6 ms·
Intel Announces Inference-Optimized Xe3P Graphics Card with 160GB VRAM
- vrighter 11mo agodoes anyone still make gpus for graphics anymore?
- RoyTyrell 11mo agoWill this have any support for open source libraries like PyTorch or will it be all Intel proprietary software that you need a license for?
- CoastalCoder 11mo agoIntel puts a huge priority on DL framework support before releasing related hardware, going back to at least 2017. I assume that hasn't changed.
- 0xfedcafe 11mo agoOpenVino is entirely open-source and can run PyTorch and ONNX models, so this is definitely not a topic of concern. PyTorch also has native Intel GPU support https://docs.pytorch.org/docs/stable/notes/get_start_xpu.html#hardware-prerequisite https://docs.pytorch.org/docs/stable/notes/get_start_xpu.htm...
- pjmlp 11mo agoThere is PyTorch support on oneAPI.
- knowitnone3 11mo agoAny business people here that can explain why companies announce products a year before their release? I can understand getting consumers excited but it also tells competitors what you are doing giving them time to make changes of their own. What's the advantage here?
- teeray 11mo ago> What's the advantage here? Stock number go up
- creaturemachine 11mo agoThe AI bubble might not last another year. Better get a few more pumps in before it blows.
- Mars008 11mo agoAI is not going anywhere. Now everyone wants to get a piece. Local inference is expected to grow. Documents, image, video, etc processing. Another obvious is driverless farm vehicles and other automated equipment. "Assisted" books, images, news,.. already and grows fast. Translation also a fact.
- thenaturalist 11mo agoThe technology, maybe - and if on local. The public co valuations of quickly depreciating chip hoarders selling expensive fever dreams to enterprises are gonna pop though. Spend 3-7 USD for 20 cents in return and 95% project failures rates for quarters on end aren't gonna go unnoticed on Wall St.
- Mars008 11mo agoSo far there is no 'plateau' in the nearest future. 'AI' as a science and its applications should develop further for the next several years. Models will get more efficient, but still the bigger the better. This is obvious. Even if models don't scale up well, they can be used collectively in parallel 'brainstorming'. This will still create demand for hardware. Stagnation is still possible in case of recession. In this case even stable businesses will suffer. As for efficiency, replacing one programmer in group of 10 with AI already will increase productivity and lower the price. In most cases. In reality adding AI accounts to existing group works better. This is _now_, not hopes or sci-fi. That's why I'm saying there is no way back. 'AI winter' is as likely as smartphones winter.
- schmorptron 11mo agoXe3P as far as I remember is built in their own fabs as opposed to xe3 at TSMC. This could give them a huge advantage by being possibly the only competitor not competing for the same TSMC wafers
- mft_ 11mo agoI have no idea of the likely price, but (IMO) this is the sort of disruption that Intel needs to aim at if it's going to make some sort of dent in this market. If they could release this for around the price of a 5090, it would be very interesting.
- schmorptron 11mo agoMaybe not that low, but given it's using LPDDR5 instead of GDDR7, at least the ram should be a lot cheaper.
- Neywiny 11mo agoCertainly an interesting choice. Dramatically worse performance but dramatically larger only time will tell how it actually goes
- schmorptron 11mo agoRumor has it (according to MLID, so no one knows whether it's accurate) that AMD is also looking to use regular LPDDR memory for some of it's lower end next gen GPUs to not have to contend with nvidia over limited and cartelled GDDR7 supply. Maybe they're going to increase parallel bandwidth to compensate it? Or have wholly different tricks up their sleeve.
- woleium 11mo agoprobably just a lot more of it, to capture that consumer ai market
- magicalhippo 11mo agoWith 160GB, surely they can add more channels to compensate?
- benob 11mo agoIs there anything preventing them from using heterogeneous memory chips, like 1/4 GDDR7 and 3/4 LPDDR? It could enable new MEO-like architectures with finer-grained performance tuning for long contexts.
- api 11mo agoA not-absurdly-priced card that can run big models (even quantized) would sell like crazy. Lots and lots of fast RAM is key.
- bigwheels 11mo agoHow does LPDDR5 (This Xe3P) compare with GDDR7 (Nvidia's flagships) when it comes to inference performance? Local inference is an interesting proposition because today in real life, the NV H300 and AMD MI-300 clusters are operated by OpenAI and Anthropic in batching mode, which slows users down as they're forced to wait for enough similar sized queries to arrive. For local inference, no waiting is required - so you could get potentially higher throughput.
