Etern4l

Yeah, another data point on Nvidia monopolistic abuse. The NVLink circuitry is there, they just removed the connector, so they can charge 2x per core for their professional 48GB cards. Good news is that AMD is getting more into the game now, from what I can see. Better late than never. Right, but 69C is pretty cool to begin with. Very curious to see a genuine OC FE vs OC AIB comparison. If these cards operate at 90-95% performance with 60% power, logic would dictate there is not much more to be gained there. Competitive overclocking will sadly be about who gets that extra 1.5-1.75% performance. Nvidia decided to kill off all that, along with those pesky AIBs, by locking out the voltage. Few people will risk shunt modding their $2k cards. Makes sense, which is where the FE blower design comes in...

Sorry, no more reactions left today, but I like the free power arrangement. Are we talking solar? Probably not, since you would need a sizeable farm to power a mining operation, and solar farms aren't normally free.

Nice, do you mind measuring the extra space required for the connector+cables, as well bent as you have them? Looks like somewhere around 20mm extra....

Are you talking about GPU crypto mining? Isn't that dead?

How are things looking with closing the case? The power connector looks crazy bent on this picture, like you are about to rip the connector off the PCB. The general advice is to minimise any bend on these high power PCIe cables, obviously next to impossible with the moronic card size and connector placement.

Source? Reviews generally state the FE card is cool and quiet. Der Bauer claims I wasn't able to OC the Aorus master over the FE. When comparing FE (stock or OC? not clear) to Aorus Master OC it's +6% in TS Extreme. Sure a gigantic radiator of a card will be somewhat cooler, but will it matter in practice. The only thing I would look for in an AIB card is less height (not available on the typical radiator AIB models) or a rear power connector. Anyone saw a 4090 card with the latter feature?

Apparently there is not much point in OCing this, but well done finding any stock at all today. Nvidia def learned their lessons on how to manage stock to extract max value 🙂

Exactly, same thing finding derBauer found. If I was an AIB, I would come up with a smaller card that still does the same job, as it would. This tech is so efficient, there is no point in putting an actual radiator in the PC.

Sure, the socket flipping is not great for the consumer, but obviously a 5nm process would have helped Intel? 🙂 Put another way, it's amazing that they can still compete with a 10 nm process. Put yet another way, AMD kind of sucks for not being able to completely blow Intel out of the water at 5nm.. Anyway, the way the world is currently going, we probably need toughbooks, solar power generators (no diesel, sorry), and a good supply of slingshots, clubs and maybe some abacuses to keep the science going when power finally goes out....

A couple of interesting points: * Nvidia advised him not to use 7950x for the test bench because of compatibility issues... FUD or real? * The card is basically factory overclocked. Setting power target to 60% yields... 90-95% performance. Insane! * He noted some issues with FE fans... Otherwise, well: 4090... love at first sight, shame about the price lol Edit: The even more gigantic AIB cards are basically pointless... probably why EVGA got out.

Good point about the benefits of chiplets, but what's IF there? I don't have any particular affinity with Intel, putting aside the fact that I have never owned AMD CPU or GPU 🙂 That said, if it can be shown that AMD can do the job more efficiently than Intel, I'm rolling a Ryzen build. Part of the problem is the popular synthetic benchmarks. You run CB23 and the only information you get back is a single number which is a proxy of how quickly the CPU would render a small scene. Since I don't do any rendering, I technically don't care, and the amount of memory utilised by CB23 is tiny, which is also inadequate for my purposes. I would need a benchmark focused on numerical computing but also answering the following questions: * How does the performance scale with the number of logical cores? * How does the performance fall off with the workload size? Anyone knows of such an off the shelf benchmark?

If we are talking about X299 platform replacement, nothing is anywhere near what would seem like close to a release, right? Intel is obviously prioritising the Xeon platform, and happy with the use of Xeons in workstations. At one point they will have to move that to DDR5, I would guess mid 2023 at the latest. I suspect "X99 HEDT" is a bit like Nvidia's Titan - they probably feel it would work across purposes with the Xeon platform. Xeon is what's meant by HEDT these days: https://www.notebookcheck.net/16-cores-32-threads-and-77-MB-of-combined-cache-Intel-Sapphire-Rapids-Fishhawk-Falls-HEDT-CPU-leaks.629930.0.html I mean Intel is in trouble. The 10nm process is really limiting them. The consumer CPUs can barely compete with AMD, then there is the ARM architecture. One thing I would like to see next in their high end consumer offering is at least 2x the amount of cache. This can work wonders even for gaming, as AMD demonstrated. L3 cache (shared): 12900K: 30MB 13900K: 36MB (a sad increase, on the bright side they doubled the L2 cache to 2MB/core) 7950x: 64MB ... Fastest gaming CPU with max boost clock of.... 4.5GHz: 96MB The cache matters even more for professional applications, so I really hope Intel catches up and fast, somehow, as we shouldn't be at the mercy of TSMC these days. My 12900KS visibly bottlenecks on cache/memory, so I'm not sure Raptor Lake with the extra e-Cores would help all that much, as the additional cores would just thrash the limited L3 cache. The Xeon platform is not the answer either, the low core variants (ca. 20 cores or less), have about the same if not less amount of cache.

