Search results

Because 4090's are made on the same process node as 5090's. If they kept making 4090's, they'd be throwing away $500 for every one they made. Also, let me know where you're finding $2200 5090's because I'd like to buy a few :)

This is hilarious Nvidia shilling. DeepSeek's capex into the hardware they used for training was a lot of money, but the actual GPU hours they burned in training this specific model was only a few million dollars, as calculated against current cloud pricing (it's actually less if you only count...

Blackwell's success will be predicated on whether neural rendering actually works for gaming. I can speak from experience as an engineer in the field that the general technique is real and should be taken seriously - basically, the idea is rather than waiting for frames 5 and 6 and then using...

AMD getting serious about AI is a good thing for everyone because AI has been the first reason for regular users to buy a new computer since like...2017. on device AI is certainly not the world changing hype the venture capitalists make it out to be, but I would put it at a comparable level to...

get the firmware update. the chip shipped with too aggressive of a power/clock target to hit benchmark targets and degrades over time. some code paths, especially shader compilation, seem to really stress the cores. it's not going to die over 4 hours but over thousands of accumulated hours you...

Do you know how hot it gets inside the dome during the day? The CPU might not mind 50C+ ambient temps, but the board and power supply might not be too happy.

The OP is free to spend their money how they want but unless you want to show off to your friends how much money you have there are much better deals on 4090 prebuilts - depending on the brand you can save anywhere from $1000-$2000.

exactly my point. the benchmark results are being skewed by ISA extensions which are extremely benchmark friendly (the GeekBench crypto benchmark is notorious for this) and E-cores which only help if your workload scales to 12+ threads. if it doesn't (and a lot of workloads don't), I would...

there's definitely something going on, there's no way a 13600K is _four times_ faster than a six core Sandy Bridge and even less chance that a quad core 15W i7 can match a 4.6GHz six core Sandy Bridge - Sunny Cove has something like 1.5x the IPC of Sandy Bridge but the 3930K has two more cores...

I think a lot of horror stories are because the techs at the RMA shop bend the pins and then reject the warranty because of bent pins. Asus was/is particularly notorious for this, they are absolutely ruthless when it comes to enforcing physical damage rules against returns and receive a lot of...

I'd rather spring for a 7800X3D. The 7900X3D fills a legitimate niche of wanting OK multicore performance, OK gaming performance, and a good price, but the problem is games are lassoed to the CCD with extra cache so for gaming, it runs like a six-core CPU. That would be fine, except...the...

A much needed rebranding. The precious metals based scheme was utter chaos, with at one point something like 100 finely segmented SKUs. Conroe to Haswell at least had feature parity across the entire stack (except for the two socket-filler SKUs each generation with no turbo or HT), but once...

yes, by lowering the boost clock or increasing Vcore (which will reduce the boost duration), both of which have a negative impact on performance.

Tricky question. 5120x1440 is the same pixel count as 4K which means nothing other than a 4090 will be futureproof at maxed out settings. The 7900 XTX is a good contender, BUT DLSS is way better than FSR - DLSS uses game-specific learned sharpening filters based on a neural network whereas FSR...

While AMD does have a fair share of issues, the latest stability issues on the Intel side are not confidence inspiring, and furthermore, are obviously traceable to what is essentially a bad factory overclock, something which can't be patched in software. On the AMD side, I wouldn't get anything...

You're saying you want to upgrade and you want Intel, so I think you answered your own question - besides the 14700K and 14900K there aren't any other sensible choices for gaming on Intel right now. That being said, Ryzen 7000X3D is the gaming champ right now and is probably a bit more stable...

they were worth $3K a piece in china right before the sanctions kicked in so there was a ton of scalping. it looks like pricing has settled back down now

yes, but at what cost? at 3 tokens per second you might as well use a CPU. also, cifar-10 in resnet50 is not representative of AI workloads in 2024. the weights and activations are so small that it becomes difficult to achieve high arithmetic intensity, which is why you only see a 2x speedup...