- qingcharles 11mo agoI asked GPT to pull real stats on both. Looks like the 50-series RAM is about 3X that of the Xe3P, but it wanted to remind me that this new Intel card is designed for data centers and is much lower power, and that the comparable Nvidia server cards (e.g. H200) have even better RAM than GDDR7, so the difference would be even higher for cloud compute.
- halJordan 11mo agoLpddr5x (not lpddr5) is 10.7 Gbps. Gddr7 is 32 Gbps. So it's going to be slower
- codedokode 11mo agoYes but in matrix multiplication there are O(N²) numbers and O(N³) multiplications, so it might be possible that you are bounded by compute speed.
- electroglyph 11mo agoboth are equally important. compute for prefill and mem bandwidth for generation
- bigmattystyles 11mo agoI remember Larabee and Xeon-Phi announcements and getting so excited at the time. So I'll wait but curb my enthusiasm.
- Analemma_ 11mo agoYeah, Intel's problem is that this is (at least) the third time they've announced a new ML accelerator platform, and the first two got shitcanned. At this point I wouldn't even glance at an Intel product in this space until it had been on the market for at least five years and several iterations, to be somewhat sure it isn't going to be killed, and Intel's current leadership inspires no confidence that they'll wait that long for success.
- wmf 11mo agoXe works much much better than Larabee or Xeon Phi ever did. Xe3 might even be good.
- throwaway173738 11mo agoI’m personally just thinking about how they treated their embedded Keem Bay line. Totally shitcanned without warning. I doubt they consider this a core market to the degree that they will endure bad sales numbers for a while.
- makapuf 11mo agoFunny they still call them graphics cards when they're really... I dont know, matmul cards ? Tensor cards ? TPU ? Well that sums it up maybe, what those are are really CUDA cards.
- halJordan 11mo agoDude, this is asinine. Graphics cards have been doing matrix and vector operations since they were invented. No one had a problem with calling matrix multiplers graphics cards until it became cool to hate AI.
- adastra22 11mo agoIt was many generations before vector operations were moved onto graphics chips.
- boomskats 11mo agoIf you s/graphics/3d graphics does that still hold true?
- gary_0 11mo agoYes. The earliest consumer PC 3D graphics cards just rasterized pre-transformed triangles and that's it; the CPU had to do pretty much all the math (but drawing the pixels was considered the hard part). Later, "Hardware Transform and Lighting (T&L)" was introduced circa 2000 by cards like the GeForce 256.
- Sesse__ 11mo agoAnd even then, you couldn't really get any sort of serious matmul out of it; they were per-vertex, not per-pixel. Per-pixel matmul (which is what you really need for anything resembling GPGPU) came with Shader Model 2.0, circa 2002; Radeon 9700, the GeForce FX series and the likes. CUDA didn't exist (nor really any other form of compute shaders), but you could wrangle it with pixel shaders, and some of us did.
- eadwu 11mo agoIt'll be either "cheap" like the DGX Spark (with crap memory bandwidth) or overpriced with the bus width of a M4 Max with the rhetoric of Intel's 50% margin.
- phonon 11mo agoOr it will be cheap, with the ability to expand 8X on a server. Particularly with PCIe 6.0 coming soon, might be a very attractive package. https://www.linkedin.com/posts/storagereview_storagereview-ai-llm-activity-7374886399185055744-Y59o/ https://www.linkedin.com/posts/storagereview_storagereview-a...
- Tepix 11mo agoSound as if it won‘t be widely available before 2027 which disappointing for a 341GB/s chip.
- storus 11mo agoIntel leadership actually reads HN? Mindblown...
- silisili 11mo agoBetween 18A becoming viable and this, it seems Intel is finally climbing out of the hole it's been in for years. Makes me wonder whether Gelsinger put all this in motion, or if the new CEO lit a fire under everyone. Kinda a shame if it's the former...
- viraptor 11mo agoGelsinger had a long term realistic plan. He was out around 11 months ago. You can't magic a new GPU in that timeframe - those projects have 3+ years pipelines for CPUs. I assume GPU will be a bit shorter, but not that much. Whatever happened with new products today must've been started before he left.
- cwillu 11mo agoAny discussion of an intel entry to discrete graphics cards needs to at least _mention_ intel's repeated history of abandoning discrete graphics cards.
- sharts 11mo agoYou’re saying it’s like the Google of graphics cards?
- cwillu 11mo agoVery much so.
- hnuser123456 11mo agoAt least larrabee's cancellation resulted in the Offset engine going to the Firefall (2014) devs, which was a really great F2P MMO game for a while.