Neither makes any sense to me from the usability perspective TBH. Guess there might be a technical reason (DB load), which would be fair enough.

Ah got it, so they went with larger fans, but lower RPM to keep the noise down. I guess that kind of makes sense, I mean those FE cards where targetting heavy OCing. On the other hand, good luck during a 40C summer heatwave! It's just that the resulting height + 16 pin connector adapter are so annoying, needs a fairly wide case with no side fans.

What's the point of the daily limit on reactions??

Does Nvidia lock out software fan control on those FE cards? Normally I would almost immediately customise the fan curve using Afterburner...

Not at all. As I was trying to explain, Titan was a different product class, above Ti. Titan RTX was released a month after 2080 Ti, and cost 2x as much at launch...

BTW let's not confuse Ti with Titan. The last true Titan card was RTX Titan. Perhaps the main difference is that FP64 performance is chopped by half in all gaming cards, including the Ti (Titan with no tan?). This is done in an effort to prevent the gaming cards from undercutting their high margin professional offerings (Axxxx, Quadro etc).

@PapusanExactly. With 3090 Ti they increased the number of cores by around 2%.... While they do have some thermal headroom due to the ridiculous card dimensions (4090 runs cooler than 3090 ti), IMHO they will struggle to offer more than +10% real performance improvement with the Ti (+20% in the keynote/official announcement :D). I guess they could throw in a sweetener such as DP 2.0 support.

Why? Does it look like they are holding a lot of performance out at the moment and will suddenly bump CUDA cores by 20%+ with the Ti? Is there a market appetite for $2000 cards anymore? Does it make sense to make the flagship card even bigger? TBH we don't even know how well the 4090 sells and judging by the poor sales of Ryzen 7 it could be a rough period for Nvidia and the market in general. Many people literally won't be able to fit these cards in their PCs.

Interesting, Processor Lasso looks like it would have been worth a try. Things are running fine now, I think I will just wait until a Linux kernel with proper AL support makes it to mainstream distros. Yeah, good advice, although tempted to just give it a test now as I am going to have to take the cooler off anyway soon. Got hit with the Arctic Freezer gasket issue. Arctic just sent me the repair kit... Haven't decided whether to jump from 12900KS to 13900K at launch or wait, but leaning towards the latter.

Yep, Nvidia def delivered the goods this time, also the FE card seems well engineered, except for perhaps the power connector placement. Need to do measurements to see if my case has enough height for it + the connector 😕 Anyone knows if card width figures are given with or without the board PCIe connector overhang? I am hoping the former lol Edit: Apparently the width includes the PCIe connector, i.e. it's the distance C below. Probably still won't fit my case, at least not without removing a 200mm side-radiator. Ridiculous design. OK AIB card it is then.

The way one uses their computers could indeed be a factor. Perhaps setting fixed core ratios does help somehow. I didn't see much difference in any standard benchmarks putting the system under 100% load such as CB23. The problem manifests itself through Windows suddenly assigning workloads to E-Cores, while P-Cores are idle, no matter what the load. Often this would only happen many hours of operation in, not sure why. High-performance Windows 10 power settings would make no difference (in contrast to Win 11 where AL system behaviour changes very noticeably in response to a change to Windows power settings, I found). If you don't see that behaviour ever, you are good. Intel themselves recommend Win 11 for Alder Lake+, and there are fairly in-depth papers online where they explain why (basically the scheduler has more information available to it in Win 11). On the flip side, Win 11 was def less stable than Win 10 for my system, as regards mem OC primarily I believe, so from an extreme OC-er's perspective definitely a concern. For anyone unsure, probably best to start Win 10, do all the tuning and benching, upgrade to Win 11, and move back to Win 10 if necessary. I flipped back and forth like that a few times before I settled on the final setup with no issues.

@jaybee83 Did you get an Arctic Freezer AIO? If so, does it have a QC sticker on it?

That's the thing, right - a lot depends on the use case. Gamers are in the fortunate situation of having everything benchmarked nicely for them, and here the situation is clear - the best CPU available currently is... the AM4 5800x 3DX 😄 I actually love Ryzen 7 in principle: * No e-Cores - these really need Windows 11 and a lot of scheduling magic to use efectively, although I have to say W11 does a great job - massive difference vs W10 (if only one didn't have to spend time manually disabling bloatware such as the Capability Access Manager Service!) * Large cache * AVX-512 - the trouble here is that this is 99% a marketing gimmick. The popular libraries which put AVX512 to some use are (mostly/exclusively) developed by Intel. These libraries detect AMD chips and severely gimp performance. Unfortuantely, on the scientifc side there is a huge reliance on Intel libraries, which means that the effective performance of the Intel platform will be much better in practice (we are talking about 2x-10x difference in some cases). Maybe AMD will finally invest in their own optimised libraries at some point, but until that happens they are removing themselves from a good chunk of the high-end market.