1660S/2060 have way too little VRAM to do anything meaningful. @OP the price of entry is $250, save up for a 3060 12GB or better yet, a 4060 TI 16 GB.

yeah 4090's are an incredibly good deal, you're looking at $500 for a yielded, tested, and packaged chip plus margins, then $100 of VRAM, plus PMICs, cooler, board, testing, packaging and distribution on the AIB side...AIB margins are not so good at $1500.

Are you sure you don't want a...laptop? They run of off DC, are generally designed to have super low idle power, are built out of wide GPUs clocked low and binned for good low power performance, etc. If you were building a new rig you could get a 4080 based laptop for under $2K which would slide...

I think its a design choice, not incompetence. AD102 has 30% more transistors than Navi 31, and performs about 30% faster. AMD's wafer starts are even more strained than Nvidia's - they have to fab Epyc, Ryzen, Navi, and Instinct, parts with wildly different markets and margins, all at the same...

The problem isn't profit, or margins, or whether Nvidia cares about you or not. The problem is foundry capacity. When you're not capacity-limited, you print as many chips as you can, then sell as many of those as high-margin enterprise parts as possible. Once you've saturated the enterprise...

A good product with, unfortunately, a bad price imposed by marketing - the die needs to be 8-core to compete in the laptop space, but by having 8 cores, it needs to have a $300+ MSRP to not cannibalize 7800X sales. As the reviews point out, $300 is the same price as a RX 6600 and a $99 CPU, a...

It's been this way for as long as I can remember. PayPal INR (item not received) cases will automatically resolve in the seller's favor if the seller can provide a tracking number for a shipment to a ZIP code that matches the buyer's ZIP. If the seller mails you an empty box, you can try to file...

4080S because ray tracing or not, DLSS gives a huge step up in performance for not a lot of visual impact.

You really need a price target, otherwise the undisputed answer is the RTX 4000 Ada but that is a $1500 4070 variant meant for the most dedicated of SFF fans and workstation users.

What workflow? Interactive 3D modeling is mostly single threaded so regardless of the CPU TDP you're not pulling much more than 60W or so. 3D scanning really depends on the software. AMD is not more efficient than Intel, AMD simply chooses a more efficient operating point. You pick the TDP -...

I've said this before and I'll say it again: the underlying price driver is NVIDIA needs to do everything it can to ensure that Geforce GPUs cannot be used to train AI models. Since 2016, they've: barred the use of Geforce drivers in datacenters, barred OEMs from making cards under 3 slots...

oof. If you want D-SUB that really limits your choices. otherwise I'm a huge fan of AMD APUs for light usage, the graphics are still a good deal faster than the equivalent priced Intel part and more importantly, the APUs share a driver stack with the mainstream Radeons meaning games will all run...

See, having owned a deck for over a year I claim it's anything but "turnkey"; not only is the underlying OS Linux, it's Arch Linux, and not only is it Arch, it's "Arch but you only get to use Flatpaks". Newer AAA games generally have good support, but older games (which are exactly the kind of...

You sure it's not a memory compatibility issue? I had a Z690 Dark and a i3-12100 that took a week of fiddling to get the DDR5 to train, and I think I saw a similar issue (only POSTing with one channel) somewhere along the way.

The real question is what games you're planning to play on an ARM Mac...

ah yes, a real classic on Supermicros. the middle standoff on the top row shorts out the topmost DIMM slot.

Only until next month. People discovered if you buy 4090s and ship them *into China* you can make $1K+ per card because of the impending sanctions so they bought up all of the 4090s in stock right now. There's a little bit of supply chain awkwardness because Nvidia can no longer build boards in...

Does the X99-UD4 have above 4G decoding by chance?

The real question is why you're trying to charge a laptop off a PCIe card...

They are a hair over $200 on eBay https://www.ebay.com/itm/394898676932, not too bad.

Works both ways, insofar as your flavor of Linux is new enough to support the latest proprietary drivers.

If you're looking for a 12GB card for inference it's hard to beat the 3060 - tensor cores, most recent architecture (Nvidia has supply chain issues so Hopper is basically MIA, meaning Ampere will be supported for some time to come), and clean upgrade path through the Ampere stack to A100. V100's...