimpeded. Especially with desktop virtualization, so the web browser VM can run without affecting anything else on the system.
AMD is looking quite attractive these days for the desktop. More cores, less cash.
Errrr SIs only? (Score:4, Insightful)
This is quite the curious move from Intel. Sure SIs have budget to buy in bulk and to auction, but the super enthusiasts who would be the likely target market for this chip aren't usually the type to go out to an SI and say "I want something off the shelf, please build for me".
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
but still, it does happen
Of course it happens, the question is does it happen in a big enough market. I'm willing to bet the majority of 7980xes were not bought by SIs. But you even said it yourself there were mitigating factors that pushed you towards that decision. In the normal case it's not cost effective to go to an SI vs building yourself and typically the type of people who are into these kinds of specs are more than capable of plugging a few things together.
Re: (Score:2)
Isn't that pretty much the Alienware market? Some dope will pay double for this tiny performance boost.
Re: (Score:2)
Isn't that pretty much the Alienware market? Some dope will pay double for this tiny performance boost.
Not really. The Alienware market is wide and varied and the "best of the best" make up a very very small portion of their sales. Yeah they do move some units, but compared to enthusiasts who don't want to pay double on an already insanely expensive piece of kit... you said it yourself.
So, does it come with the industrial chiller? (Score:1)
Obligatory https://www.youtube.com/watch?v=ozcEel1rNKM
Binned Over-Clocked Processor - Meh (Score:2)
Re: (Score:2)
With a 255-watt power draw, you know it's overvolted and overclocked.
Overvolted and overclocked imply you're pushing something beyond it's vendor spec. This is the vendor spec so by definition not "overclocked". In case you don't know basically every processor on the market is binned so by your standard every processor on the market except for the slowest model in every line is "overclocked".
Nein! (Score:1)
That sure is a lot of 9s. Looking at the CPU lists though, they haven't really used 9 very much in their model names. Is 2019 the year of the Number Justice Warriors?
Would I even notice the CPU speedup if I upgraded? (Score:1)
Like many people, I want my PC to keep up with the newest games, but none of these are CPU constrained. With dual graphics cards, even a cheapish CPU will not degrade performance. If people do a lot of rendering video or working in Photoshop, they can use Ouadro or FirePro rendering accelerators, and even a $170 cpu [ebay.com] will not be the limiting factor.
I think it's time for tech journalists to do the equivalent of a "taste test" to see if even power users can really tell whether they're running one of these s
Re: (Score:2)
I would probably assert that moving to a NVMe SSD might more important than going to a new CPU barring specific tasks like Photoshop or other rendering. For almost all interactive use, disk I/O is what causes slowdowns.
In my experience, it is usually disk I/O -> RAM -> GPU -> CPU, in that order for performance items, for most things.
Re: (Score:2)
There's only so much you can offload on the GPU. If you have a 6 year old GPU and throw a Quadro at it you're going to have a pretty bad day, even in Photoshop or Premier. I'm currently running an encode offloaded on hardware. My hardware video encoder is only sitting at around 60% many thanks to one of my CPU cores being pegged.
You're quite right about the law of diminishing returns for gaming though. The difference between a 9700K and an 8700K is only a few FPS. But hey if that matters to you then more po
Re: (Score:2)
Or they do things that require the fastest single thread perf you can get like VR and Flight sim, or VR+FlightSim.
Re: (Score:2)
Re: (Score:2)
If the flight sim isn't scaling with the cores, it's a software issue, not a hardware issue.
Fine. You fix the software, and in the meantime, we'll buy fucking fast CPUs so that we can continue to use it.
At 255W, systems integrators are required (Score:4, Informative)
I understand they don't want to give this one to the masses. 255W from an area the size of an oversized postal stamp will require some very good cooling and having it destroyed due to just poor installation of your water cooler, they'll probably won't be allowed returns at all.
I know Intel wants to give AMD the finger but this is 1/4 kW in the processor alone. Give it a good GPU and you're looking at a 2kW space heater. This moves makes it obvious they are desperate to increase both profitability and get to a smaller scale.
Re: (Score:2)
Re: (Score:2)
The thermal cycles of a modern CPU are no different to those from 10 years ago, we ran them up to 100C back then, we run them up to 100C now. The difference is that a modern CPU getting to 100C on die will cause it to protect itself whereas the old ones happily let the smoke come out.
TDP limit is a figure given to designers of cooling systems and my ancient Pentium 4 had thermal cycles almost identical to my Threadripper (which has 4x the TDP). Oh and my Pentium 4 still works 15 years later.
Re: (Score:2)
I've seen Core i7's melt and warp the surrounding PCB due to poorly installed (water)coolers. I've also seen people over-tighten them, cracking the chip or having liquid leak all over. If they're on there but not working good, they will dissipate the heat into the tiny watercooler sink which then becomes a radiator to the neighboring areas, melting out capacitors and various other things. Intel's CPU can work up to 100C before throttling, that's a LOT of heat.
Re: (Score:2)
Intel's CPU can work up to 100C before throttling, that's a LOT of heat.
