BTC $64,571.53 -0.50%
ETH $1,813.10 -0.05%
BNB $589.79 -0.69%
XRP $0.9534 -0.79%
SOL $71.45 -0.48%
TRX $0.3051 +0.65%
DOGE $0.0587 -0.91%
ADA $0.1520 -0.93%
BCH $207.38 -0.97%
LINK $9.09 -0.98%
HYPE $69.06 +3.28%
AAVE $108.35 -0.20%
SUI $0.7139 -1.96%
XLM $0.1370 -1.85%
ZEC $900.79 +0.27%
AAPL $353.96 -1.08%
AMZN $258.56 -2.91%
GOOGL $284.27 -3.86%
MSFT $520.02 -0.99%
META $626.20 -2.43%
NVDA $156.31 -4.54%
TSLA $379.57 -2.66%
SNDK $812.11 -5.75%
INTC $53.76 -8.23%
SPCX $137.68 -1.81%
MU $722.38 -5.83%
AMD $504.47 -3.62%
BTC $64,571.53 -0.50%
ETH $1,813.10 -0.05%
BNB $589.79 -0.69%
XRP $0.9534 -0.79%
SOL $71.45 -0.48%
TRX $0.3051 +0.65%
DOGE $0.0587 -0.91%
ADA $0.1520 -0.93%
BCH $207.38 -0.97%
LINK $9.09 -0.98%
HYPE $69.06 +3.28%
AAVE $108.35 -0.20%
SUI $0.7139 -1.96%
XLM $0.1370 -1.85%
ZEC $900.79 +0.27%
AAPL $353.96 -1.08%
AMZN $258.56 -2.91%
GOOGL $284.27 -3.86%
MSFT $520.02 -0.99%
META $626.20 -2.43%
NVDA $156.31 -4.54%
TSLA $379.57 -2.66%
SNDK $812.11 -5.75%
INTC $53.76 -8.23%
SPCX $137.68 -1.81%
MU $722.38 -5.83%
AMD $504.47 -3.62%

consumption

All
Article
Flash

hot_img Cerebras releases the fourth generation AI inference system CS-4: performance doubled, power consumption doubled, more flexible deployment

Cerebras released its fourth-generation AI inference system CS-4 this week, based on the same 5nm WSE-3 wafer, achieving double the performance by doubling the clock frequency and power consumption. A single CS-4 cabinet accommodates 3 wafers (CS-3 has 2), featuring a modular "backpack" design that simplifies manufacturing and deployment, with a TDP of approximately 125 to 135kW. The CS-4 can provide an inference speed of nearly 4000 tokens/second/user, about twice that of the CS-3, and supports decomposed inference with heterogeneous systems such as AMD and AWS Trainium.Cerebras claims that the CS-4 offers about 2000 times the on-chip memory bandwidth of NVIDIA's Rubin (43PB/s), but the 44GB SRAM capacity remains unchanged, and long-context inference still requires multi-wafer stacking. For example, with the DeepSeek V4 Pro (1.6T parameters), approximately 20 systems are needed for a 1M context window, and about 40 systems are required for 256 concurrent users, corresponding to a CAPEX exceeding 20 million USD. Cerebras is collaborating with clients such as OpenAI and plans to achieve approximately double performance improvements each year, aiming for a 20-fold throughput increase by 2027. The "backpack" cabinet design of the CS-4 will continue into the next-generation "Nexus" platform.
app_icon
ChainCatcher Building the Web3 world with innovations.