- kobalsky 11mo agothe GPU market is not what it used to be, it's not some checkbox some executive needs to check to say "we are doing something". the chips are so valuable now NVIDIA will end up owning a chunk of every major tech company, everyone is throwing cash and shares at them as fast as they can.
- mawadev 11mo agoHonestly, Intel just has to build a GPU with insane amount of VRAM. It doesn't even have to be the fastest to compete... just a ton of vram for dirt cheap
- jychang 11mo agoIt’s LPDDR5x It’s gonna be slowwww It’s gonna be what, 273GB/sec vram bandwidth at most? Might as well as buy an AND 395+ 128GB right now for the same inference performance and slightly less VRAM.
- hengheng 11mo agoHow can you tell without knowing the bus width?
- maeln 11mo agoSlow is better than nothing. A card with this much VRAM in a "prosumer" price range would be really interesting right now for workstation, to work with big models.
- jychang 11mo agoSlow is worse than nothing. What's the point of this card that's going to be released around the same time as DDR6, and DDR6 will be faster? Might as well as use cheaper system RAM if you system RAM is slower than VRAM.
- kingstnap 11mo agoBandwidth depends very much on on bus width. If its fast LPDDR5x (9600 MT/s) with 512 bit bus width (8 64bit channels (actually multiples of quad 16 bit subchannel nonsense)) it could be upwards of 600 GB/s. Lots of bandwidth like the beefy macs have.
- jychang 11mo ago1. 600GB/sec is still slow as hell. You might as well as use regular DDR5 RAM then if you're so slow, you can spec regular system DDR5 RAM faster than 600GB/sec. The half decade old consumer 3090 is 1.5x that speed. The current 5090 is 1,792 GB/s. The current nvidia datacenter cards are 8 TB/s. What's the point of having lots of VRAM if your system RAM is faster? For context: if you have a 160GB dense ML model in VRAM and you're just running 600GB/sec, you can do... roughly 4 tokens per second AT BEST. That massive amount of VRAM is unusable if it's slow. 2. 512 bit LPDDR5x is most likely just 512GB/sec with typical LPDDR5x that's not overly expensive. I would be HIGHLY surprised if they gave it the more expensive RAM that'd break 600GB/sec. The Intel B60 is at 456 GB/s and that's using GDDR6. Honestly, you're better off waiting for regular DDR6 to come out in a year and just build a system using that.
- lillecarl 11mo agoI'm hopeful for the second hand market, imagine when these have paid for themselves and you can do local inference of crazy capable models!?
- tonetegeatinst 11mo agoWhat price is this sitting at? Because if its software support is decent then Intel might have just managed to break into the hardware for AI on the edge. Examples like self hosted LLM finetuning and RAG on a old dell or HP server with these type of cards on them.
- Aurornis 11mo ago> Examples like self hosted LLM finetuning and RAG on an old dell or HP server with these type of cards on them. This won’t be in the price range of an old Dell server or a fun impulse buy for a hobbyist. 160GB of raw LPDDR5X chips alone is not cheap. This is a server/workstation grade card and the price is going where the market will allow. Consider that an nVidia card with almost half the RAM is going to cost $8K or more. That price point is probably the starting point for where this will be priced, too.
- vardump 11mo agoThat nVidia card is going to have 5x the memory bandwidth. LPDDR5X is going to be rather low bandwidth. (My guess is Intel's card is only going to have about 400 GB/s bandwidth.)
- matt-p 11mo agoLast year's M4 max MacBook is 520GB/s and (I expect) that should be closer to 1TB/s in a year or two by the time they are using ddr5. It would be deeply embarrassing if they had worse performance than apples cheaper laptop.
- matt-p 11mo agoMaybe not old but if this was say a 6k card that would make it accessible to pretty much any business and at least some hobbyists. 160GB of Lpddr5 should be less than 2k, so it's easily doable if they've got the will. 4 x 5090s is 128GB and probs much more powerful at ~8k, so it would need to be 6/7k to make it make sense.
- thedudeabides5 11mo agothe mad lad leopold did it, props
- DrNosferatu 11mo agoAnyone has any idea about the price?
- incomingpain 11mo agoA year out, in that time nvidia and amd; not to mention huawei and others are going to hit the market as well. Intel are quite behind. To me, the price point is what matters. It's going to be slow with ddr5. The 5090 today is much faster. But sure big ram. RTX pro 6000 with 96gb of ram will be much faster. So I'm thinking price point is below the 6000, above the 5090.
- g42gregory 11mo agoAnybody knows memory bandwidth?
- DrNosferatu 11mo agoIt would be great if they would greatly undercut the price of the NVIDIA DGX Spark.
- nullsmack 11mo agowhoa, shoot this directly into my veins