No it's not. It's quite a standard amount of heat that CPU's have experienced since the days of Pentium 3 and the first generation of Athlons. Additionally most of this temperature does not couple into the motherboard where even if your die is at 100C you will read barely more than a 20C rise on the board itself. On the otherhand VRMs do couple heat directly into the motherboard and they are rated to 150C though most of them will sensibly throttle at around 130C mainly to prevent cooking other components on
Re: (Score:2)
I understand they don't want to give this one to the masses.
What "masses" spend in excess of $2000 on a CPU and don't understand or look into the cooling solution?
While we're at it, what 1990s era CPU is capable of being damaged due to poor cooling? Thermal throttling is a thing that occurs on a millisecond basis and failure for a CPU to protect itself these days would be grounds for returning it as being defective and not fit for purpose.
I know Intel wants to give AMD the finger but this is 1/4 kW in the processor alone.
Multiple Threadrippers have a 250W TDP, as do many Xeon chips. This isn't anything new. It's also nothing difficult to cool. Ther
Re: (Score:3)
Re: (Score:2)
The RTX 2080 TI is a 250 watt TDP chip
So are the Threadrippers and many Xeon chips.
Re: (Score:2)
Not for 15-20 years. Athlons famously caught fire without a sink as did other P4-era chips, Intel started requiring temp diodes on the motherboard around that time but the BIOS had to regulate it. Nowadays, temp sensors are integrated but they only throttle to 25% and they produce enough heat before throttling that you can easily, with the help of a tiny water-cooler heatsink destroy or warp the motherboard and surrounding components, eventually destroying the chip.
And yes, people will install the chip the
Re: (Score:2)
And yes, people will install the chip the wrong way, overtighten the cooling, not applying or too much cooling paste (resulting in uneven heat dissipation) or somehow destroy the chip and simply return it to the store and ask for their money back.
And? That applies to the cooler on a Celeron as well and has nothing to do with high TDP.
The GPU is a lot larger
No it's not. The GPU is firstly direct die and is smaller than a heatspreadder. On top of that the cooler is required to connect to multiple parts of the card. If you're worried about enthusiasts destroying any component while installing a heatsink, then the GPU is orders of magnitude more likely to be ruined.
They also have the heatsink already assembled in most cases and removing it voids the warranty.
You do realise that installing the heatsink on e.g. a modern 250W TDP TR4 chip or an LGA 3647 chip is no differe
Re: (Score:2)
Maybe he's suggesting that a GPU with over twice the surface area of a CPU offers substantially greater heat dissipation capabilities and is thus far easier to cool for a given TDP.
Or is that too complicated for you?
What about clock skew? (Score:2)
I know I'm a bit behind the times but when I was in school in the aughts, my TA was doing research on how to overcome the large amount of clock skew [wikipedia.org] you'd have on such fast chips. It seems like a 5 GHz chip would have problems with that.
Re: (Score:2)
5GHz is not even close to the record extreme overclocks have stably achieved on CoffeeLake. 7.3GHz is getting close to the limit where no amount of magic can make it run stably anymore.
Re: (Score:3)
Re: (Score:2)
I mean, Intel still leads in single-threaded performance.
If I have a workload that can be multi-threaded and take advantage of all the cores, though, AMD all the way, especially for the price you're going to pay. Even for mixed workloads that involve some single-threaded and some multi-threaded, it's probably worth considering AMD from a price perspective.
Re: (Score:3)
AMD 7nm chips should change that where AMD will also win single threaded as well as multi-threaded workloads. The IPC improvements from the Zen2 cores, combined with the 7nm fab process which will allow for higher clock speeds/lower power draw definitely have the potential to do this. Right now though, we need to wait another 5-6 months to find out if that will be true, but it is very possible. The unconfirmed leaked lineup of Ryzen 3000 series processors has 12 and 16 core parts, including boost that
Re: (Score:2)
AMD 7nm chips should change that
I hope they do, but AMD has a lot to prove on the IPC spec. They promised to catch up with Intel with every generation and while they've made leaps and bounds they still aren't there yet.
Re: (Score:1)
While I wouldn't go to one of these hyper high-end processors (too expensive, uncertain benefits) I still go with Intel. I've had issues with AMD parts locking up when running certain physics simulations (under FreeBSD) which haven't been alleviated with microcode updates. The Intel parts, on the other hand, seem rock-solid.
Re:Novelty (Score:4, Interesting)
A lot of money just to act like a big shot. I suppose it is like people who buy mansions or 100k sports cars.
It's an odd market. There are enthusiasts who push the limits on what's possible for overclocking, building ridiculous systems (e.g., liquid nitrogen cooling) with prices that are crazy, but still way cheaper than cars or photography as a middle-age-man hobby. Thing is, enthusiasts don't buy systems from integrators. Building your on is the entire point.
This is a dangerous marketing move for Intell IMO. These chips will go in systems for kids with rich parents. That could kill the enthusiast market for Intel. Chasing away your hardcore fans is rarely wise.